diff options
Diffstat (limited to 'erts/emulator/pcre/pcre_compile.c')
| -rw-r--r-- | erts/emulator/pcre/pcre_compile.c | 6052 |
1 files changed, 4131 insertions, 1921 deletions
diff --git a/erts/emulator/pcre/pcre_compile.c b/erts/emulator/pcre/pcre_compile.c index 9508c5a697..5652bad0ce 100644 --- a/erts/emulator/pcre/pcre_compile.c +++ b/erts/emulator/pcre/pcre_compile.c @@ -6,7 +6,7 @@ and semantics are as close as possible to those of the Perl 5 language. Written by Philip Hazel - Copyright (c) 1997-2008 University of Cambridge + Copyright (c) 1997-2013 University of Cambridge ----------------------------------------------------------------------------- Redistribution and use in source and binary forms, with or without @@ -38,7 +38,7 @@ POSSIBILITY OF SUCH DAMAGE. */ -/* This module contains the external function erts_pcre_compile(), along with +/* This module contains the external function pcre_compile(), along with supporting internal functions that are not used by other modules. */ /* %ExternalCopyright% */ @@ -54,17 +54,22 @@ supporting internal functions that are not used by other modules. */ #include "pcre_internal.h" -/* When DEBUG is defined, we need the pcre_printint() function, which is also -used by pcretest. DEBUG is not defined when building a production library. */ +/* When PCRE_DEBUG is defined, we need the pcre(16|32)_printint() function, which +is also used by pcretest. PCRE_DEBUG is not defined when building a production +library. We do not need to select pcre16_printint.c specially, because the +COMPILE_PCREx macro will already be appropriately set. */ -#ifdef DEBUG -#include "pcre_printint.src" +#ifdef PCRE_DEBUG +/* pcre_printint.c should not include any headers */ +#define PCRE_INCLUDED +#include "pcre_printint.c" +#undef PCRE_INCLUDED #endif /* Macro for setting individual bits in class bitmaps. */ -#define SETBIT(a,b) a[b/8] |= (1 << (b%8)) +#define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7)) /* Maximum length value to check against when making sure that the integer that holds the compiled pattern length does not overflow. We make it a bit less than @@ -73,6 +78,18 @@ to check them every time. */ #define OFLOW_MAX (INT_MAX - 20) +/* Definitions to allow mutual recursion */ + +static int + add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *, + const pcre_uint32 *, unsigned int); + +static BOOL + compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int, + pcre_uint32 *, pcre_int32 *, pcre_uint32 *, pcre_int32 *, branch_chain *, + compile_data *, int *); + + /************************************************* * Code parameters and static tables * @@ -88,36 +105,89 @@ so this number is very generous. The same workspace is used during the second, actual compile phase for remembering forward references to groups so that they can be filled in at the end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE -is 4 there is plenty of room. */ +is 4 there is plenty of room for most patterns. However, the memory can get +filled up by repetitions of forward references, for example patterns like +/(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so +that the workspace is expanded using malloc() in this situation. The value +below is therefore a minimum, and we put a maximum on it for safety. The +minimum is now also defined in terms of LINK_SIZE so that the use of malloc() +kicks in at the same number of forward references in all cases. */ -#define COMPILE_WORK_SIZE (4096) +#define COMPILE_WORK_SIZE (2048*LINK_SIZE) +#define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE) /* The overrun tests check for a slightly smaller size so that they detect the overrun before it actually does run off the end of the data block. */ -#define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100) +#define WORK_SIZE_SAFETY_MARGIN (100) +/* Private flags added to firstchar and reqchar. */ + +#define REQ_CASELESS (1 << 0) /* Indicates caselessness */ +#define REQ_VARY (1 << 1) /* Reqchar followed non-literal item */ +/* Negative values for the firstchar and reqchar flags */ +#define REQ_UNSET (-2) +#define REQ_NONE (-1) + +/* Repeated character flags. */ + +#define UTF_LENGTH 0x10000000l /* The char contains its length. */ /* Table for handling escaped characters in the range '0'-'z'. Positive returns are simple data values; negative values are for special things like \d and so on. Zero means further processing is needed (for things like \x), or the escape is invalid. */ -#ifndef EBCDIC /* This is the "normal" table for ASCII systems */ +#ifndef EBCDIC + +/* This is the "normal" table for ASCII systems or for EBCDIC systems running +in UTF-8 mode. */ + static const short int escapes[] = { - 0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 7 */ - 0, 0, ':', ';', '<', '=', '>', '?', /* 8 - ? */ - '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E, 0, -ESC_G, /* @ - G */ --ESC_H, 0, 0, -ESC_K, 0, 0, 0, 0, /* H - O */ --ESC_P, -ESC_Q, -ESC_R, -ESC_S, 0, 0, -ESC_V, -ESC_W, /* P - W */ --ESC_X, 0, -ESC_Z, '[', '\\', ']', '^', '_', /* X - _ */ - '`', 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0, /* ` - g */ --ESC_h, 0, 0, -ESC_k, 0, 0, ESC_n, 0, /* h - o */ --ESC_p, 0, ESC_r, -ESC_s, ESC_tee, 0, -ESC_v, -ESC_w, /* p - w */ - 0, 0, -ESC_z /* x - z */ + 0, 0, + 0, 0, + 0, 0, + 0, 0, + 0, 0, + CHAR_COLON, CHAR_SEMICOLON, + CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN, + CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK, + CHAR_COMMERCIAL_AT, -ESC_A, + -ESC_B, -ESC_C, + -ESC_D, -ESC_E, + 0, -ESC_G, + -ESC_H, 0, + 0, -ESC_K, + 0, 0, + -ESC_N, 0, + -ESC_P, -ESC_Q, + -ESC_R, -ESC_S, + 0, 0, + -ESC_V, -ESC_W, + -ESC_X, 0, + -ESC_Z, CHAR_LEFT_SQUARE_BRACKET, + CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET, + CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE, + CHAR_GRAVE_ACCENT, 7, + -ESC_b, 0, + -ESC_d, ESC_e, + ESC_f, 0, + -ESC_h, 0, + 0, -ESC_k, + 0, 0, + ESC_n, 0, + -ESC_p, 0, + ESC_r, -ESC_s, + ESC_tee, 0, + -ESC_v, -ESC_w, + 0, 0, + -ESC_z }; -#else /* This is the "abnormal" table for EBCDIC systems */ +#else + +/* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */ + static const short int escapes[] = { /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|', /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0, @@ -136,7 +206,7 @@ static const short int escapes[] = { /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-', /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G, /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0, -/* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P, +/* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P, /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0, /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X, /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0, @@ -148,33 +218,40 @@ static const short int escapes[] = { /* Table of special "verbs" like (*PRUNE). This is a short table, so it is searched linearly. Put all the names into a single string, in order to reduce -the number of relocations when a shared library is dynamically linked. */ +the number of relocations when a shared library is dynamically linked. The +string is built from string macros so that it works in UTF-8 mode on EBCDIC +platforms. */ typedef struct verbitem { - int len; - int op; + int len; /* Length of verb name */ + int op; /* Op when no arg, or -1 if arg mandatory */ + int op_arg; /* Op when arg present, or -1 if not allowed */ } verbitem; static const char verbnames[] = - "ACCEPT\0" - "COMMIT\0" - "F\0" - "FAIL\0" - "PRUNE\0" - "SKIP\0" - "THEN"; - -static verbitem verbs[] = { - { 6, OP_ACCEPT }, - { 6, OP_COMMIT }, - { 1, OP_FAIL }, - { 4, OP_FAIL }, - { 5, OP_PRUNE }, - { 4, OP_SKIP }, - { 4, OP_THEN } + "\0" /* Empty name is a shorthand for MARK */ + STRING_MARK0 + STRING_ACCEPT0 + STRING_COMMIT0 + STRING_F0 + STRING_FAIL0 + STRING_PRUNE0 + STRING_SKIP0 + STRING_THEN; + +static const verbitem verbs[] = { + { 0, -1, OP_MARK }, + { 4, -1, OP_MARK }, + { 6, OP_ACCEPT, -1 }, + { 6, OP_COMMIT, -1 }, + { 1, OP_FAIL, -1 }, + { 4, OP_FAIL, -1 }, + { 5, OP_PRUNE, OP_PRUNE_ARG }, + { 4, OP_SKIP, OP_SKIP_ARG }, + { 4, OP_THEN, OP_THEN_ARG } }; -static int verbcount = sizeof(verbs)/sizeof(verbitem); +static const int verbcount = sizeof(verbs)/sizeof(verbitem); /* Tables of names of POSIX character classes and their lengths. The names are @@ -184,11 +261,12 @@ length entry. The first three must be alpha, lower, upper, as this is assumed for handling case independence. */ static const char posix_names[] = - "alpha\0" "lower\0" "upper\0" "alnum\0" "ascii\0" "blank\0" - "cntrl\0" "digit\0" "graph\0" "print\0" "punct\0" "space\0" - "word\0" "xdigit"; + STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0 + STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0 + STRING_graph0 STRING_print0 STRING_punct0 STRING_space0 + STRING_word0 STRING_xdigit; -static const uschar posix_name_lengths[] = { +static const pcre_uint8 posix_name_lengths[] = { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 }; /* Table of class bit maps for each POSIX class. Each class is formed from a @@ -218,6 +296,107 @@ static const int posix_class_maps[] = { cbit_xdigit,-1, 0 /* xdigit */ }; +/* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class +substitutes must be in the order of the names, defined above, and there are +both positive and negative cases. NULL means no substitute. */ + +#ifdef SUPPORT_UCP +static const pcre_uchar string_PNd[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_pNd[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_PXsp[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_pXsp[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_PXwd[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_pXwd[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' }; + +static const pcre_uchar *substitutes[] = { + string_PNd, /* \D */ + string_pNd, /* \d */ + string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */ + string_pXsp, /* \s */ + string_PXwd, /* \W */ + string_pXwd /* \w */ +}; + +static const pcre_uchar string_pL[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_pLl[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_pLu[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_pXan[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_h[] = { + CHAR_BACKSLASH, CHAR_h, '\0' }; +static const pcre_uchar string_pXps[] = { + CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_PL[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_PLl[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_PLu[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_PXan[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' }; +static const pcre_uchar string_H[] = { + CHAR_BACKSLASH, CHAR_H, '\0' }; +static const pcre_uchar string_PXps[] = { + CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET, + CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' }; + +static const pcre_uchar *posix_substitutes[] = { + string_pL, /* alpha */ + string_pLl, /* lower */ + string_pLu, /* upper */ + string_pXan, /* alnum */ + NULL, /* ascii */ + string_h, /* blank */ + NULL, /* cntrl */ + string_pNd, /* digit */ + NULL, /* graph */ + NULL, /* print */ + NULL, /* punct */ + string_pXps, /* space */ /* NOTE: Xps is POSIX space */ + string_pXwd, /* word */ + NULL, /* xdigit */ + /* Negated cases */ + string_PL, /* ^alpha */ + string_PLl, /* ^lower */ + string_PLu, /* ^upper */ + string_PXan, /* ^alnum */ + NULL, /* ^ascii */ + string_H, /* ^blank */ + NULL, /* ^cntrl */ + string_PNd, /* ^digit */ + NULL, /* ^graph */ + NULL, /* ^print */ + NULL, /* ^punct */ + string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */ + string_PXwd, /* ^word */ + NULL /* ^xdigit */ +}; +#define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *)) +#endif #define STRING(a) # a #define XSTRING(s) STRING(s) @@ -230,7 +409,11 @@ the number of relocations needed when a shared library is loaded dynamically, it is now one long string. We cannot use a table of offsets, because the lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we simply count through to the one we want - this isn't a performance issue -because these strings are used only when there is a compilation error. */ +because these strings are used only when there is a compilation error. + +Each substring ends with \0 to insert a null character. This includes the final +substring, so that the whole string ends with \0\0, which can be detected when +counting through. */ static const char error_texts[] = "no error\0" @@ -271,13 +454,13 @@ static const char error_texts[] = /* 30 */ "unknown POSIX class name\0" "POSIX collating elements are not supported\0" - "this version of PCRE is not compiled with PCRE_UTF8 support\0" + "this version of PCRE is compiled without UTF support\0" "spare error\0" /** DEAD **/ "character value in \\x{...} sequence is too large\0" /* 35 */ "invalid condition (?(0)\0" "\\C not allowed in lookbehind assertion\0" - "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0" + "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0" "number after (?C is > 255\0" "closing ) for (?C expected\0" /* 40 */ @@ -294,22 +477,40 @@ static const char error_texts[] = "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0" /* 50 */ "repeated subpattern is too long\0" /** DEAD **/ - "octal value is greater than \\377 (not in UTF-8 mode)\0" + "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0" "internal error: overran compiling workspace\0" "internal error: previously-checked referenced subpattern not found\0" "DEFINE group contains more than one branch\0" /* 55 */ - "repeating a DEFINE group is not allowed\0" + "repeating a DEFINE group is not allowed\0" /** DEAD **/ "inconsistent NEWLINE options\0" - "\\g is not followed by a braced name or an optionally braced non-zero number\0" - "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number\0" - "(*VERB) with an argument is not supported\0" + "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0" + "a numbered reference must not be zero\0" + "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0" /* 60 */ - "(*VERB) not recognized\0" + "(*VERB) not recognized or malformed\0" "number is too big\0" "subpattern name expected\0" - "digit expected after (?+"; - + "digit expected after (?+\0" + "] is an invalid data character in JavaScript compatibility mode\0" + /* 65 */ + "different names for subpatterns of the same number are not allowed\0" + "(*MARK) must have an argument\0" + "this version of PCRE is not compiled with Unicode property support\0" + "\\c must be followed by an ASCII character\0" + "\\k is not followed by a braced, angle-bracketed, or quoted name\0" + /* 70 */ + "internal error: unknown opcode in find_fixedlength()\0" + "\\N is not supported in a class\0" + "too many forward references\0" + "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0" + "invalid UTF-16 string\0" + /* 75 */ + "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0" + "character value in \\u.... sequence is too large\0" + "invalid UTF-32 string\0" + "setting UTF is disabled by the application\0" + ; /* Table to identify digits and hex digits. This is used when compiling patterns. Note that the tables in chartables are dependent on the locale, and @@ -327,8 +528,18 @@ For convenience, we use the same bit definitions as in chartables: Then we can use ctype_digit and ctype_xdigit in the code. */ -#ifndef EBCDIC /* This is the "normal" case, for ASCII systems */ -static const unsigned char digitab[] = +/* Using a simple comparison for decimal numbers rather than a memory read +is much faster, and the resulting code is simpler (the compiler turns it +into a subtraction and unsigned comparison). */ + +#define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9) + +#ifndef EBCDIC + +/* This is the "normal" case, for ASCII systems, and EBCDIC systems running in +UTF-8 mode. */ + +static const pcre_uint8 digitab[] = { 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */ @@ -363,8 +574,11 @@ static const unsigned char digitab[] = 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */ -#else /* This is the "abnormal" case, for EBCDIC systems */ -static const unsigned char digitab[] = +#else + +/* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */ + +static const pcre_uint8 digitab[] = { 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */ @@ -399,7 +613,7 @@ static const unsigned char digitab[] = 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */ 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */ -static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */ +static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */ 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */ 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */ 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */ @@ -435,13 +649,6 @@ static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */ #endif -/* Definition to allow mutual recursion */ - -static BOOL - compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int, - int *, int *, branch_chain *, compile_data *, int *); - - /************************************************* * Find an error text * @@ -460,98 +667,245 @@ static const char * find_error_text(int n) { const char *s = error_texts; -for (; n > 0; n--) while (*s++ != 0); +for (; n > 0; n--) + { + while (*s++ != CHAR_NULL) {}; + if (*s == CHAR_NULL) return "Error text not found (please report)"; + } return s; } /************************************************* +* Expand the workspace * +*************************************************/ + +/* This function is called during the second compiling phase, if the number of +forward references fills the existing workspace, which is originally a block on +the stack. A larger block is obtained from malloc() unless the ultimate limit +has been reached or the increase will be rather small. + +Argument: pointer to the compile data block +Returns: 0 if all went well, else an error number +*/ + +static int +expand_workspace(compile_data *cd) +{ +pcre_uchar *newspace; +int newsize = cd->workspace_size * 2; + +if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX; +if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX || + newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN) + return ERR72; + +newspace = (PUBL(malloc))(IN_UCHARS(newsize)); +if (newspace == NULL) return ERR21; +memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar)); +cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace); +if (cd->workspace_size > COMPILE_WORK_SIZE) + (PUBL(free))((void *)cd->start_workspace); +cd->start_workspace = newspace; +cd->workspace_size = newsize; +return 0; +} + + + +/************************************************* +* Check for counted repeat * +*************************************************/ + +/* This function is called when a '{' is encountered in a place where it might +start a quantifier. It looks ahead to see if it really is a quantifier or not. +It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd} +where the ddds are digits. + +Arguments: + p pointer to the first char after '{' + +Returns: TRUE or FALSE +*/ + +static BOOL +is_counted_repeat(const pcre_uchar *p) +{ +if (!IS_DIGIT(*p)) return FALSE; +p++; +while (IS_DIGIT(*p)) p++; +if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE; + +if (*p++ != CHAR_COMMA) return FALSE; +if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE; + +if (!IS_DIGIT(*p)) return FALSE; +p++; +while (IS_DIGIT(*p)) p++; + +return (*p == CHAR_RIGHT_CURLY_BRACKET); +} + + + +/************************************************* * Handle escapes * *************************************************/ /* This function is called when a \ has been encountered. It either returns a -positive value for a simple escape such as \n, or a negative value which -encodes one of the more complicated things such as \d. A backreference to group -n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When -UTF-8 is enabled, a positive value greater than 255 may be returned. On entry, -ptr is pointing at the \. On exit, it is on the final character of the escape -sequence. +positive value for a simple escape such as \n, or 0 for a data character +which will be placed in chptr. A backreference to group n is returned as +negative n. When UTF-8 is enabled, a positive value greater than 255 may +be returned in chptr. +On entry,ptr is pointing at the \. On exit, it is on the final character of the +escape sequence. Arguments: ptrptr points to the pattern position pointer + chptr points to the data character errorcodeptr points to the errorcode variable bracount number of previous extracting brackets options the options bits isclass TRUE if inside a character class -Returns: zero or positive => a data character - negative => a special escape sequence +Returns: zero => a data character + positive => a special escape sequence + negative => a back reference on error, errorcodeptr is set */ static int -check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount, - int options, BOOL isclass) +check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr, + int bracount, int options, BOOL isclass) { -BOOL utf8 = (options & PCRE_UTF8) != 0; -const uschar *ptr = *ptrptr + 1; -int c, i; +/* PCRE_UTF16 has the same value as PCRE_UTF8. */ +BOOL utf = (options & PCRE_UTF8) != 0; +const pcre_uchar *ptr = *ptrptr + 1; +pcre_uint32 c; +int escape = 0; +int i; GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */ ptr--; /* Set pointer back to the last byte */ /* If backslash is at the end of the pattern, it's an error. */ -if (c == 0) *errorcodeptr = ERR1; + +if (c == CHAR_NULL) *errorcodeptr = ERR1; /* Non-alphanumerics are literals. For digits or letters, do an initial lookup in a table. A non-zero result is something that can be returned immediately. Otherwise further processing may be required. */ -#ifndef EBCDIC /* ASCII coding */ -else if (c < '0' || c > 'z') {} /* Not alphanumeric */ -else if ((i = escapes[c - '0']) != 0) c = i; +#ifndef EBCDIC /* ASCII/UTF-8 coding */ +/* Not alphanumeric */ +else if (c < CHAR_0 || c > CHAR_z) {} +else if ((i = escapes[c - CHAR_0]) != 0) + { if (i > 0) c = (pcre_uint32)i; else escape = -i; } #else /* EBCDIC coding */ -else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */ -else if ((i = escapes[c - 0x48]) != 0) c = i; +/* Not alphanumeric */ +else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {} +else if ((i = escapes[c - 0x48]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; } #endif /* Escapes that need further processing, or are illegal. */ else { - const uschar *oldptr; - BOOL braced, negated; + const pcre_uchar *oldptr; + BOOL braced, negated, overflow; + int s; switch (c) { /* A number of Perl escapes are not handled by PCRE. We give an explicit error. */ - case 'l': - case 'L': - case 'N': - case 'u': - case 'U': + case CHAR_l: + case CHAR_L: *errorcodeptr = ERR37; break; - /* \g must be followed by a number, either plain or braced. If positive, it - is an absolute backreference. If negative, it is a relative backreference. - This is a Perl 5.10 feature. Perl 5.10 also supports \g{name} as a - reference to a named group. This is part of Perl's movement towards a - unified syntax for back references. As this is synonymous with \k{name}, we - fudge it up by pretending it really was \k. */ + case CHAR_u: + if ((options & PCRE_JAVASCRIPT_COMPAT) != 0) + { + /* In JavaScript, \u must be followed by four hexadecimal numbers. + Otherwise it is a lowercase u letter. */ + if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0 + && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0 + && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0 + && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0) + { + c = 0; + for (i = 0; i < 4; ++i) + { + register pcre_uint32 cc = *(++ptr); +#ifndef EBCDIC /* ASCII/UTF-8 coding */ + if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ + c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); +#else /* EBCDIC coding */ + if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */ + c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); +#endif + } - case 'g': - if (ptr[1] == '{') +#if defined COMPILE_PCRE8 + if (c > (utf ? 0x10ffffU : 0xffU)) +#elif defined COMPILE_PCRE16 + if (c > (utf ? 0x10ffffU : 0xffffU)) +#elif defined COMPILE_PCRE32 + if (utf && c > 0x10ffffU) +#endif + { + *errorcodeptr = ERR76; + } + else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73; + } + } + else + *errorcodeptr = ERR37; + break; + + case CHAR_U: + /* In JavaScript, \U is an uppercase U letter. */ + if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37; + break; + + /* In a character class, \g is just a literal "g". Outside a character + class, \g must be followed by one of a number of specific things: + + (1) A number, either plain or braced. If positive, it is an absolute + backreference. If negative, it is a relative backreference. This is a Perl + 5.10 feature. + + (2) Perl 5.10 also supports \g{name} as a reference to a named group. This + is part of Perl's movement towards a unified syntax for back references. As + this is synonymous with \k{name}, we fudge it up by pretending it really + was \k. + + (3) For Oniguruma compatibility we also support \g followed by a name or a + number either in angle brackets or in single quotes. However, these are + (possibly recursive) subroutine calls, _not_ backreferences. Just return + the ESC_g code (cf \k). */ + + case CHAR_g: + if (isclass) break; + if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE) { - const uschar *p; - for (p = ptr+2; *p != 0 && *p != '}'; p++) - if (*p != '-' && (digitab[*p] & ctype_digit) == 0) break; - if (*p != 0 && *p != '}') + escape = ESC_g; + break; + } + + /* Handle the Perl-compatible cases */ + + if (ptr[1] == CHAR_LEFT_CURLY_BRACKET) + { + const pcre_uchar *p; + for (p = ptr+2; *p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET; p++) + if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break; + if (*p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET) { - c = -ESC_k; + escape = ESC_k; break; } braced = TRUE; @@ -559,40 +913,56 @@ else } else braced = FALSE; - if (ptr[1] == '-') + if (ptr[1] == CHAR_MINUS) { negated = TRUE; ptr++; } else negated = FALSE; - c = 0; - while ((digitab[ptr[1]] & ctype_digit) != 0) - c = c * 10 + *(++ptr) - '0'; - - if (c < 0) + /* The integer range is limited by the machine's int representation. */ + s = 0; + overflow = FALSE; + while (IS_DIGIT(ptr[1])) { + if (s > INT_MAX / 10 - 1) /* Integer overflow */ + { + overflow = TRUE; + break; + } + s = s * 10 + (int)(*(++ptr) - CHAR_0); + } + if (overflow) /* Integer overflow */ + { + while (IS_DIGIT(ptr[1])) + ptr++; *errorcodeptr = ERR61; break; } - if (c == 0 || (braced && *(++ptr) != '}')) + if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET) { *errorcodeptr = ERR57; break; } + if (s == 0) + { + *errorcodeptr = ERR58; + break; + } + if (negated) { - if (c > bracount) + if (s > bracount) { *errorcodeptr = ERR15; break; } - c = bracount - (c - 1); + s = bracount - (s - 1); } - c = -(ESC_REF + c); + escape = -s; break; /* The handling of escape sequences consisting of a string of digits @@ -607,23 +977,34 @@ else value is greater than 377, the least significant 8 bits are taken. Inside a character class, \ followed by a digit is always an octal number. */ - case '1': case '2': case '3': case '4': case '5': - case '6': case '7': case '8': case '9': + case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5: + case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9: if (!isclass) { oldptr = ptr; - c -= '0'; - while ((digitab[ptr[1]] & ctype_digit) != 0) - c = c * 10 + *(++ptr) - '0'; - if (c < 0) + /* The integer range is limited by the machine's int representation. */ + s = (int)(c -CHAR_0); + overflow = FALSE; + while (IS_DIGIT(ptr[1])) + { + if (s > INT_MAX / 10 - 1) /* Integer overflow */ + { + overflow = TRUE; + break; + } + s = s * 10 + (int)(*(++ptr) - CHAR_0); + } + if (overflow) /* Integer overflow */ { + while (IS_DIGIT(ptr[1])) + ptr++; *errorcodeptr = ERR61; break; } - if (c < 10 || c <= bracount) + if (s < 10 || s <= bracount) { - c = -(ESC_REF + c); + escape = -s; break; } ptr = oldptr; /* Put the pointer back and fall through */ @@ -633,7 +1014,7 @@ else generates a binary zero byte and treats the digit as a following literal. Thus we have to pull back the pointer by one. */ - if ((c = *ptr) >= '8') + if ((c = *ptr) >= CHAR_8) { ptr--; c = 0; @@ -643,45 +1024,87 @@ else /* \0 always starts an octal number, but we may drop through to here with a larger first octal digit. The original code used just to take the least significant 8 bits of octal numbers (I think this is what early Perls used - to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more - than 3 octal digits. */ - - case '0': - c -= '0'; - while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7') - c = c * 8 + *(++ptr) - '0'; - if (!utf8 && c > 255) *errorcodeptr = ERR51; + to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode, + but no more than 3 octal digits. */ + + case CHAR_0: + c -= CHAR_0; + while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7) + c = c * 8 + *(++ptr) - CHAR_0; +#ifdef COMPILE_PCRE8 + if (!utf && c > 0xff) *errorcodeptr = ERR51; +#endif break; /* \x is complicated. \x{ddd} is a character number which can be greater - than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is - treated as a data character. */ + than 0xff in utf or non-8bit mode, but only if the ddd are hex digits. + If not, { is treated as a data character. */ + + case CHAR_x: + if ((options & PCRE_JAVASCRIPT_COMPAT) != 0) + { + /* In JavaScript, \x must be followed by two hexadecimal numbers. + Otherwise it is a lowercase x letter. */ + if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0 + && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0) + { + c = 0; + for (i = 0; i < 2; ++i) + { + register pcre_uint32 cc = *(++ptr); +#ifndef EBCDIC /* ASCII/UTF-8 coding */ + if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ + c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); +#else /* EBCDIC coding */ + if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */ + c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); +#endif + } + } + break; + } - case 'x': - if (ptr[1] == '{') + if (ptr[1] == CHAR_LEFT_CURLY_BRACKET) { - const uschar *pt = ptr + 2; - int count = 0; + const pcre_uchar *pt = ptr + 2; c = 0; - while ((digitab[*pt] & ctype_xdigit) != 0) + overflow = FALSE; + while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) { - register int cc = *pt++; - if (c == 0 && cc == '0') continue; /* Leading zeroes */ - count++; + register pcre_uint32 cc = *pt++; + if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */ -#ifndef EBCDIC /* ASCII coding */ - if (cc >= 'a') cc -= 32; /* Convert to upper case */ - c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10)); +#ifdef COMPILE_PCRE32 + if (c >= 0x10000000l) { overflow = TRUE; break; } +#endif + +#ifndef EBCDIC /* ASCII/UTF-8 coding */ + if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ + c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); #else /* EBCDIC coding */ - if (cc >= 'a' && cc <= 'z') cc += 64; /* Convert to upper case */ - c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10)); + if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */ + c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); +#endif + +#if defined COMPILE_PCRE8 + if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; } +#elif defined COMPILE_PCRE16 + if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; } +#elif defined COMPILE_PCRE32 + if (utf && c > 0x10ffffU) { overflow = TRUE; break; } #endif } - if (*pt == '}') + if (overflow) { - if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34; + while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++; + *errorcodeptr = ERR34; + } + + if (*pt == CHAR_RIGHT_CURLY_BRACKET) + { + if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73; ptr = pt; break; } @@ -693,37 +1116,42 @@ else /* Read just a single-byte hex-defined char */ c = 0; - while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0) + while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0) { - int cc; /* Some compilers don't like ++ */ - cc = *(++ptr); /* in initializers */ -#ifndef EBCDIC /* ASCII coding */ - if (cc >= 'a') cc -= 32; /* Convert to upper case */ - c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10)); + pcre_uint32 cc; /* Some compilers don't like */ + cc = *(++ptr); /* ++ in initializers */ +#ifndef EBCDIC /* ASCII/UTF-8 coding */ + if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ + c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); #else /* EBCDIC coding */ - if (cc <= 'z') cc += 64; /* Convert to upper case */ - c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10)); + if (cc <= CHAR_z) cc += 64; /* Convert to upper case */ + c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); #endif } break; /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped. - This coding is ASCII-specific, but then the whole concept of \cx is + An error is given if the byte following \c is not an ASCII character. This + coding is ASCII-specific, but then the whole concept of \cx is ASCII-specific. (However, an EBCDIC equivalent has now been added.) */ - case 'c': + case CHAR_c: c = *(++ptr); - if (c == 0) + if (c == CHAR_NULL) { *errorcodeptr = ERR2; break; } - -#ifndef EBCDIC /* ASCII coding */ - if (c >= 'a' && c <= 'z') c -= 32; +#ifndef EBCDIC /* ASCII/UTF-8 coding */ + if (c > 127) /* Excludes all non-ASCII in either mode */ + { + *errorcodeptr = ERR68; + break; + } + if (c >= CHAR_a && c <= CHAR_z) c -= 32; c ^= 0x40; -#else /* EBCDIC coding */ - if (c >= 'a' && c <= 'z') c += 64; +#else /* EBCDIC coding */ + if (c >= CHAR_a && c <= CHAR_z) c += 64; c ^= 0xC0; #endif break; @@ -745,11 +1173,25 @@ else } } -*ptrptr = ptr; -return c; -} +/* Perl supports \N{name} for character names, as well as plain \N for "not +newline". PCRE does not support \N{name}. However, it does support +quantification such as \N{2,3}. */ + +if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET && + !is_counted_repeat(ptr+2)) + *errorcodeptr = ERR37; +/* If PCRE_UCP is set, we change the values for \d etc. */ +if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w) + escape += (ESC_DU - ESC_D); + +/* Set the pointer to the final character before returning. */ + +*ptrptr = ptr; +*chptr = c; +return escape; +} #ifdef SUPPORT_UCP /************************************************* @@ -764,42 +1206,45 @@ escape sequence. Argument: ptrptr points to the pattern position pointer negptr points to a boolean that is set TRUE for negation else FALSE - dptr points to an int that is set to the detailed property value + ptypeptr points to an unsigned int that is set to the type value + pdataptr points to an unsigned int that is set to the detailed property value errorcodeptr points to the error code variable -Returns: type value from ucp_type_table, or -1 for an invalid type +Returns: TRUE if the type value was found, or FALSE for an invalid type */ -static int -get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr) +static BOOL +get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, unsigned int *ptypeptr, + unsigned int *pdataptr, int *errorcodeptr) { -int c, i, bot, top; -const uschar *ptr = *ptrptr; -char name[32]; +pcre_uchar c; +int i, bot, top; +const pcre_uchar *ptr = *ptrptr; +pcre_uchar name[32]; c = *(++ptr); -if (c == 0) goto ERROR_RETURN; +if (c == CHAR_NULL) goto ERROR_RETURN; *negptr = FALSE; /* \P or \p can be followed by a name in {}, optionally preceded by ^ for negation. */ -if (c == '{') +if (c == CHAR_LEFT_CURLY_BRACKET) { - if (ptr[1] == '^') + if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT) { *negptr = TRUE; ptr++; } - for (i = 0; i < (int)sizeof(name) - 1; i++) + for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++) { c = *(++ptr); - if (c == 0) goto ERROR_RETURN; - if (c == '}') break; + if (c == CHAR_NULL) goto ERROR_RETURN; + if (c == CHAR_RIGHT_CURLY_BRACKET) break; name[i] = c; } - if (c !='}') goto ERROR_RETURN; + if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN; name[i] = 0; } @@ -816,28 +1261,30 @@ else /* Search for a recognized property name using binary chop */ bot = 0; -top = _erts_pcre_utt_size; +top = PRIV(utt_size); while (bot < top) { + int r; i = (bot + top) >> 1; - c = strcmp(name, _erts_pcre_utt_names + _erts_pcre_utt[i].name_offset); - if (c == 0) + r = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset); + if (r == 0) { - *dptr = _erts_pcre_utt[i].value; - return _erts_pcre_utt[i].type; + *ptypeptr = PRIV(utt)[i].type; + *pdataptr = PRIV(utt)[i].value; + return TRUE; } - if (c > 0) bot = i + 1; else top = i; + if (r > 0) bot = i + 1; else top = i; } *errorcodeptr = ERR47; *ptrptr = ptr; -return -1; +return FALSE; ERROR_RETURN: *errorcodeptr = ERR46; *ptrptr = ptr; -return -1; +return FALSE; } #endif @@ -845,39 +1292,6 @@ return -1; /************************************************* -* Check for counted repeat * -*************************************************/ - -/* This function is called when a '{' is encountered in a place where it might -start a quantifier. It looks ahead to see if it really is a quantifier or not. -It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd} -where the ddds are digits. - -Arguments: - p pointer to the first char after '{' - -Returns: TRUE or FALSE -*/ - -static BOOL -is_counted_repeat(const uschar *p) -{ -if ((digitab[*p++] & ctype_digit) == 0) return FALSE; -while ((digitab[*p] & ctype_digit) != 0) p++; -if (*p == '}') return TRUE; - -if (*p++ != ',') return FALSE; -if (*p == '}') return TRUE; - -if ((digitab[*p++] & ctype_digit) == 0) return FALSE; -while ((digitab[*p] & ctype_digit) != 0) p++; - -return (*p == '}'); -} - - - -/************************************************* * Read repeat counts * *************************************************/ @@ -896,8 +1310,8 @@ Returns: pointer to '}' on success; current ptr on error, with errorcodeptr set non-zero */ -static const uschar * -read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr) +static const pcre_uchar * +read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr) { int min = 0; int max = -1; @@ -905,7 +1319,7 @@ int max = -1; /* Read the minimum value and do a paranoid check: a negative value indicates an integer overflow. */ -while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0'; +while (IS_DIGIT(*p)) min = min * 10 + (int)(*p++ - CHAR_0); if (min < 0 || min > 65535) { *errorcodeptr = ERR5; @@ -915,12 +1329,12 @@ if (min < 0 || min > 65535) /* Read the maximum value if there is one, and again do a paranoid on its size. Also, max must not be less than min. */ -if (*p == '}') max = min; else +if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else { - if (*(++p) != '}') + if (*(++p) != CHAR_RIGHT_CURLY_BRACKET) { max = 0; - while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0'; + while(IS_DIGIT(*p)) max = max * 10 + (int)(*p++ - CHAR_0); if (max < 0 || max > 65535) { *errorcodeptr = ERR5; @@ -945,65 +1359,201 @@ return p; /************************************************* -* Find forward referenced subpattern * +* Subroutine for finding forward reference * *************************************************/ -/* This function scans along a pattern's text looking for capturing +/* This recursive function is called only from find_parens() below. The +top-level call starts at the beginning of the pattern. All other calls must +start at a parenthesis. It scans along a pattern's text looking for capturing subpatterns, and counting them. If it finds a named pattern that matches the name it is given, it returns its number. Alternatively, if the name is NULL, it -returns when it reaches a given numbered subpattern. This is used for forward -references to subpatterns. We know that if (?P< is encountered, the name will -be terminated by '>' because that is checked in the first pass. +returns when it reaches a given numbered subpattern. Recursion is used to keep +track of subpatterns that reset the capturing group numbers - the (?| feature. + +This function was originally called only from the second pass, in which we know +that if (?< or (?' or (?P< is encountered, the name will be correctly +terminated because that is checked in the first pass. There is now one call to +this function in the first pass, to check for a recursive back reference by +name (so that we can make the whole group atomic). In this case, we need check +only up to the current position in the pattern, and that is still OK because +and previous occurrences will have been checked. To make this work, the test +for "end of pattern" is a check against cd->end_pattern in the main loop, +instead of looking for a binary zero. This means that the special first-pass +call can adjust cd->end_pattern temporarily. (Checks for binary zero while +processing items within the loop are OK, because afterwards the main loop will +terminate.) Arguments: - ptr current position in the pattern - count current count of capturing parens so far encountered + ptrptr address of the current character pointer (updated) + cd compile background data name name to seek, or NULL if seeking a numbered subpattern lorn name length, or subpattern number if name is NULL xmode TRUE if we are in /x mode + utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode + count pointer to the current capturing subpattern number (updated) Returns: the number of the named subpattern, or -1 if not found */ static int -find_parens(const uschar *ptr, int count, const uschar *name, int lorn, - BOOL xmode) +find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn, + BOOL xmode, BOOL utf, int *count) { -const uschar *thisname; +pcre_uchar *ptr = *ptrptr; +int start_count = *count; +int hwm_count = start_count; +BOOL dup_parens = FALSE; -for (; *ptr != 0; ptr++) +/* If the first character is a parenthesis, check on the type of group we are +dealing with. The very first call may not start with a parenthesis. */ + +if (ptr[0] == CHAR_LEFT_PARENTHESIS) { - int term; + /* Handle specials such as (*SKIP) or (*UTF8) etc. */ + + if (ptr[1] == CHAR_ASTERISK) + { + ptr += 2; + while (ptr < cd->end_pattern && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++; + } + + /* Handle a normal, unnamed capturing parenthesis. */ + + else if (ptr[1] != CHAR_QUESTION_MARK) + { + *count += 1; + if (name == NULL && *count == lorn) return *count; + ptr++; + } + + /* All cases now have (? at the start. Remember when we are in a group + where the parenthesis numbers are duplicated. */ + + else if (ptr[2] == CHAR_VERTICAL_LINE) + { + ptr += 3; + dup_parens = TRUE; + } + + /* Handle comments; all characters are allowed until a ket is reached. */ + + else if (ptr[2] == CHAR_NUMBER_SIGN) + { + for (ptr += 3; *ptr != CHAR_NULL; ptr++) + if (*ptr == CHAR_RIGHT_PARENTHESIS) break; + goto FAIL_EXIT; + } + + /* Handle a condition. If it is an assertion, just carry on so that it + is processed as normal. If not, skip to the closing parenthesis of the + condition (there can't be any nested parens). */ + + else if (ptr[2] == CHAR_LEFT_PARENTHESIS) + { + ptr += 2; + if (ptr[1] != CHAR_QUESTION_MARK) + { + while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++; + if (*ptr != CHAR_NULL) ptr++; + } + } + + /* Start with (? but not a condition. */ + + else + { + ptr += 2; + if (*ptr == CHAR_P) ptr++; /* Allow optional P */ + + /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */ + if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK && + ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE) + { + pcre_uchar term; + const pcre_uchar *thisname; + *count += 1; + if (name == NULL && *count == lorn) return *count; + term = *ptr++; + if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN; + thisname = ptr; + while (*ptr != term) ptr++; + if (name != NULL && lorn == (int)(ptr - thisname) && + STRNCMP_UC_UC(name, thisname, (unsigned int)lorn) == 0) + return *count; + term++; + } + } + } + +/* Past any initial parenthesis handling, scan for parentheses or vertical +bars. Stop if we get to cd->end_pattern. Note that this is important for the +first-pass call when this value is temporarily adjusted to stop at the current +position. So DO NOT change this to a test for binary zero. */ + +for (; ptr < cd->end_pattern; ptr++) + { /* Skip over backslashed characters and also entire \Q...\E */ - if (*ptr == '\\') + if (*ptr == CHAR_BACKSLASH) { - if (*(++ptr) == 0) return -1; - if (*ptr == 'Q') for (;;) + if (*(++ptr) == CHAR_NULL) goto FAIL_EXIT; + if (*ptr == CHAR_Q) for (;;) { - while (*(++ptr) != 0 && *ptr != '\\'); - if (*ptr == 0) return -1; - if (*(++ptr) == 'E') break; + while (*(++ptr) != CHAR_NULL && *ptr != CHAR_BACKSLASH) {}; + if (*ptr == CHAR_NULL) goto FAIL_EXIT; + if (*(++ptr) == CHAR_E) break; } continue; } - /* Skip over character classes */ + /* Skip over character classes; this logic must be similar to the way they + are handled for real. If the first character is '^', skip it. Also, if the + first few characters (either before or after ^) are \Q\E or \E we skip them + too. This makes for compatibility with Perl. Note the use of STR macros to + encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */ - if (*ptr == '[') + if (*ptr == CHAR_LEFT_SQUARE_BRACKET) { - while (*(++ptr) != ']') + BOOL negate_class = FALSE; + for (;;) { - if (*ptr == 0) return -1; - if (*ptr == '\\') + if (ptr[1] == CHAR_BACKSLASH) { - if (*(++ptr) == 0) return -1; - if (*ptr == 'Q') for (;;) + if (ptr[2] == CHAR_E) + ptr+= 2; + else if (STRNCMP_UC_C8(ptr + 2, + STR_Q STR_BACKSLASH STR_E, 3) == 0) + ptr += 4; + else + break; + } + else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT) + { + negate_class = TRUE; + ptr++; + } + else break; + } + + /* If the next character is ']', it is a data character that must be + skipped, except in JavaScript compatibility mode. */ + + if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET && + (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0) + ptr++; + + while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET) + { + if (*ptr == CHAR_NULL) return -1; + if (*ptr == CHAR_BACKSLASH) + { + if (*(++ptr) == CHAR_NULL) goto FAIL_EXIT; + if (*ptr == CHAR_Q) for (;;) { - while (*(++ptr) != 0 && *ptr != '\\'); - if (*ptr == 0) return -1; - if (*(++ptr) == 'E') break; + while (*(++ptr) != CHAR_NULL && *ptr != CHAR_BACKSLASH) {}; + if (*ptr == CHAR_NULL) goto FAIL_EXIT; + if (*(++ptr) == CHAR_E) break; } continue; } @@ -1013,89 +1563,131 @@ for (; *ptr != 0; ptr++) /* Skip comments in /x mode */ - if (xmode && *ptr == '#') + if (xmode && *ptr == CHAR_NUMBER_SIGN) { - while (*(++ptr) != 0 && *ptr != '\n'); - if (*ptr == 0) return -1; + ptr++; + while (*ptr != CHAR_NULL) + { + if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; } + ptr++; +#ifdef SUPPORT_UTF + if (utf) FORWARDCHAR(ptr); +#endif + } + if (*ptr == CHAR_NULL) goto FAIL_EXIT; continue; } - /* An opening parens must now be a real metacharacter */ + /* Check for the special metacharacters */ - if (*ptr != '(') continue; - if (ptr[1] != '?' && ptr[1] != '*') + if (*ptr == CHAR_LEFT_PARENTHESIS) { - count++; - if (name == NULL && count == lorn) return count; - continue; + int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count); + if (rc > 0) return rc; + if (*ptr == CHAR_NULL) goto FAIL_EXIT; } - ptr += 2; - if (*ptr == 'P') ptr++; /* Allow optional P */ + else if (*ptr == CHAR_RIGHT_PARENTHESIS) + { + if (dup_parens && *count < hwm_count) *count = hwm_count; + goto FAIL_EXIT; + } - /* We have to disambiguate (?<! and (?<= from (?<name> */ + else if (*ptr == CHAR_VERTICAL_LINE && dup_parens) + { + if (*count > hwm_count) hwm_count = *count; + *count = start_count; + } + } - if ((*ptr != '<' || ptr[1] == '!' || ptr[1] == '=') && - *ptr != '\'') - continue; +FAIL_EXIT: +*ptrptr = ptr; +return -1; +} - count++; - if (name == NULL && count == lorn) return count; - term = *ptr++; - if (term == '<') term = '>'; - thisname = ptr; - while (*ptr != term) ptr++; - if (name != NULL && lorn == ptr - thisname && - strncmp((const char *)name, (const char *)thisname, lorn) == 0) - return count; + + +/************************************************* +* Find forward referenced subpattern * +*************************************************/ + +/* This function scans along a pattern's text looking for capturing +subpatterns, and counting them. If it finds a named pattern that matches the +name it is given, it returns its number. Alternatively, if the name is NULL, it +returns when it reaches a given numbered subpattern. This is used for forward +references to subpatterns. We used to be able to start this scan from the +current compiling point, using the current count value from cd->bracount, and +do it all in a single loop, but the addition of the possibility of duplicate +subpattern numbers means that we have to scan from the very start, in order to +take account of such duplicates, and to use a recursive function to keep track +of the different types of group. + +Arguments: + cd compile background data + name name to seek, or NULL if seeking a numbered subpattern + lorn name length, or subpattern number if name is NULL + xmode TRUE if we are in /x mode + utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode + +Returns: the number of the found subpattern, or -1 if not found +*/ + +static int +find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode, + BOOL utf) +{ +pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern; +int count = 0; +int rc; + +/* If the pattern does not start with an opening parenthesis, the first call +to find_parens_sub() will scan right to the end (if necessary). However, if it +does start with a parenthesis, find_parens_sub() will return when it hits the +matching closing parens. That is why we have to have a loop. */ + +for (;;) + { + rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count); + if (rc > 0 || *ptr++ == CHAR_NULL) break; } -return -1; +return rc; } + /************************************************* * Find first significant op code * *************************************************/ /* This is called by several functions that scan a compiled expression looking for a fixed first character, or an anchoring op code etc. It skips over things -that do not influence this. For some calls, a change of option is important. -For some calls, it makes sense to skip negative forward and all backward -assertions, and also the \b assertion; for others it does not. +that do not influence this. For some calls, it makes sense to skip negative +forward and all backward assertions, and also the \b assertion; for others it +does not. Arguments: code pointer to the start of the group - options pointer to external options - optbit the option bit whose changing is significant, or - zero if none are skipassert TRUE if certain assertions are to be skipped Returns: pointer to the first significant opcode */ -static const uschar* -first_significant_code(const uschar *code, int *options, int optbit, - BOOL skipassert) +static const pcre_uchar* +first_significant_code(const pcre_uchar *code, BOOL skipassert) { for (;;) { switch ((int)*code) { - case OP_OPT: - if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit)) - *options = (int)code[1]; - code += 2; - break; - case OP_ASSERT_NOT: case OP_ASSERTBACK: case OP_ASSERTBACK_NOT: if (!skipassert) return code; do code += GET(code, 1); while (*code == OP_ALT); - code += _erts_pcre_OP_lengths[*code]; + code += PRIV(OP_lengths)[*code]; break; case OP_WORD_BOUNDARY: @@ -1105,9 +1697,11 @@ for (;;) case OP_CALLOUT: case OP_CREF: + case OP_NCREF: case OP_RREF: + case OP_NRREF: case OP_DEF: - code += _erts_pcre_OP_lengths[*code]; + code += PRIV(OP_lengths)[*code]; break; default: @@ -1121,28 +1715,40 @@ for (;;) /************************************************* -* Find the fixed length of a pattern * +* Find the fixed length of a branch * *************************************************/ -/* Scan a pattern and compute the fixed length of subject that will match it, +/* Scan a branch and compute the fixed length of subject that will match it, if the length is fixed. This is needed for dealing with backward assertions. -In UTF8 mode, the result is in characters rather than bytes. +In UTF8 mode, the result is in characters rather than bytes. The branch is +temporarily terminated with OP_END when this function is called. + +This function is called when a backward assertion is encountered, so that if it +fails, the error message can point to the correct place in the pattern. +However, we cannot do this when the assertion contains subroutine calls, +because they can be forward references. We solve this by remembering this case +and doing the check at the end; a flag specifies which mode we are running in. Arguments: code points to the start of the pattern (the bracket) - options the compiling options - -Returns: the fixed length, or -1 if there is no fixed length, - or -2 if \C was encountered + utf TRUE in UTF-8 / UTF-16 / UTF-32 mode + atend TRUE if called when the pattern is complete + cd the "compile data" structure + +Returns: the fixed length, + or -1 if there is no fixed length, + or -2 if \C was encountered (in UTF-8 mode only) + or -3 if an OP_RECURSE item was encountered and atend is FALSE + or -4 if an unknown opcode was encountered (internal error) */ static int -find_fixedlength(uschar *code, int options) +find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd) { int length = -1; register int branchlength = 0; -register uschar *cc = code + 1 + LINK_SIZE; +register pcre_uchar *cc = code + 1 + LINK_SIZE; /* Scan along the opcodes for this branch. If we get to the end of the branch, check the length against that of the other branches. */ @@ -1150,29 +1756,39 @@ branch, check the length against that of the other branches. */ for (;;) { int d; - register int op = *cc; + pcre_uchar *ce, *cs; + register pcre_uchar op = *cc; + switch (op) { + /* We only need to continue for OP_CBRA (normal capturing bracket) and + OP_BRA (normal non-capturing bracket) because the other variants of these + opcodes are all concerned with unlimited repeated groups, which of course + are not of fixed length. */ + case OP_CBRA: case OP_BRA: case OP_ONCE: + case OP_ONCE_NC: case OP_COND: - d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options); + d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd); if (d < 0) return d; branchlength += d; do cc += GET(cc, 1); while (*cc == OP_ALT); cc += 1 + LINK_SIZE; break; - /* Reached end of a branch; if it's a ket it is the end of a nested - call. If it's ALT it is an alternation in a nested call. If it is - END it's the end of the outer call. All can be handled by the same code. */ + /* Reached end of a branch; if it's a ket it is the end of a nested call. + If it's ALT it is an alternation in a nested call. An ACCEPT is effectively + an ALT. If it is END it's the end of the outer call. All can be handled by + the same code. Note that we must not include the OP_KETRxxx opcodes here, + because they all imply an unlimited repeat. */ case OP_ALT: case OP_KET: - case OP_KETRMAX: - case OP_KETRMIN: case OP_END: + case OP_ACCEPT: + case OP_ASSERT_ACCEPT: if (length < 0) length = branchlength; else if (length != branchlength) return -1; if (*cc != OP_ALT) return length; @@ -1180,6 +1796,21 @@ for (;;) branchlength = 0; break; + /* A true recursion implies not fixed length, but a subroutine call may + be OK. If the subroutine is a forward reference, we can't deal with + it until the end of the pattern, so return -3. */ + + case OP_RECURSE: + if (!atend) return -3; + cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */ + do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */ + if (cc > cs && cc < ce) return -1; /* Recursion */ + d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd); + if (d < 0) return d; + branchlength += d; + cc += 1 + LINK_SIZE; + break; + /* Skip over assertive subpatterns */ case OP_ASSERT: @@ -1187,39 +1818,55 @@ for (;;) case OP_ASSERTBACK: case OP_ASSERTBACK_NOT: do cc += GET(cc, 1); while (*cc == OP_ALT); - /* Fall through */ + cc += PRIV(OP_lengths)[*cc]; + break; /* Skip over things that don't match chars */ - case OP_REVERSE: + case OP_MARK: + case OP_PRUNE_ARG: + case OP_SKIP_ARG: + case OP_THEN_ARG: + cc += cc[1] + PRIV(OP_lengths)[*cc]; + break; + + case OP_CALLOUT: + case OP_CIRC: + case OP_CIRCM: + case OP_CLOSE: + case OP_COMMIT: case OP_CREF: - case OP_RREF: case OP_DEF: - case OP_OPT: - case OP_CALLOUT: - case OP_SOD: - case OP_SOM: + case OP_DOLL: + case OP_DOLLM: case OP_EOD: case OP_EODN: - case OP_CIRC: - case OP_DOLL: + case OP_FAIL: + case OP_NCREF: + case OP_NRREF: case OP_NOT_WORD_BOUNDARY: + case OP_PRUNE: + case OP_REVERSE: + case OP_RREF: + case OP_SET_SOM: + case OP_SKIP: + case OP_SOD: + case OP_SOM: + case OP_THEN: case OP_WORD_BOUNDARY: - cc += _erts_pcre_OP_lengths[*cc]; + cc += PRIV(OP_lengths)[*cc]; break; /* Handle literal characters */ case OP_CHAR: - case OP_CHARNC: + case OP_CHARI: case OP_NOT: + case OP_NOTI: branchlength++; cc += 2; -#ifdef SUPPORT_UTF8 - if ((options & PCRE_UTF8) != 0) - { - while ((*cc & 0xc0) == 0x80) cc++; - } +#ifdef SUPPORT_UTF + if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]); #endif break; @@ -1227,20 +1874,21 @@ for (;;) need to skip over a multibyte character in UTF8 mode. */ case OP_EXACT: - branchlength += GET2(cc,1); - cc += 4; -#ifdef SUPPORT_UTF8 - if ((options & PCRE_UTF8) != 0) - { - while((*cc & 0x80) == 0x80) cc++; - } + case OP_EXACTI: + case OP_NOTEXACT: + case OP_NOTEXACTI: + branchlength += (int)GET2(cc,1); + cc += 2 + IMM2_SIZE; +#ifdef SUPPORT_UTF + if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]); #endif break; case OP_TYPEEXACT: branchlength += GET2(cc,1); - if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2; - cc += 4; + if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) + cc += 2; + cc += 1 + IMM2_SIZE + 1; break; /* Handle single-char matchers */ @@ -1250,6 +1898,10 @@ for (;;) cc += 2; /* Fall through */ + case OP_HSPACE: + case OP_VSPACE: + case OP_NOT_HSPACE: + case OP_NOT_VSPACE: case OP_NOT_DIGIT: case OP_DIGIT: case OP_NOT_WHITESPACE: @@ -1257,29 +1909,37 @@ for (;;) case OP_NOT_WORDCHAR: case OP_WORDCHAR: case OP_ANY: + case OP_ALLANY: branchlength++; cc++; break; - /* The single-byte matcher isn't allowed */ + /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode; + otherwise \C is coded as OP_ALLANY. */ case OP_ANYBYTE: return -2; /* Check a class for variable quantification */ -#ifdef SUPPORT_UTF8 - case OP_XCLASS: - cc += GET(cc, 1) - 33; - /* Fall through */ -#endif - case OP_CLASS: case OP_NCLASS: - cc += 33; +#if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32 + case OP_XCLASS: + /* The original code caused an unsigned overflow in 64 bit systems, + so now we use a conditional statement. */ + if (op == OP_XCLASS) + cc += GET(cc, 1); + else + cc += PRIV(OP_lengths)[OP_CLASS]; +#else + cc += PRIV(OP_lengths)[OP_CLASS]; +#endif switch (*cc) { + case OP_CRPLUS: + case OP_CRMINPLUS: case OP_CRSTAR: case OP_CRMINSTAR: case OP_CRQUERY: @@ -1288,9 +1948,9 @@ for (;;) case OP_CRRANGE: case OP_CRMINRANGE: - if (GET2(cc,1) != GET2(cc,3)) return -1; - branchlength += GET2(cc,1); - cc += 5; + if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1; + branchlength += (int)GET2(cc,1); + cc += 1 + 2 * IMM2_SIZE; break; default: @@ -1300,8 +1960,91 @@ for (;;) /* Anything else is variable length */ - default: + case OP_ANYNL: + case OP_BRAMINZERO: + case OP_BRAPOS: + case OP_BRAPOSZERO: + case OP_BRAZERO: + case OP_CBRAPOS: + case OP_EXTUNI: + case OP_KETRMAX: + case OP_KETRMIN: + case OP_KETRPOS: + case OP_MINPLUS: + case OP_MINPLUSI: + case OP_MINQUERY: + case OP_MINQUERYI: + case OP_MINSTAR: + case OP_MINSTARI: + case OP_MINUPTO: + case OP_MINUPTOI: + case OP_NOTMINPLUS: + case OP_NOTMINPLUSI: + case OP_NOTMINQUERY: + case OP_NOTMINQUERYI: + case OP_NOTMINSTAR: + case OP_NOTMINSTARI: + case OP_NOTMINUPTO: + case OP_NOTMINUPTOI: + case OP_NOTPLUS: + case OP_NOTPLUSI: + case OP_NOTPOSPLUS: + case OP_NOTPOSPLUSI: + case OP_NOTPOSQUERY: + case OP_NOTPOSQUERYI: + case OP_NOTPOSSTAR: + case OP_NOTPOSSTARI: + case OP_NOTPOSUPTO: + case OP_NOTPOSUPTOI: + case OP_NOTQUERY: + case OP_NOTQUERYI: + case OP_NOTSTAR: + case OP_NOTSTARI: + case OP_NOTUPTO: + case OP_NOTUPTOI: + case OP_PLUS: + case OP_PLUSI: + case OP_POSPLUS: + case OP_POSPLUSI: + case OP_POSQUERY: + case OP_POSQUERYI: + case OP_POSSTAR: + case OP_POSSTARI: + case OP_POSUPTO: + case OP_POSUPTOI: + case OP_QUERY: + case OP_QUERYI: + case OP_REF: + case OP_REFI: + case OP_SBRA: + case OP_SBRAPOS: + case OP_SCBRA: + case OP_SCBRAPOS: + case OP_SCOND: + case OP_SKIPZERO: + case OP_STAR: + case OP_STARI: + case OP_TYPEMINPLUS: + case OP_TYPEMINQUERY: + case OP_TYPEMINSTAR: + case OP_TYPEMINUPTO: + case OP_TYPEPLUS: + case OP_TYPEPOSPLUS: + case OP_TYPEPOSQUERY: + case OP_TYPEPOSSTAR: + case OP_TYPEPOSUPTO: + case OP_TYPEQUERY: + case OP_TYPESTAR: + case OP_TYPEUPTO: + case OP_UPTO: + case OP_UPTOI: return -1; + + /* Catch unrecognized opcodes so that when new ones are added they + are not forgotten, as has happened in the past. */ + + default: + return -4; } } /* Control never gets here */ @@ -1311,26 +2054,30 @@ for (;;) /************************************************* -* Scan compiled regex for numbered bracket * +* Scan compiled regex for specific bracket * *************************************************/ /* This little function scans through a compiled pattern until it finds a -capturing bracket with the given number. +capturing bracket with the given number, or, if the number is negative, an +instance of OP_REVERSE for a lookbehind. The function is global in the C sense +so that it can be called from pcre_study() when finding the minimum matching +length. Arguments: code points to start of expression - utf8 TRUE in UTF-8 mode - number the required bracket number + utf TRUE in UTF-8 / UTF-16 / UTF-32 mode + number the required bracket number or negative to find a lookbehind Returns: pointer to the opcode for the bracket, or NULL if not found */ -static const uschar * -find_bracket(const uschar *code, BOOL utf8, int number) +const pcre_uchar * +PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number) { for (;;) { - register int c = *code; + register pcre_uchar c = *code; + if (c == OP_END) return NULL; /* XCLASS is used for classes that cannot be represented just by a bit @@ -1339,18 +2086,28 @@ for (;;) if (c == OP_XCLASS) code += GET(code, 1); + /* Handle recursion */ + + else if (c == OP_REVERSE) + { + if (number < 0) return (pcre_uchar *)code; + code += PRIV(OP_lengths)[c]; + } + /* Handle capturing bracket */ - else if (c == OP_CBRA) + else if (c == OP_CBRA || c == OP_SCBRA || + c == OP_CBRAPOS || c == OP_SCBRAPOS) { - int n = GET2(code, 1+LINK_SIZE); - if (n == number) return (uschar *)code; - code += _erts_pcre_OP_lengths[c]; + int n = (int)GET2(code, 1+LINK_SIZE); + if (n == number) return (pcre_uchar *)code; + code += PRIV(OP_lengths)[c]; } /* Otherwise, we can get the item's length from the table, except that for repeated character types, we have to test for \p and \P, which have an extra - two bytes of parameters. */ + two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we + must add in its length. */ else { @@ -1372,39 +2129,62 @@ for (;;) case OP_TYPEMINUPTO: case OP_TYPEEXACT: case OP_TYPEPOSUPTO: - if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2; + if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP) + code += 2; + break; + + case OP_MARK: + case OP_PRUNE_ARG: + case OP_SKIP_ARG: + case OP_THEN_ARG: + code += code[1]; break; } /* Add in the fixed length from the table */ - code += _erts_pcre_OP_lengths[c]; + code += PRIV(OP_lengths)[c]; /* In UTF-8 mode, opcodes that are followed by a character may be followed by a multi-byte character. The length in the table is a minimum, so we have to arrange to skip the extra bytes. */ -#ifdef SUPPORT_UTF8 - if (utf8) switch(c) +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf) switch(c) { case OP_CHAR: - case OP_CHARNC: + case OP_CHARI: case OP_EXACT: + case OP_EXACTI: case OP_UPTO: + case OP_UPTOI: case OP_MINUPTO: + case OP_MINUPTOI: case OP_POSUPTO: + case OP_POSUPTOI: case OP_STAR: + case OP_STARI: case OP_MINSTAR: + case OP_MINSTARI: case OP_POSSTAR: + case OP_POSSTARI: case OP_PLUS: + case OP_PLUSI: case OP_MINPLUS: + case OP_MINPLUSI: case OP_POSPLUS: + case OP_POSPLUSI: case OP_QUERY: + case OP_QUERYI: case OP_MINQUERY: + case OP_MINQUERYI: case OP_POSQUERY: - if (code[-1] >= 0xc0) code += _erts_pcre_utf8_table4[code[-1] & 0x3f]; + case OP_POSQUERYI: + if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]); break; } +#else + (void)(utf); /* Keep compiler happy by referencing function argument */ #endif } } @@ -1421,17 +2201,17 @@ instance of OP_RECURSE. Arguments: code points to start of expression - utf8 TRUE in UTF-8 mode + utf TRUE in UTF-8 / UTF-16 / UTF-32 mode Returns: pointer to the opcode for OP_RECURSE, or NULL if not found */ -static const uschar * -find_recurse(const uschar *code, BOOL utf8) +static const pcre_uchar * +find_recurse(const pcre_uchar *code, BOOL utf) { for (;;) { - register int c = *code; + register pcre_uchar c = *code; if (c == OP_END) return NULL; if (c == OP_RECURSE) return code; @@ -1443,7 +2223,8 @@ for (;;) /* Otherwise, we can get the item's length from the table, except that for repeated character types, we have to test for \p and \P, which have an extra - two bytes of parameters. */ + two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we + must add in its length. */ else { @@ -1465,39 +2246,90 @@ for (;;) case OP_TYPEUPTO: case OP_TYPEMINUPTO: case OP_TYPEEXACT: - if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2; + if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP) + code += 2; + break; + + case OP_MARK: + case OP_PRUNE_ARG: + case OP_SKIP_ARG: + case OP_THEN_ARG: + code += code[1]; break; } /* Add in the fixed length from the table */ - code += _erts_pcre_OP_lengths[c]; + code += PRIV(OP_lengths)[c]; /* In UTF-8 mode, opcodes that are followed by a character may be followed by a multi-byte character. The length in the table is a minimum, so we have to arrange to skip the extra bytes. */ -#ifdef SUPPORT_UTF8 - if (utf8) switch(c) +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf) switch(c) { case OP_CHAR: - case OP_CHARNC: + case OP_CHARI: + case OP_NOT: + case OP_NOTI: case OP_EXACT: + case OP_EXACTI: + case OP_NOTEXACT: + case OP_NOTEXACTI: case OP_UPTO: + case OP_UPTOI: + case OP_NOTUPTO: + case OP_NOTUPTOI: case OP_MINUPTO: + case OP_MINUPTOI: + case OP_NOTMINUPTO: + case OP_NOTMINUPTOI: case OP_POSUPTO: + case OP_POSUPTOI: + case OP_NOTPOSUPTO: + case OP_NOTPOSUPTOI: case OP_STAR: + case OP_STARI: + case OP_NOTSTAR: + case OP_NOTSTARI: case OP_MINSTAR: + case OP_MINSTARI: + case OP_NOTMINSTAR: + case OP_NOTMINSTARI: case OP_POSSTAR: + case OP_POSSTARI: + case OP_NOTPOSSTAR: + case OP_NOTPOSSTARI: case OP_PLUS: + case OP_PLUSI: + case OP_NOTPLUS: + case OP_NOTPLUSI: case OP_MINPLUS: + case OP_MINPLUSI: + case OP_NOTMINPLUS: + case OP_NOTMINPLUSI: case OP_POSPLUS: + case OP_POSPLUSI: + case OP_NOTPOSPLUS: + case OP_NOTPOSPLUSI: case OP_QUERY: + case OP_QUERYI: + case OP_NOTQUERY: + case OP_NOTQUERYI: case OP_MINQUERY: + case OP_MINQUERYI: + case OP_NOTMINQUERY: + case OP_NOTMINQUERYI: case OP_POSQUERY: - if (code[-1] >= 0xc0) code += _erts_pcre_utf8_table4[code[-1] & 0x3f]; + case OP_POSQUERYI: + case OP_NOTPOSQUERY: + case OP_NOTPOSQUERYI: + if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]); break; } +#else + (void)(utf); /* Keep compiler happy by referencing function argument */ #endif } } @@ -1520,20 +2352,22 @@ bracket whose current branch will already have been scanned. Arguments: code points to start of search endcode points to where to stop - utf8 TRUE if in UTF8 mode + utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode + cd contains pointers to tables etc. Returns: TRUE if what is matched could be empty */ static BOOL -could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8) +could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode, + BOOL utf, compile_data *cd) { -register int c; -for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, TRUE); +register pcre_uchar c; +for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE); code < endcode; - code = first_significant_code(code + _erts_pcre_OP_lengths[c], NULL, 0, TRUE)) + code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE)) { - const uschar *ccode; + const pcre_uchar *ccode; c = *code; @@ -1547,11 +2381,63 @@ for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, continue; } + /* For a recursion/subroutine call, if its end has been reached, which + implies a backward reference subroutine call, we can scan it. If it's a + forward reference subroutine call, we can't. To detect forward reference + we have to scan up the list that is kept in the workspace. This function is + called only when doing the real compile, not during the pre-compile that + measures the size of the compiled pattern. */ + + if (c == OP_RECURSE) + { + const pcre_uchar *scode; + BOOL empty_branch; + + /* Test for forward reference */ + + for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE) + if ((int)GET(scode, 0) == (int)(code + 1 - cd->start_code)) return TRUE; + + /* Not a forward reference, test for completed backward reference */ + + empty_branch = FALSE; + scode = cd->start_code + GET(code, 1); + if (GET(scode, 1) == 0) return TRUE; /* Unclosed */ + + /* Completed backwards reference */ + + do + { + if (could_be_empty_branch(scode, endcode, utf, cd)) + { + empty_branch = TRUE; + break; + } + scode += GET(scode, 1); + } + while (*scode == OP_ALT); + + if (!empty_branch) return FALSE; /* All branches are non-empty */ + continue; + } + /* Groups with zero repeats can of course be empty; skip them. */ - if (c == OP_BRAZERO || c == OP_BRAMINZERO) + if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO || + c == OP_BRAPOSZERO) + { + code += PRIV(OP_lengths)[c]; + do code += GET(code, 1); while (*code == OP_ALT); + c = *code; + continue; + } + + /* A nested group that is already marked as "could be empty" can just be + skipped. */ + + if (c == OP_SBRA || c == OP_SBRAPOS || + c == OP_SCBRA || c == OP_SCBRAPOS) { - code += _erts_pcre_OP_lengths[c]; do code += GET(code, 1); while (*code == OP_ALT); c = *code; continue; @@ -1559,22 +2445,33 @@ for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, /* For other groups, scan the branches. */ - if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND) + if (c == OP_BRA || c == OP_BRAPOS || + c == OP_CBRA || c == OP_CBRAPOS || + c == OP_ONCE || c == OP_ONCE_NC || + c == OP_COND) { BOOL empty_branch; if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */ - /* Scan a closed bracket */ + /* If a conditional group has only one branch, there is a second, implied, + empty branch, so just skip over the conditional, because it could be empty. + Otherwise, scan the individual branches of the group. */ - empty_branch = FALSE; - do - { - if (!empty_branch && could_be_empty_branch(code, endcode, utf8)) - empty_branch = TRUE; + if (c == OP_COND && code[GET(code, 1)] != OP_ALT) code += GET(code, 1); + else + { + empty_branch = FALSE; + do + { + if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd)) + empty_branch = TRUE; + code += GET(code, 1); + } + while (*code == OP_ALT); + if (!empty_branch) return FALSE; /* All branches are non-empty */ } - while (*code == OP_ALT); - if (!empty_branch) return FALSE; /* All branches are non-empty */ + c = *code; continue; } @@ -1585,11 +2482,11 @@ for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, { /* Check for quantifiers after a class. XCLASS is used for classes that cannot be represented just by a bit map. This includes negated single - high-valued characters. The length in _erts_pcre_OP_lengths[] is zero; the + high-valued characters. The length in PRIV(OP_lengths)[] is zero; the actual length is stored in the compiled code, so we must update "code" here. */ -#ifdef SUPPORT_UTF8 +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 case OP_XCLASS: ccode = code += GET(code, 1); goto CHECK_CLASS_REPEAT; @@ -1597,9 +2494,9 @@ for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, case OP_CLASS: case OP_NCLASS: - ccode = code + 33; + ccode = code + PRIV(OP_lengths)[OP_CLASS]; -#ifdef SUPPORT_UTF8 +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 CHECK_CLASS_REPEAT: #endif @@ -1635,10 +2532,12 @@ for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, case OP_NOT_WORDCHAR: case OP_WORDCHAR: case OP_ANY: + case OP_ALLANY: case OP_ANYBYTE: case OP_CHAR: - case OP_CHARNC: + case OP_CHARI: case OP_NOT: + case OP_NOTI: case OP_PLUS: case OP_MINPLUS: case OP_POSPLUS: @@ -1670,7 +2569,8 @@ for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, case OP_TYPEUPTO: case OP_TYPEMINUPTO: case OP_TYPEPOSUPTO: - if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2; + if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP) + code += 2; break; /* End of branch */ @@ -1678,25 +2578,53 @@ for (code = first_significant_code(code + _erts_pcre_OP_lengths[*code], NULL, 0, case OP_KET: case OP_KETRMAX: case OP_KETRMIN: + case OP_KETRPOS: case OP_ALT: return TRUE; /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO, MINUPTO, and POSUPTO may be followed by a multibyte character */ -#ifdef SUPPORT_UTF8 +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 case OP_STAR: + case OP_STARI: case OP_MINSTAR: + case OP_MINSTARI: case OP_POSSTAR: + case OP_POSSTARI: case OP_QUERY: + case OP_QUERYI: case OP_MINQUERY: + case OP_MINQUERYI: case OP_POSQUERY: + case OP_POSQUERYI: + if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]); + break; + case OP_UPTO: + case OP_UPTOI: case OP_MINUPTO: + case OP_MINUPTOI: case OP_POSUPTO: - if (utf8) while ((code[2] & 0xc0) == 0x80) code++; + case OP_POSUPTOI: + if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]); break; #endif + + /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument + string. */ + + case OP_MARK: + case OP_PRUNE_ARG: + case OP_SKIP_ARG: + case OP_THEN_ARG: + code += code[1]; + break; + + /* None of the remaining opcodes are required to match a character. */ + + default: + break; } } @@ -1713,23 +2641,27 @@ return TRUE; the current branch of the current pattern to see if it could match the empty string. If it could, we must look outwards for branches at other levels, stopping when we pass beyond the bracket which is the subject of the recursion. +This function is called only during the real compile, not during the +pre-compile. Arguments: code points to start of the recursion endcode points to where to stop (current RECURSE item) bcptr points to the chain of current (unclosed) branch starts - utf8 TRUE if in UTF-8 mode + utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode + cd pointers to tables etc Returns: TRUE if what is matched could be empty */ static BOOL -could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr, - BOOL utf8) +could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode, + branch_chain *bcptr, BOOL utf, compile_data *cd) { -while (bcptr != NULL && bcptr->current >= code) +while (bcptr != NULL && bcptr->current_branch >= code) { - if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE; + if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd)) + return FALSE; bcptr = bcptr->outer; } return TRUE; @@ -1761,6 +2693,17 @@ where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does, I think. +A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not. +It seems that the appearance of a nested POSIX class supersedes an apparent +external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or +a digit. + +In Perl, unescaped square brackets may also appear as part of class names. For +example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for +[:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not +seem right at all. PCRE does not allow closing square brackets in POSIX class +names. + Arguments: ptr pointer to the initial [ endptr where to return the end pointer @@ -1769,20 +2712,27 @@ Returns: TRUE or FALSE */ static BOOL -check_posix_syntax(const uschar *ptr, const uschar **endptr) +check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr) { -int terminator; /* Don't combine these lines; the Solaris cc */ +pcre_uchar terminator; /* Don't combine these lines; the Solaris cc */ terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */ -for (++ptr; *ptr != 0; ptr++) +for (++ptr; *ptr != CHAR_NULL; ptr++) { - if (*ptr == '\\' && ptr[1] == ']') ptr++; else + if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) + ptr++; + else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE; + else { - if (*ptr == ']') return FALSE; - if (*ptr == terminator && ptr[1] == ']') + if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) { *endptr = ptr; return TRUE; } + if (*ptr == CHAR_LEFT_SQUARE_BRACKET && + (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT || + ptr[1] == CHAR_EQUALS_SIGN) && + check_posix_syntax(ptr, endptr)) + return FALSE; } } return FALSE; @@ -1806,14 +2756,14 @@ Returns: a value representing the name, or -1 if unknown */ static int -check_posix_name(const uschar *ptr, int len) +check_posix_name(const pcre_uchar *ptr, int len) { const char *pn = posix_names; register int yield = 0; while (posix_name_lengths[yield] != 0) { if (len == posix_name_lengths[yield] && - strncmp((const char *)ptr, pn, len) == 0) return yield; + STRNCMP_UC_C8(ptr, pn, (unsigned int)len) == 0) return yield; pn += posix_name_lengths[yield] + 1; yield++; } @@ -1829,11 +2779,12 @@ return -1; that is referenced. This means that groups can be replicated for fixed repetition simply by copying (because the recursion is allowed to refer to earlier groups that are outside the current group). However, when a group is -optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before -it, after it has been compiled. This means that any OP_RECURSE items within it -that refer to the group itself or any contained groups have to have their -offsets adjusted. That one of the jobs of this function. Before it is called, -the partially compiled regex must be temporarily terminated with OP_END. +optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is +inserted before it, after it has been compiled. This means that any OP_RECURSE +items within it that refer to the group itself or any contained groups have to +have their offsets adjusted. That one of the jobs of this function. Before it +is called, the partially compiled regex must be temporarily terminated with +OP_END. This function has been extended with the possibility of forward references for recursions and subroutine calls. It must also check the list of such references @@ -1844,7 +2795,7 @@ value in the reference (which is a group number). Arguments: group points to the start of the group adjust the amount by which the group is to be moved - utf8 TRUE in UTF-8 mode + utf TRUE in UTF-8 / UTF-16 / UTF-32 mode cd contains pointers to tables etc. save_hwm the hwm forward reference pointer at the start of the group @@ -1852,22 +2803,22 @@ Returns: nothing */ static void -adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd, - uschar *save_hwm) +adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd, + pcre_uchar *save_hwm) { -uschar *ptr = group; +pcre_uchar *ptr = group; -while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL) +while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL) { int offset; - uschar *hc; + pcre_uchar *hc; /* See if this recursion is on the forward reference list. If so, adjust the reference. */ for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE) { - offset = GET(hc, 0); + offset = (int)GET(hc, 0); if (cd->start_code + offset == ptr + 1) { PUT(hc, 0, offset + adjust); @@ -1880,7 +2831,7 @@ while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL) if (hc >= cd->hwm) { - offset = GET(ptr, 1); + offset = (int)GET(ptr, 1); if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust); } @@ -1905,14 +2856,14 @@ Arguments: Returns: new code pointer */ -static uschar * -auto_callout(uschar *code, const uschar *ptr, compile_data *cd) +static pcre_uchar * +auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd) { *code++ = OP_CALLOUT; *code++ = 255; -PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */ -PUT(code, LINK_SIZE, 0); /* Default length */ -return code + 2*LINK_SIZE; +PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */ +PUT(code, LINK_SIZE, 0); /* Default length */ +return code + 2 * LINK_SIZE; } @@ -1934,9 +2885,9 @@ Returns: nothing */ static void -complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd) +complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd) { -int length = ptr - cd->start_pattern - GET(previous_callout, 2); +int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2)); PUT(previous_callout, 2 + LINK_SIZE, length); } @@ -1948,9 +2899,10 @@ PUT(previous_callout, 2 + LINK_SIZE, length); *************************************************/ /* This function is passed the start and end of a class range, in UTF-8 mode -with UCP support. It searches up the characters, looking for internal ranges of +with UCP support. It searches up the characters, looking for ranges of characters in the "other" case. Each call returns the next one, updating the -start address. +start address. A character with multiple other cases is returned on its own +with a special return value. Arguments: cptr points to starting character value; updated @@ -1958,33 +2910,127 @@ Arguments: ocptr where to put start of othercase range odptr where to put end of othercase range -Yield: TRUE when range returned; FALSE when no more +Yield: -1 when no more + 0 when a range is returned + >0 the CASESET offset for char with multiple other cases + in this case, ocptr contains the original */ -static BOOL -get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr, - unsigned int *odptr) +static int +get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr, + pcre_uint32 *odptr) { -unsigned int c, othercase, next; +pcre_uint32 c, othercase, next; +unsigned int co; + +/* Find the first character that has an other case. If it has multiple other +cases, return its case offset value. */ for (c = *cptr; c <= d; c++) - { if ((othercase = _erts_pcre_ucp_othercase(c)) != NOTACHAR) break; } + { + if ((co = UCD_CASESET(c)) != 0) + { + *ocptr = c++; /* Character that has the set */ + *cptr = c; /* Rest of input range */ + return (int)co; + } + if ((othercase = UCD_OTHERCASE(c)) != c) break; + } -if (c > d) return FALSE; +if (c > d) return -1; /* Reached end of range */ *ocptr = othercase; next = othercase + 1; for (++c; c <= d; c++) { - if (_erts_pcre_ucp_othercase(c) != next) break; + if (UCD_OTHERCASE(c) != next) break; next++; } -*odptr = next - 1; -*cptr = c; +*odptr = next - 1; /* End of othercase range */ +*cptr = c; /* Rest of input range */ +return 0; +} -return TRUE; + + +/************************************************* +* Check a character and a property * +*************************************************/ + +/* This function is called by check_auto_possessive() when a property item +is adjacent to a fixed character. + +Arguments: + c the character + ptype the property type + pdata the data for the type + negated TRUE if it's a negated property (\P or \p{^) + +Returns: TRUE if auto-possessifying is OK +*/ + +static BOOL +check_char_prop(pcre_uint32 c, unsigned int ptype, unsigned int pdata, BOOL negated) +{ +#ifdef SUPPORT_UCP +const pcre_uint32 *p; +#endif + +const ucd_record *prop = GET_UCD(c); + +switch(ptype) + { + case PT_LAMP: + return (prop->chartype == ucp_Lu || + prop->chartype == ucp_Ll || + prop->chartype == ucp_Lt) == negated; + + case PT_GC: + return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated; + + case PT_PC: + return (pdata == prop->chartype) == negated; + + case PT_SC: + return (pdata == prop->script) == negated; + + /* These are specials */ + + case PT_ALNUM: + return (PRIV(ucp_gentype)[prop->chartype] == ucp_L || + PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated; + + case PT_SPACE: /* Perl space */ + return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z || + c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR) + == negated; + + case PT_PXSPACE: /* POSIX space */ + return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z || + c == CHAR_HT || c == CHAR_NL || c == CHAR_VT || + c == CHAR_FF || c == CHAR_CR) + == negated; + + case PT_WORD: + return (PRIV(ucp_gentype)[prop->chartype] == ucp_L || + PRIV(ucp_gentype)[prop->chartype] == ucp_N || + c == CHAR_UNDERSCORE) == negated; + +#ifdef SUPPORT_UCP + case PT_CLIST: + p = PRIV(ucd_caseless_sets) + prop->caseset; + for (;;) + { + if (c < *p) return !negated; + if (c == *p++) return negated; + } + break; /* Control never reaches here */ +#endif + } + +return FALSE; } #endif /* SUPPORT_UCP */ @@ -1999,10 +3045,8 @@ whether the next thing could possibly match the repeated item. If not, it makes sense to automatically possessify the repeated item. Arguments: - op_code the repeated op code - this data for this item, depends on the opcode - utf8 TRUE in UTF-8 mode - utf8_char used for utf8 character bytes, NULL if not relevant + previous pointer to the repeated opcode + utf TRUE in UTF-8 / UTF-16 / UTF-32 mode ptr next character in pattern options options bits cd contains pointers to tables etc. @@ -2011,10 +3055,13 @@ Returns: TRUE if possessifying is wanted */ static BOOL -check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char, - const uschar *ptr, int options, compile_data *cd) +check_auto_possessive(const pcre_uchar *previous, BOOL utf, + const pcre_uchar *ptr, int options, compile_data *cd) { -int next; +pcre_uint32 c = NOTACHAR; +pcre_uint32 next; +int escape; +pcre_uchar op_code = *previous++; /* Skip whitespace and comments in extended mode */ @@ -2022,11 +3069,18 @@ if ((options & PCRE_EXTENDED) != 0) { for (;;) { - while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++; - if (*ptr == '#') + while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++; + if (*ptr == CHAR_NUMBER_SIGN) { - while (*(++ptr) != 0) + ptr++; + while (*ptr != CHAR_NULL) + { if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; } + ptr++; +#ifdef SUPPORT_UTF + if (utf) FORWARDCHAR(ptr); +#endif + } } else break; } @@ -2035,22 +3089,22 @@ if ((options & PCRE_EXTENDED) != 0) /* If the next item is one that we can handle, get its value. A non-negative value is a character, a negative value is an escape value. */ -if (*ptr == '\\') +if (*ptr == CHAR_BACKSLASH) { int temperrorcode = 0; - next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE); + escape = check_escape(&ptr, &next, &temperrorcode, cd->bracount, options, + FALSE); if (temperrorcode != 0) return FALSE; ptr++; /* Point after the escape sequence */ } - -else if ((cd->ctypes[*ptr] & ctype_meta) == 0) +else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0) { -#ifdef SUPPORT_UTF8 - if (utf8) { GETCHARINC(next, ptr); } else + escape = 0; +#ifdef SUPPORT_UTF + if (utf) { GETCHARINC(next, ptr); } else #endif next = *ptr++; } - else return FALSE; /* Skip whitespace and comments in extended mode */ @@ -2059,11 +3113,18 @@ if ((options & PCRE_EXTENDED) != 0) { for (;;) { - while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++; - if (*ptr == '#') + while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++; + if (*ptr == CHAR_NUMBER_SIGN) { - while (*(++ptr) != 0) + ptr++; + while (*ptr != CHAR_NULL) + { if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; } + ptr++; +#ifdef SUPPORT_UTF + if (utf) FORWARDCHAR(ptr); +#endif + } } else break; } @@ -2071,247 +3132,286 @@ if ((options & PCRE_EXTENDED) != 0) /* If the next thing is itself optional, we have to give up. */ -if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0) - return FALSE; - -/* Now compare the next item with the previous opcode. If the previous is a -positive single character match, "item" either contains the character or, if -"item" is greater than 127 in utf8 mode, the character's bytes are in -utf8_char. */ - +if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK || + STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0) + return FALSE; -/* Handle cases when the next item is a character. */ +/* If the previous item is a character, get its value. */ -if (next >= 0) switch(op_code) +if (op_code == OP_CHAR || op_code == OP_CHARI || + op_code == OP_NOT || op_code == OP_NOTI) { - case OP_CHAR: -#ifdef SUPPORT_UTF8 - if (utf8 && item > 127) { GETCHAR(item, utf8_char); } +#ifdef SUPPORT_UTF + GETCHARTEST(c, previous); +#else + c = *previous; #endif - return item != next; + } - /* For CHARNC (caseless character) we must check the other case. If we have - Unicode property support, we can use it to test the other case of - high-valued characters. */ +/* Now compare the next item with the previous opcode. First, handle cases when +the next item is a character. */ - case OP_CHARNC: -#ifdef SUPPORT_UTF8 - if (utf8 && item > 127) { GETCHAR(item, utf8_char); } +if (escape == 0) + { + /* For a caseless UTF match, the next character may have more than one other + case, which maps to the special PT_CLIST property. Check this first. */ + +#ifdef SUPPORT_UCP + if (utf && c != NOTACHAR && (options & PCRE_CASELESS) != 0) + { + unsigned int ocs = UCD_CASESET(next); + if (ocs > 0) return check_char_prop(c, PT_CLIST, ocs, op_code >= OP_NOT); + } #endif - if (item == next) return FALSE; -#ifdef SUPPORT_UTF8 - if (utf8) + + switch(op_code) { - unsigned int othercase; - if (next < 128) othercase = cd->fcc[next]; else + case OP_CHAR: + return c != next; + + /* For CHARI (caseless character) we must check the other case. If we have + Unicode property support, we can use it to test the other case of + high-valued characters. We know that next can have only one other case, + because multi-other-case characters are dealt with above. */ + + case OP_CHARI: + if (c == next) return FALSE; +#ifdef SUPPORT_UTF + if (utf) + { + pcre_uint32 othercase; + if (next < 128) othercase = cd->fcc[next]; else #ifdef SUPPORT_UCP - othercase = _erts_pcre_ucp_othercase((unsigned int)next); + othercase = UCD_OTHERCASE(next); #else - othercase = NOTACHAR; + othercase = NOTACHAR; #endif - return (unsigned int)item != othercase; - } - else -#endif /* SUPPORT_UTF8 */ - return (item != cd->fcc[next]); /* Non-UTF-8 mode */ + return c != othercase; + } + else +#endif /* SUPPORT_UTF */ + return (c != TABLE_GET(next, cd->fcc, next)); /* Not UTF */ - /* For OP_NOT, "item" must be a single-byte character. */ + case OP_NOT: + return c == next; - case OP_NOT: - if (next < 0) return FALSE; /* Not a character */ - if (item == next) return TRUE; - if ((options & PCRE_CASELESS) == 0) return FALSE; -#ifdef SUPPORT_UTF8 - if (utf8) - { - unsigned int othercase; - if (next < 128) othercase = cd->fcc[next]; else + case OP_NOTI: + if (c == next) return TRUE; +#ifdef SUPPORT_UTF + if (utf) + { + pcre_uint32 othercase; + if (next < 128) othercase = cd->fcc[next]; else #ifdef SUPPORT_UCP - othercase = _erts_pcre_ucp_othercase(next); + othercase = UCD_OTHERCASE(next); #else - othercase = NOTACHAR; + othercase = NOTACHAR; #endif - return (unsigned int)item == othercase; - } - else -#endif /* SUPPORT_UTF8 */ - return (item == cd->fcc[next]); /* Non-UTF-8 mode */ + return c == othercase; + } + else +#endif /* SUPPORT_UTF */ + return (c == TABLE_GET(next, cd->fcc, next)); /* Not UTF */ - case OP_DIGIT: - return next > 127 || (cd->ctypes[next] & ctype_digit) == 0; + /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set. + When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */ - case OP_NOT_DIGIT: - return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0; + case OP_DIGIT: + return next > 255 || (cd->ctypes[next] & ctype_digit) == 0; - case OP_WHITESPACE: - return next > 127 || (cd->ctypes[next] & ctype_space) == 0; + case OP_NOT_DIGIT: + return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0; - case OP_NOT_WHITESPACE: - return next <= 127 && (cd->ctypes[next] & ctype_space) != 0; + case OP_WHITESPACE: + return next > 255 || (cd->ctypes[next] & ctype_space) == 0; - case OP_WORDCHAR: - return next > 127 || (cd->ctypes[next] & ctype_word) == 0; + case OP_NOT_WHITESPACE: + return next <= 255 && (cd->ctypes[next] & ctype_space) != 0; - case OP_NOT_WORDCHAR: - return next <= 127 && (cd->ctypes[next] & ctype_word) != 0; + case OP_WORDCHAR: + return next > 255 || (cd->ctypes[next] & ctype_word) == 0; - case OP_HSPACE: - case OP_NOT_HSPACE: - switch(next) - { - case 0x09: - case 0x20: - case 0xa0: - case 0x1680: - case 0x180e: - case 0x2000: - case 0x2001: - case 0x2002: - case 0x2003: - case 0x2004: - case 0x2005: - case 0x2006: - case 0x2007: - case 0x2008: - case 0x2009: - case 0x200A: - case 0x202f: - case 0x205f: - case 0x3000: - return op_code != OP_HSPACE; - default: - return op_code == OP_HSPACE; - } + case OP_NOT_WORDCHAR: + return next <= 255 && (cd->ctypes[next] & ctype_word) != 0; + + case OP_HSPACE: + case OP_NOT_HSPACE: + switch(next) + { + HSPACE_CASES: + return op_code == OP_NOT_HSPACE; + + default: + return op_code != OP_NOT_HSPACE; + } + + case OP_ANYNL: + case OP_VSPACE: + case OP_NOT_VSPACE: + switch(next) + { + VSPACE_CASES: + return op_code == OP_NOT_VSPACE; + + default: + return op_code != OP_NOT_VSPACE; + } + +#ifdef SUPPORT_UCP + case OP_PROP: + return check_char_prop(next, previous[0], previous[1], FALSE); + + case OP_NOTPROP: + return check_char_prop(next, previous[0], previous[1], TRUE); +#endif - case OP_VSPACE: - case OP_NOT_VSPACE: - switch(next) - { - case 0x0a: - case 0x0b: - case 0x0c: - case 0x0d: - case 0x85: - case 0x2028: - case 0x2029: - return op_code != OP_VSPACE; default: - return op_code == OP_VSPACE; + return FALSE; } - - default: - return FALSE; } - -/* Handle the case when the next item is \d, \s, etc. */ +/* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP +is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are +generated only when PCRE_UCP is *not* set, that is, when only ASCII +characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are +replaced by OP_PROP codes when PCRE_UCP is set. */ switch(op_code) { case OP_CHAR: - case OP_CHARNC: -#ifdef SUPPORT_UTF8 - if (utf8 && item > 127) { GETCHAR(item, utf8_char); } -#endif - switch(-next) + case OP_CHARI: + switch(escape) { case ESC_d: - return item > 127 || (cd->ctypes[item] & ctype_digit) == 0; + return c > 255 || (cd->ctypes[c] & ctype_digit) == 0; case ESC_D: - return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0; + return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0; case ESC_s: - return item > 127 || (cd->ctypes[item] & ctype_space) == 0; + return c > 255 || (cd->ctypes[c] & ctype_space) == 0; case ESC_S: - return item <= 127 && (cd->ctypes[item] & ctype_space) != 0; + return c <= 255 && (cd->ctypes[c] & ctype_space) != 0; case ESC_w: - return item > 127 || (cd->ctypes[item] & ctype_word) == 0; + return c > 255 || (cd->ctypes[c] & ctype_word) == 0; case ESC_W: - return item <= 127 && (cd->ctypes[item] & ctype_word) != 0; + return c <= 255 && (cd->ctypes[c] & ctype_word) != 0; case ESC_h: case ESC_H: - switch(item) + switch(c) { - case 0x09: - case 0x20: - case 0xa0: - case 0x1680: - case 0x180e: - case 0x2000: - case 0x2001: - case 0x2002: - case 0x2003: - case 0x2004: - case 0x2005: - case 0x2006: - case 0x2007: - case 0x2008: - case 0x2009: - case 0x200A: - case 0x202f: - case 0x205f: - case 0x3000: - return -next != ESC_h; + HSPACE_CASES: + return escape != ESC_h; + default: - return -next == ESC_h; + return escape == ESC_h; } case ESC_v: case ESC_V: - switch(item) + switch(c) { - case 0x0a: - case 0x0b: - case 0x0c: - case 0x0d: - case 0x85: - case 0x2028: - case 0x2029: - return -next != ESC_v; + VSPACE_CASES: + return escape != ESC_v; + default: - return -next == ESC_v; + return escape == ESC_v; + } + + /* When PCRE_UCP is set, these values get generated for \d etc. Find + their substitutions and process them. The result will always be either + ESC_p or ESC_P. Then fall through to process those values. */ + +#ifdef SUPPORT_UCP + case ESC_du: + case ESC_DU: + case ESC_wu: + case ESC_WU: + case ESC_su: + case ESC_SU: + { + int temperrorcode = 0; + ptr = substitutes[escape - ESC_DU]; + escape = check_escape(&ptr, &next, &temperrorcode, 0, options, FALSE); + if (temperrorcode != 0) return FALSE; + ptr++; /* For compatibility */ } + /* Fall through */ + + case ESC_p: + case ESC_P: + { + unsigned int ptype = 0, pdata = 0; + int errorcodeptr; + BOOL negated; + + ptr--; /* Make ptr point at the p or P */ + if (!get_ucp(&ptr, &negated, &ptype, &pdata, &errorcodeptr)) + return FALSE; + ptr++; /* Point past the final curly ket */ + + /* If the property item is optional, we have to give up. (When generated + from \d etc by PCRE_UCP, this test will have been applied much earlier, + to the original \d etc. At this point, ptr will point to a zero byte. */ + + if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK || + STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0) + return FALSE; + + /* Do the property check. */ + + return check_char_prop(c, ptype, pdata, (escape == ESC_P) != negated); + } +#endif default: return FALSE; } + /* In principle, support for Unicode properties should be integrated here as + well. It means re-organizing the above code so as to get hold of the property + values before switching on the op-code. However, I wonder how many patterns + combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set, + these op-codes are never generated.) */ + case OP_DIGIT: - return next == -ESC_D || next == -ESC_s || next == -ESC_W || - next == -ESC_h || next == -ESC_v; + return escape == ESC_D || escape == ESC_s || escape == ESC_W || + escape == ESC_h || escape == ESC_v || escape == ESC_R; case OP_NOT_DIGIT: - return next == -ESC_d; + return escape == ESC_d; case OP_WHITESPACE: - return next == -ESC_S || next == -ESC_d || next == -ESC_w; + return escape == ESC_S || escape == ESC_d || escape == ESC_w; case OP_NOT_WHITESPACE: - return next == -ESC_s || next == -ESC_h || next == -ESC_v; + return escape == ESC_s || escape == ESC_h || escape == ESC_v || escape == ESC_R; case OP_HSPACE: - return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w; + return escape == ESC_S || escape == ESC_H || escape == ESC_d || + escape == ESC_w || escape == ESC_v || escape == ESC_R; case OP_NOT_HSPACE: - return next == -ESC_h; + return escape == ESC_h; /* Can't have \S in here because VT matches \S (Perl anomaly) */ + case OP_ANYNL: case OP_VSPACE: - return next == -ESC_V || next == -ESC_d || next == -ESC_w; + return escape == ESC_V || escape == ESC_d || escape == ESC_w; case OP_NOT_VSPACE: - return next == -ESC_v; + return escape == ESC_v || escape == ESC_R; case OP_WORDCHAR: - return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v; + return escape == ESC_W || escape == ESC_s || escape == ESC_h || + escape == ESC_v || escape == ESC_R; case OP_NOT_WORDCHAR: - return next == -ESC_w || next == -ESC_d; + return escape == ESC_w || escape == ESC_d; default: return FALSE; @@ -2323,6 +3423,244 @@ switch(op_code) /************************************************* +* Add a character or range to a class * +*************************************************/ + +/* This function packages up the logic of adding a character or range of +characters to a class. The character values in the arguments will be within the +valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is +mutually recursive with the function immediately below. + +Arguments: + classbits the bit map for characters < 256 + uchardptr points to the pointer for extra data + options the options word + cd contains pointers to tables etc. + start start of range character + end end of range character + +Returns: the number of < 256 characters added + the pointer to extra data is updated +*/ + +static int +add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options, + compile_data *cd, pcre_uint32 start, pcre_uint32 end) +{ +pcre_uint32 c; +int n8 = 0; + +/* If caseless matching is required, scan the range and process alternate +cases. In Unicode, there are 8-bit characters that have alternate cases that +are greater than 255 and vice-versa. Sometimes we can just extend the original +range. */ + +if ((options & PCRE_CASELESS) != 0) + { +#ifdef SUPPORT_UCP + if ((options & PCRE_UTF8) != 0) + { + int rc; + pcre_uint32 oc, od; + + options &= ~PCRE_CASELESS; /* Remove for recursive calls */ + c = start; + + while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0) + { + /* Handle a single character that has more than one other case. */ + + if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd, + PRIV(ucd_caseless_sets) + rc, oc); + + /* Do nothing if the other case range is within the original range. */ + + else if (oc >= start && od <= end) continue; + + /* Extend the original range if there is overlap, noting that if oc < c, we + can't have od > end because a subrange is always shorter than the basic + range. Otherwise, use a recursive call to add the additional range. */ + + else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */ + else if (od > end && oc <= end + 1) end = od; /* Extend upwards */ + else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od); + } + } + else +#endif /* SUPPORT_UCP */ + + /* Not UTF-mode, or no UCP */ + + for (c = start; c <= end && c < 256; c++) + { + SETBIT(classbits, cd->fcc[c]); + n8++; + } + } + +/* Now handle the original range. Adjust the final value according to the bit +length - this means that the same lists of (e.g.) horizontal spaces can be used +in all cases. */ + +#if defined COMPILE_PCRE8 +#ifdef SUPPORT_UTF + if ((options & PCRE_UTF8) == 0) +#endif + if (end > 0xff) end = 0xff; + +#elif defined COMPILE_PCRE16 +#ifdef SUPPORT_UTF + if ((options & PCRE_UTF16) == 0) +#endif + if (end > 0xffff) end = 0xffff; + +#endif /* COMPILE_PCRE[8|16] */ + +/* If all characters are less than 256, use the bit map. Otherwise use extra +data. */ + +if (end < 0x100) + { + for (c = start; c <= end; c++) + { + n8++; + SETBIT(classbits, c); + } + } + +else + { + pcre_uchar *uchardata = *uchardptr; + +#ifdef SUPPORT_UTF + if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */ + { + if (start < end) + { + *uchardata++ = XCL_RANGE; + uchardata += PRIV(ord2utf)(start, uchardata); + uchardata += PRIV(ord2utf)(end, uchardata); + } + else if (start == end) + { + *uchardata++ = XCL_SINGLE; + uchardata += PRIV(ord2utf)(start, uchardata); + } + } + else +#endif /* SUPPORT_UTF */ + + /* Without UTF support, character values are constrained by the bit length, + and can only be > 256 for 16-bit and 32-bit libraries. */ + +#ifdef COMPILE_PCRE8 + {} +#else + if (start < end) + { + *uchardata++ = XCL_RANGE; + *uchardata++ = start; + *uchardata++ = end; + } + else if (start == end) + { + *uchardata++ = XCL_SINGLE; + *uchardata++ = start; + } +#endif + + *uchardptr = uchardata; /* Updata extra data pointer */ + } + +return n8; /* Number of 8-bit characters */ +} + + + + +/************************************************* +* Add a list of characters to a class * +*************************************************/ + +/* This function is used for adding a list of case-equivalent characters to a +class, and also for adding a list of horizontal or vertical whitespace. If the +list is in order (which it should be), ranges of characters are detected and +handled appropriately. This function is mutually recursive with the function +above. + +Arguments: + classbits the bit map for characters < 256 + uchardptr points to the pointer for extra data + options the options word + cd contains pointers to tables etc. + p points to row of 32-bit values, terminated by NOTACHAR + except character to omit; this is used when adding lists of + case-equivalent characters to avoid including the one we + already know about + +Returns: the number of < 256 characters added + the pointer to extra data is updated +*/ + +static int +add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options, + compile_data *cd, const pcre_uint32 *p, unsigned int except) +{ +int n8 = 0; +while (p[0] < NOTACHAR) + { + int n = 0; + if (p[0] != except) + { + while(p[n+1] == p[0] + n + 1) n++; + n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]); + } + p += n + 1; + } +return n8; +} + + + +/************************************************* +* Add characters not in a list to a class * +*************************************************/ + +/* This function is used for adding the complement of a list of horizontal or +vertical whitespace to a class. The list must be in order. + +Arguments: + classbits the bit map for characters < 256 + uchardptr points to the pointer for extra data + options the options word + cd contains pointers to tables etc. + p points to row of 32-bit values, terminated by NOTACHAR + +Returns: the number of < 256 characters added + the pointer to extra data is updated +*/ + +static int +add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, + int options, compile_data *cd, const pcre_uint32 *p) +{ +BOOL utf = (options & PCRE_UTF8) != 0; +int n8 = 0; +if (p[0] > 0) + n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1); +while (p[0] < NOTACHAR) + { + while (p[1] == p[0] + 1) p++; + n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1, + (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1); + p++; + } +return n8; +} + + + +/************************************************* * Compile one branch * *************************************************/ @@ -2337,9 +3675,12 @@ Arguments: codeptr points to the pointer to the current code point ptrptr points to the current pattern pointer errorcodeptr points to error code variable - firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE) - reqbyteptr set to the last literal character required, else < 0 + firstcharptr place to put the first required character + firstcharflagsptr place to put the first character flags, or a negative number + reqcharptr place to put the last required character + reqcharflagsptr place to put the last required character flags, or a negative number bcptr points to current branch chain + cond_depth conditional nesting depth cd contains pointers to tables etc. lengthptr NULL during the real compile phase points to length accumulator during pre-compile phase @@ -2349,46 +3690,67 @@ Returns: TRUE on success */ static BOOL -compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr, - int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, +compile_branch(int *optionsptr, pcre_uchar **codeptr, + const pcre_uchar **ptrptr, int *errorcodeptr, + pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr, + pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr, + branch_chain *bcptr, int cond_depth, compile_data *cd, int *lengthptr) { int repeat_type, op_type; int repeat_min = 0, repeat_max = 0; /* To please picky compilers */ int bravalue = 0; int greedy_default, greedy_non_default; -int firstbyte, reqbyte; -int zeroreqbyte, zerofirstbyte; -int req_caseopt, reqvary, tempreqvary; -int options = *optionsptr; +pcre_uint32 firstchar, reqchar; +pcre_int32 firstcharflags, reqcharflags; +pcre_uint32 zeroreqchar, zerofirstchar; +pcre_int32 zeroreqcharflags, zerofirstcharflags; +pcre_int32 req_caseopt, reqvary, tempreqvary; +int options = *optionsptr; /* May change dynamically */ int after_manual_callout = 0; int length_prevgroup = 0; -register int c; -register uschar *code = *codeptr; -uschar *last_code = code; -uschar *orig_code = code; -uschar *tempcode; +register pcre_uint32 c; +int escape; +register pcre_uchar *code = *codeptr; +pcre_uchar *last_code = code; +pcre_uchar *orig_code = code; +pcre_uchar *tempcode; BOOL inescq = FALSE; -BOOL groupsetfirstbyte = FALSE; -const uschar *ptr = *ptrptr; -const uschar *tempptr; -uschar *previous = NULL; -uschar *previous_callout = NULL; -uschar *save_hwm = NULL; -uschar classbits[32]; - -#ifdef SUPPORT_UTF8 -BOOL class_utf8; -BOOL utf8 = (options & PCRE_UTF8) != 0; -uschar *class_utf8data; -uschar *class_utf8data_base; -uschar utf8_char[6]; +BOOL groupsetfirstchar = FALSE; +const pcre_uchar *ptr = *ptrptr; +const pcre_uchar *tempptr; +const pcre_uchar *nestptr = NULL; +pcre_uchar *previous = NULL; +pcre_uchar *previous_callout = NULL; +pcre_uchar *save_hwm = NULL; +pcre_uint8 classbits[32]; + +/* We can fish out the UTF-8 setting once and for all into a BOOL, but we +must not do this for other options (e.g. PCRE_EXTENDED) because they may change +dynamically as we process the pattern. */ + +#ifdef SUPPORT_UTF +/* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */ +BOOL utf = (options & PCRE_UTF8) != 0; +#ifndef COMPILE_PCRE32 +pcre_uchar utf_chars[6]; +#endif #else -BOOL utf8 = FALSE; -uschar *utf8_char = NULL; +BOOL utf = FALSE; #endif -#ifdef DEBUG +/* Helper variables for OP_XCLASS opcode (for characters > 255). We define +class_uchardata always so that it can be passed to add_to_class() always, +though it will not be used in non-UTF 8-bit cases. This avoids having to supply +alternative calls for the different cases. */ + +pcre_uchar *class_uchardata; +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 +BOOL xclass; +pcre_uchar *class_uchardata_base; +#endif + +#ifdef PCRE_DEBUG if (lengthptr != NULL) DPRINTF((">> start branch\n")); #endif @@ -2399,22 +3761,24 @@ greedy_non_default = greedy_default ^ 1; /* Initialize no first byte, no required byte. REQ_UNSET means "no char matching encountered yet". It gets changed to REQ_NONE if we hit something that -matches a non-fixed char first char; reqbyte just remains unset if we never +matches a non-fixed char first char; reqchar just remains unset if we never find one. When we hit a repeat whose minimum is zero, we may have to adjust these values to take the zero repeat into account. This is implemented by setting them to -zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual +zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual item types that can be repeated set these backoff variables appropriately. */ -firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET; +firstchar = reqchar = zerofirstchar = zeroreqchar = 0; +firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET; -/* The variable req_caseopt contains either the REQ_CASELESS value or zero, -according to the current setting of the caseless flag. REQ_CASELESS is a bit -value > 255. It is added into the firstbyte or reqbyte variables to record the -case status of the value. This is used only for ASCII characters. */ +/* The variable req_caseopt contains either the REQ_CASELESS value +or zero, according to the current setting of the caseless flag. The +REQ_CASELESS leaves the lower 28 bit empty. It is added into the +firstchar or reqchar variables to record the case status of the +value. This is used only for ASCII characters. */ -req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0; +req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0; /* Switch on next character until the end of the branch */ @@ -2426,31 +3790,44 @@ for (;; ptr++) BOOL is_quantifier; BOOL is_recurse; BOOL reset_bracount; - int class_charcount; - int class_lastchar; + int class_has_8bitchar; + int class_one_char; int newoptions; int recno; int refsign; int skipbytes; - int subreqbyte; - int subfirstbyte; + pcre_uint32 subreqchar, subfirstchar; + pcre_int32 subreqcharflags, subfirstcharflags; int terminator; - int mclength; - uschar mcbuffer[8]; + unsigned int mclength; + unsigned int tempbracount; + pcre_uint32 ec; + pcre_uchar mcbuffer[8]; - /* Get next byte in the pattern */ + /* Get next character in the pattern */ c = *ptr; + /* If we are at the end of a nested substitution, revert to the outer level + string. Nesting only happens one level deep. */ + + if (c == CHAR_NULL && nestptr != NULL) + { + ptr = nestptr; + nestptr = NULL; + c = *ptr; + } + /* If we are in the pre-compile phase, accumulate the length used for the previous cycle of this loop. */ if (lengthptr != NULL) { -#ifdef DEBUG +#ifdef PCRE_DEBUG if (code > cd->hwm) cd->hwm = code; /* High water info */ #endif - if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */ + if (code > cd->start_workspace + cd->workspace_size - + WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */ { *errorcodeptr = ERR52; goto FAILED; @@ -2472,8 +3849,9 @@ for (;; ptr++) goto FAILED; } - *lengthptr += code - last_code; - DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c)); + *lengthptr += (int)(code - last_code); + DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr, + (int)(code - last_code), c, c)); /* If "previous" is set and it is not at the start of the work space, move it back to there, in order to avoid filling up the work space. Otherwise, @@ -2483,7 +3861,7 @@ for (;; ptr++) { if (previous > orig_code) { - memmove(orig_code, previous, code - previous); + memmove(orig_code, previous, IN_UCHARS(code - previous)); code -= previous - orig_code; previous = orig_code; } @@ -2499,7 +3877,8 @@ for (;; ptr++) /* In the real compile phase, just check the workspace used by the forward reference list. */ - else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK) + else if (cd->hwm > cd->start_workspace + cd->workspace_size - + WORK_SIZE_SAFETY_MARGIN) { *errorcodeptr = ERR52; goto FAILED; @@ -2507,9 +3886,9 @@ for (;; ptr++) /* If in \Q...\E, check for the end; if not, we have a literal */ - if (inescq && c != 0) + if (inescq && c != CHAR_NULL) { - if (c == '\\' && ptr[1] == 'E') + if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) { inescq = FALSE; ptr++; @@ -2535,8 +3914,9 @@ for (;; ptr++) /* Fill in length of a previous callout, except when the next thing is a quantifier. */ - is_quantifier = c == '*' || c == '+' || c == '?' || - (c == '{' && is_counted_repeat(ptr+1)); + is_quantifier = + c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK || + (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1)); if (!is_quantifier && previous_callout != NULL && after_manual_callout-- <= 0) @@ -2546,18 +3926,23 @@ for (;; ptr++) previous_callout = NULL; } - /* In extended mode, skip white space and comments */ + /* In extended mode, skip white space and comments. */ if ((options & PCRE_EXTENDED) != 0) { - if ((cd->ctypes[c] & ctype_space) != 0) continue; - if (c == '#') + if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue; + if (c == CHAR_NUMBER_SIGN) { - while (*(++ptr) != 0) + ptr++; + while (*ptr != CHAR_NULL) { if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; } + ptr++; +#ifdef SUPPORT_UTF + if (utf) FORWARDCHAR(ptr); +#endif } - if (*ptr != 0) continue; + if (*ptr != CHAR_NULL) continue; /* Else fall through to handle end of string */ c = 0; @@ -2576,10 +3961,12 @@ for (;; ptr++) { /* ===================================================================*/ case 0: /* The branch terminates at string end */ - case '|': /* or | or ) */ - case ')': - *firstbyteptr = firstbyte; - *reqbyteptr = reqbyte; + case CHAR_VERTICAL_LINE: /* or | or ) */ + case CHAR_RIGHT_PARENTHESIS: + *firstcharptr = firstchar; + *firstcharflagsptr = firstcharflags; + *reqcharptr = reqchar; + *reqcharflagsptr = reqcharflags; *codeptr = code; *ptrptr = ptr; if (lengthptr != NULL) @@ -2589,7 +3976,7 @@ for (;; ptr++) *errorcodeptr = ERR20; goto FAILED; } - *lengthptr += code - last_code; /* To include callout length */ + *lengthptr += (int)(code - last_code); /* To include callout length */ DPRINTF((">> end branch\n")); } return TRUE; @@ -2599,29 +3986,32 @@ for (;; ptr++) /* Handle single-character metacharacters. In multiline mode, ^ disables the setting of any following char as a first character. */ - case '^': + case CHAR_CIRCUMFLEX_ACCENT: + previous = NULL; if ((options & PCRE_MULTILINE) != 0) { - if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; + *code++ = OP_CIRCM; } - previous = NULL; - *code++ = OP_CIRC; + else *code++ = OP_CIRC; break; - case '$': + case CHAR_DOLLAR_SIGN: previous = NULL; - *code++ = OP_DOLL; + *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL; break; /* There can never be a first char if '.' is first, whatever happens about - repeats. The value of reqbyte doesn't change either. */ - - case '.': - if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; - zerofirstbyte = firstbyte; - zeroreqbyte = reqbyte; + repeats. The value of reqchar doesn't change either. */ + + case CHAR_DOT: + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; + zerofirstchar = firstchar; + zerofirstcharflags = firstcharflags; + zeroreqchar = reqchar; + zeroreqcharflags = reqcharflags; previous = code; - *code++ = OP_ANY; + *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY; break; @@ -2636,18 +4026,29 @@ for (;; ptr++) opcode is compiled. It may optionally have a bit map for characters < 256, but those above are are explicitly listed afterwards. A flag byte tells whether the bitmap is present, and whether this is a negated class or not. - */ - case '[': + In JavaScript compatibility mode, an isolated ']' causes an error. In + default (Perl) mode, it is treated as a data character. */ + + case CHAR_RIGHT_SQUARE_BRACKET: + if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0) + { + *errorcodeptr = ERR64; + goto FAILED; + } + goto NORMAL_CHAR; + + case CHAR_LEFT_SQUARE_BRACKET: previous = code; /* PCRE supports POSIX class stuff inside a class. Perl gives an error if they are encountered at the top level, so we'll do that too. */ - if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') && + if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT || + ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr)) { - *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31; + *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31; goto FAILED; } @@ -2659,75 +4060,97 @@ for (;; ptr++) for (;;) { c = *(++ptr); - if (c == '\\') + if (c == CHAR_BACKSLASH) { - if (ptr[1] == 'E') ptr++; - else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3; - else break; + if (ptr[1] == CHAR_E) + ptr++; + else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0) + ptr += 3; + else + break; } - else if (!negate_class && c == '^') + else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT) negate_class = TRUE; else break; } + /* Empty classes are allowed in JavaScript compatibility mode. Otherwise, + an initial ']' is taken as a data character -- the code below handles + that. In JS mode, [] must always fail, so generate OP_FAIL, whereas + [^] must match any character, so generate OP_ALLANY. */ + + if (c == CHAR_RIGHT_SQUARE_BRACKET && + (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0) + { + *code++ = negate_class? OP_ALLANY : OP_FAIL; + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; + zerofirstchar = firstchar; + zerofirstcharflags = firstcharflags; + break; + } + /* If a class contains a negative special such as \S, we need to flip the negation flag at the end, so that support for characters > 255 works correctly (they are all included in the class). */ should_flip_negation = FALSE; - /* Keep a count of chars with values < 256 so that we can optimize the case - of just a single character (as long as it's < 256). However, For higher - valued UTF-8 characters, we don't yet do any optimization. */ + /* For optimization purposes, we track some properties of the class: + class_has_8bitchar will be non-zero if the class contains at least one < + 256 character; class_one_char will be 1 if the class contains just one + character. */ - class_charcount = 0; - class_lastchar = -1; + class_has_8bitchar = 0; + class_one_char = 0; /* Initialize the 32-char bit map to all zeros. We build the map in a - temporary bit of memory, in case the class contains only 1 character (less - than 256), because in that case the compiled code doesn't use the bit map. - */ + temporary bit of memory, in case the class contains fewer than two + 8-bit characters because in that case the compiled code doesn't use the bit + map. */ - memset(classbits, 0, 32 * sizeof(uschar)); + memset(classbits, 0, 32 * sizeof(pcre_uint8)); -#ifdef SUPPORT_UTF8 - class_utf8 = FALSE; /* No chars >= 256 */ - class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */ - class_utf8data_base = class_utf8data; /* For resetting in pass 1 */ +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 + xclass = FALSE; + class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */ + class_uchardata_base = class_uchardata; /* Save the start */ #endif /* Process characters until ] is reached. By writing this as a "do" it means that an initial ] is taken as a data character. At the start of the loop, c contains the first byte of the character. */ - if (c != 0) do + if (c != CHAR_NULL) do { - const uschar *oldptr; + const pcre_uchar *oldptr; -#ifdef SUPPORT_UTF8 - if (utf8 && c > 127) +#ifdef SUPPORT_UTF + if (utf && HAS_EXTRALEN(c)) { /* Braces are required because the */ GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */ } +#endif - /* In the pre-compile phase, accumulate the length of any UTF-8 extra +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 + /* In the pre-compile phase, accumulate the length of any extra data and reset the pointer. This is so that very large classes that - contain a zillion UTF-8 characters no longer overwrite the work space - (which is on the stack). */ + contain a zillion > 255 characters no longer overwrite the work space + (which is on the stack). We have to remember that there was XCLASS data, + however. */ - if (lengthptr != NULL) + if (lengthptr != NULL && class_uchardata > class_uchardata_base) { - *lengthptr += class_utf8data - class_utf8data_base; - class_utf8data = class_utf8data_base; + xclass = TRUE; + *lengthptr += class_uchardata - class_uchardata_base; + class_uchardata = class_uchardata_base; } - #endif /* Inside \Q...\E everything is literal except \E */ if (inescq) { - if (c == '\\' && ptr[1] == 'E') /* If we are at \E */ + if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */ { inescq = FALSE; /* Reset literal state */ ptr++; /* Skip the 'E' */ @@ -2742,30 +4165,30 @@ for (;; ptr++) [.ch.] and [=ch=] ("collating elements") and fault them, as Perl 5.6 and 5.8 do. */ - if (c == '[' && - (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') && - check_posix_syntax(ptr, &tempptr)) + if (c == CHAR_LEFT_SQUARE_BRACKET && + (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT || + ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr)) { BOOL local_negate = FALSE; int posix_class, taboffset, tabopt; - register const uschar *cbits = cd->cbits; - uschar pbits[32]; + register const pcre_uint8 *cbits = cd->cbits; + pcre_uint8 pbits[32]; - if (ptr[1] != ':') + if (ptr[1] != CHAR_COLON) { *errorcodeptr = ERR31; goto FAILED; } ptr += 2; - if (*ptr == '^') + if (*ptr == CHAR_CIRCUMFLEX_ACCENT) { local_negate = TRUE; should_flip_negation = TRUE; /* Note negative special */ ptr++; } - posix_class = check_posix_name(ptr, tempptr - ptr); + posix_class = check_posix_name(ptr, (int)(tempptr - ptr)); if (posix_class < 0) { *errorcodeptr = ERR30; @@ -2779,17 +4202,32 @@ for (;; ptr++) if ((options & PCRE_CASELESS) != 0 && posix_class <= 2) posix_class = 0; - /* We build the bit map for the POSIX class in a chunk of local store - because we may be adding and subtracting from it, and we don't want to - subtract bits that may be in the main map already. At the end we or the - result into the bit map that is being built. */ + /* When PCRE_UCP is set, some of the POSIX classes are converted to + different escape sequences that use Unicode properties. */ + +#ifdef SUPPORT_UCP + if ((options & PCRE_UCP) != 0) + { + int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0); + if (posix_substitutes[pc] != NULL) + { + nestptr = tempptr + 1; + ptr = posix_substitutes[pc] - 1; + continue; + } + } +#endif + /* In the non-UCP case, we build the bit map for the POSIX class in a + chunk of local store because we may be adding and subtracting from it, + and we don't want to subtract bits that may be in the main map already. + At the end we or the result into the bit map that is being built. */ posix_class *= 3; /* Copy in the first table (always present) */ memcpy(pbits, cbits + posix_class_maps[posix_class], - 32 * sizeof(uschar)); + 32 * sizeof(pcre_uint8)); /* If there is a second table, add or remove it as required. */ @@ -2804,7 +4242,7 @@ for (;; ptr++) for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset]; } - /* Not see if we need to remove any special characters. An option + /* Now see if we need to remove any special characters. An option value of 1 removes vertical space and 2 removes underscore. */ if (tabopt < 0) tabopt = -tabopt; @@ -2820,45 +4258,67 @@ for (;; ptr++) for (c = 0; c < 32; c++) classbits[c] |= pbits[c]; ptr = tempptr + 1; - class_charcount = 10; /* Set > 1; assumes more than 1 per class */ + /* Every class contains at least one < 256 character. */ + class_has_8bitchar = 1; + /* Every class contains at least two characters. */ + class_one_char = 2; continue; /* End of POSIX syntax handling */ } /* Backslash may introduce a single character, or it may introduce one of the specials, which just set a flag. The sequence \b is a special - case. Inside a class (and only there) it is treated as backspace. - Elsewhere it marks a word boundary. Other escapes have preset maps ready - to 'or' into the one we are building. We assume they have more than one - character in them, so set class_charcount bigger than one. */ - - if (c == '\\') + case. Inside a class (and only there) it is treated as backspace. We + assume that other escapes have more than one character in them, so + speculatively set both class_has_8bitchar and class_one_char bigger + than one. Unrecognized escapes fall through and are either treated + as literal characters (by default), or are faulted if + PCRE_EXTRA is set. */ + + if (c == CHAR_BACKSLASH) { - c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE); + escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, + TRUE); if (*errorcodeptr != 0) goto FAILED; - - if (-c == ESC_b) c = '\b'; /* \b is backspace in a class */ - else if (-c == ESC_X) c = 'X'; /* \X is literal X in a class */ - else if (-c == ESC_R) c = 'R'; /* \R is literal R in a class */ - else if (-c == ESC_Q) /* Handle start of quoted string */ + if (escape == 0) c = ec; + else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */ + else if (escape == ESC_N) /* \N is not supported in a class */ + { + *errorcodeptr = ERR71; + goto FAILED; + } + else if (escape == ESC_Q) /* Handle start of quoted string */ { - if (ptr[1] == '\\' && ptr[2] == 'E') + if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E) { ptr += 2; /* avoid empty string */ } else inescq = TRUE; continue; } - else if (-c == ESC_E) continue; /* Ignore orphan \E */ + else if (escape == ESC_E) continue; /* Ignore orphan \E */ - if (c < 0) + else { - register const uschar *cbits = cd->cbits; - class_charcount += 2; /* Greater than 1 is what matters */ + register const pcre_uint8 *cbits = cd->cbits; + /* Every class contains at least two < 256 characters. */ + class_has_8bitchar++; + /* Every class contains at least two characters. */ + class_one_char += 2; - /* Save time by not doing this in the pre-compile phase. */ - - if (lengthptr == NULL) switch (-c) + switch (escape) { +#ifdef SUPPORT_UCP + case ESC_du: /* These are the values given for \d etc */ + case ESC_DU: /* when PCRE_UCP is set. We replace the */ + case ESC_wu: /* escape sequence with an appropriate \p */ + case ESC_WU: /* or \P to test Unicode properties instead */ + case ESC_su: /* of the default ASCII testing. */ + case ESC_SU: + nestptr = ptr; + ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */ + class_has_8bitchar--; /* Undo! */ + continue; +#endif case ESC_d: for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit]; continue; @@ -2877,9 +4337,15 @@ for (;; ptr++) for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word]; continue; + /* Perl 5.004 onwards omits VT from \s, but we must preserve it + if it was previously set by something earlier in the character + class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and + EBCDIC, so we lazily just adjust the appropriate bit. */ + case ESC_s: - for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space]; - classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */ + classbits[0] |= cbits[cbit_space]; + classbits[1] |= cbits[cbit_space+1] & ~0x08; + for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space]; continue; case ESC_S: @@ -2888,224 +4354,118 @@ for (;; ptr++) classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */ continue; - default: /* Not recognized; fall through */ - break; /* Need "default" setting to stop compiler warning. */ - } - - /* In the pre-compile phase, just do the recognition. */ + /* The rest apply in both UCP and non-UCP cases. */ - else if (c == -ESC_d || c == -ESC_D || c == -ESC_w || - c == -ESC_W || c == -ESC_s || c == -ESC_S) continue; - - /* We need to deal with \H, \h, \V, and \v in both phases because - they use extra memory. */ - - if (-c == ESC_h) - { - SETBIT(classbits, 0x09); /* VT */ - SETBIT(classbits, 0x20); /* SPACE */ - SETBIT(classbits, 0xa0); /* NSBP */ -#ifdef SUPPORT_UTF8 - if (utf8) - { - class_utf8 = TRUE; - *class_utf8data++ = XCL_SINGLE; - class_utf8data += _erts_pcre_ord2utf8(0x1680, class_utf8data); - *class_utf8data++ = XCL_SINGLE; - class_utf8data += _erts_pcre_ord2utf8(0x180e, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x2000, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x200A, class_utf8data); - *class_utf8data++ = XCL_SINGLE; - class_utf8data += _erts_pcre_ord2utf8(0x202f, class_utf8data); - *class_utf8data++ = XCL_SINGLE; - class_utf8data += _erts_pcre_ord2utf8(0x205f, class_utf8data); - *class_utf8data++ = XCL_SINGLE; - class_utf8data += _erts_pcre_ord2utf8(0x3000, class_utf8data); - } -#endif + case ESC_h: + (void)add_list_to_class(classbits, &class_uchardata, options, cd, + PRIV(hspace_list), NOTACHAR); continue; - } - if (-c == ESC_H) - { - for (c = 0; c < 32; c++) - { - int x = 0xff; - switch (c) - { - case 0x09/8: x ^= 1 << (0x09%8); break; - case 0x20/8: x ^= 1 << (0x20%8); break; - case 0xa0/8: x ^= 1 << (0xa0%8); break; - default: break; - } - classbits[c] |= x; - } + case ESC_H: + (void)add_not_list_to_class(classbits, &class_uchardata, options, + cd, PRIV(hspace_list)); + continue; -#ifdef SUPPORT_UTF8 - if (utf8) - { - class_utf8 = TRUE; - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x0100, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x167f, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x1681, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x180d, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x180f, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x1fff, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x200B, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x202e, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x2030, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x205e, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x2060, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x2fff, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x3001, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x7fffffff, class_utf8data); - } -#endif + case ESC_v: + (void)add_list_to_class(classbits, &class_uchardata, options, cd, + PRIV(vspace_list), NOTACHAR); continue; - } - if (-c == ESC_v) - { - SETBIT(classbits, 0x0a); /* LF */ - SETBIT(classbits, 0x0b); /* VT */ - SETBIT(classbits, 0x0c); /* FF */ - SETBIT(classbits, 0x0d); /* CR */ - SETBIT(classbits, 0x85); /* NEL */ -#ifdef SUPPORT_UTF8 - if (utf8) - { - class_utf8 = TRUE; - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x2028, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x2029, class_utf8data); - } -#endif + case ESC_V: + (void)add_not_list_to_class(classbits, &class_uchardata, options, + cd, PRIV(vspace_list)); continue; - } - if (-c == ESC_V) - { - for (c = 0; c < 32; c++) +#ifdef SUPPORT_UCP + case ESC_p: + case ESC_P: { - int x = 0xff; - switch (c) - { - case 0x0a/8: x ^= 1 << (0x0a%8); - x ^= 1 << (0x0b%8); - x ^= 1 << (0x0c%8); - x ^= 1 << (0x0d%8); - break; - case 0x85/8: x ^= 1 << (0x85%8); break; - default: break; - } - classbits[c] |= x; + BOOL negated; + unsigned int ptype = 0, pdata = 0; + if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr)) + goto FAILED; + *class_uchardata++ = ((escape == ESC_p) != negated)? + XCL_PROP : XCL_NOTPROP; + *class_uchardata++ = ptype; + *class_uchardata++ = pdata; + class_has_8bitchar--; /* Undo! */ + continue; } +#endif + /* Unrecognized escapes are faulted if PCRE is running in its + strict mode. By default, for compatibility with Perl, they are + treated as literals. */ -#ifdef SUPPORT_UTF8 - if (utf8) + default: + if ((options & PCRE_EXTRA) != 0) { - class_utf8 = TRUE; - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x0100, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x2027, class_utf8data); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(0x2029, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(0x7fffffff, class_utf8data); + *errorcodeptr = ERR7; + goto FAILED; } -#endif - continue; - } - - /* We need to deal with \P and \p in both phases. */ - -#ifdef SUPPORT_UCP - if (-c == ESC_p || -c == ESC_P) - { - BOOL negated; - int pdata; - int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr); - if (ptype < 0) goto FAILED; - class_utf8 = TRUE; - *class_utf8data++ = ((-c == ESC_p) != negated)? - XCL_PROP : XCL_NOTPROP; - *class_utf8data++ = ptype; - *class_utf8data++ = pdata; - class_charcount -= 2; /* Not a < 256 character */ - continue; - } -#endif - /* Unrecognized escapes are faulted if PCRE is running in its - strict mode. By default, for compatibility with Perl, they are - treated as literals. */ - - if ((options & PCRE_EXTRA) != 0) - { - *errorcodeptr = ERR7; - goto FAILED; + class_has_8bitchar--; /* Undo the speculative increase. */ + class_one_char -= 2; /* Undo the speculative increase. */ + c = *ptr; /* Get the final character and fall through */ + break; } - - class_charcount -= 2; /* Undo the default count from above */ - c = *ptr; /* Get the final character and fall through */ } - /* Fall through if we have a single character (c >= 0). This may be - greater than 256 in UTF-8 mode. */ + /* Fall through if the escape just defined a single character (c >= 0). + This may be greater than 256. */ + + escape = 0; } /* End of backslash handling */ - /* A single character may be followed by '-' to form a range. However, - Perl does not permit ']' to be the end of the range. A '-' character - at the end is treated as a literal. Perl ignores orphaned \E sequences - entirely. The code for handling \Q and \E is messy. */ + /* A character may be followed by '-' to form a range. However, Perl does + not permit ']' to be the end of the range. A '-' character at the end is + treated as a literal. Perl ignores orphaned \E sequences entirely. The + code for handling \Q and \E is messy. */ CHECK_RANGE: - while (ptr[1] == '\\' && ptr[2] == 'E') + while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E) { inescq = FALSE; ptr += 2; } - oldptr = ptr; - /* Remember \r or \n */ + /* Remember if \r or \n were explicitly used */ - if (c == '\r' || c == '\n') cd->external_flags |= PCRE_HASCRORLF; + if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF; /* Check for range */ - if (!inescq && ptr[1] == '-') + if (!inescq && ptr[1] == CHAR_MINUS) { - int d; + pcre_uint32 d; ptr += 2; - while (*ptr == '\\' && ptr[1] == 'E') ptr += 2; + while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2; /* If we hit \Q (not followed by \E) at this point, go into escaped mode. */ - while (*ptr == '\\' && ptr[1] == 'Q') + while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q) { ptr += 2; - if (*ptr == '\\' && ptr[1] == 'E') { ptr += 2; continue; } + if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) + { ptr += 2; continue; } inescq = TRUE; break; } - if (*ptr == 0 || (!inescq && *ptr == ']')) + /* Minus (hyphen) at the end of a class is treated as a literal, so put + back the pointer and jump to handle the character that preceded it. */ + + if (*ptr == CHAR_NULL || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET)) { ptr = oldptr; - goto LONE_SINGLE_CHARACTER; + goto CLASS_SINGLE_CHARACTER; } -#ifdef SUPPORT_UTF8 - if (utf8) + /* Otherwise, we have a potential range; pick up the next character */ + +#ifdef SUPPORT_UTF + if (utf) { /* Braces are required because the */ GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */ } @@ -3117,324 +4477,231 @@ for (;; ptr++) not any of the other escapes. Perl 5.6 treats a hyphen as a literal in such circumstances. */ - if (!inescq && d == '\\') + if (!inescq && d == CHAR_BACKSLASH) { - d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE); + int descape; + descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE); if (*errorcodeptr != 0) goto FAILED; - /* \b is backspace; \X is literal X; \R is literal R; any other - special means the '-' was literal */ + /* \b is backspace; any other special means the '-' was literal. */ - if (d < 0) + if (descape != 0) { - if (d == -ESC_b) d = '\b'; - else if (d == -ESC_X) d = 'X'; - else if (d == -ESC_R) d = 'R'; else + if (descape == ESC_b) d = CHAR_BS; else { ptr = oldptr; - goto LONE_SINGLE_CHARACTER; /* A few lines below */ + goto CLASS_SINGLE_CHARACTER; /* A few lines below */ } } } /* Check that the two values are in the correct order. Optimize - one-character ranges */ + one-character ranges. */ if (d < c) { *errorcodeptr = ERR8; goto FAILED; } + if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */ - if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */ - - /* Remember \r or \n */ - - if (d == '\r' || d == '\n') cd->external_flags |= PCRE_HASCRORLF; - - /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless - matching, we have to use an XCLASS with extra data items. Caseless - matching for characters > 127 is available only if UCP support is - available. */ + /* We have found a character range, so single character optimizations + cannot be done anymore. Any value greater than 1 indicates that there + is more than one character. */ -#ifdef SUPPORT_UTF8 - if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127))) - { - class_utf8 = TRUE; - - /* With UCP support, we can find the other case equivalents of - the relevant characters. There may be several ranges. Optimize how - they fit with the basic range. */ + class_one_char = 2; -#ifdef SUPPORT_UCP - if ((options & PCRE_CASELESS) != 0) - { - unsigned int occ, ocd; - unsigned int cc = c; - unsigned int origd = d; - while (get_othercase_range(&cc, origd, &occ, &ocd)) - { - if (occ >= (unsigned int)c && - ocd <= (unsigned int)d) - continue; /* Skip embedded ranges */ - - if (occ < (unsigned int)c && - ocd >= (unsigned int)c - 1) /* Extend the basic range */ - { /* if there is overlap, */ - c = occ; /* noting that if occ < c */ - continue; /* we can't have ocd > d */ - } /* because a subrange is */ - if (ocd > (unsigned int)d && - occ <= (unsigned int)d + 1) /* always shorter than */ - { /* the basic range. */ - d = ocd; - continue; - } + /* Remember an explicit \r or \n, and add the range to the class. */ - if (occ == ocd) - { - *class_utf8data++ = XCL_SINGLE; - } - else - { - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(occ, class_utf8data); - } - class_utf8data += _erts_pcre_ord2utf8(ocd, class_utf8data); - } - } -#endif /* SUPPORT_UCP */ + if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF; - /* Now record the original range, possibly modified for UCP caseless - overlapping ranges. */ + class_has_8bitchar += + add_to_class(classbits, &class_uchardata, options, cd, c, d); - *class_utf8data++ = XCL_RANGE; - class_utf8data += _erts_pcre_ord2utf8(c, class_utf8data); - class_utf8data += _erts_pcre_ord2utf8(d, class_utf8data); + continue; /* Go get the next char in the class */ + } - /* With UCP support, we are done. Without UCP support, there is no - caseless matching for UTF-8 characters > 127; we can use the bit map - for the smaller ones. */ + /* Handle a single character - we can get here for a normal non-escape + char, or after \ that introduces a single character or for an apparent + range that isn't. Only the value 1 matters for class_one_char, so don't + increase it if it is already 2 or more ... just in case there's a class + with a zillion characters in it. */ + + CLASS_SINGLE_CHARACTER: + if (class_one_char < 2) class_one_char++; + + /* If class_one_char is 1, we have the first single character in the + class, and there have been no prior ranges, or XCLASS items generated by + escapes. If this is the final character in the class, we can optimize by + turning the item into a 1-character OP_CHAR[I] if it's positive, or + OP_NOT[I] if it's negative. In the positive case, it can cause firstchar + to be set. Otherwise, there can be no first char if this item is first, + whatever repeat count may follow. In the case of reqchar, save the + previous value for reinstating. */ + + if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) + { + ptr++; + zeroreqchar = reqchar; + zeroreqcharflags = reqcharflags; + if (negate_class) + { #ifdef SUPPORT_UCP - continue; /* With next character in the class */ -#else - if ((options & PCRE_CASELESS) == 0 || c > 127) continue; - - /* Adjust upper limit and fall through to set up the map */ - - d = 127; - -#endif /* SUPPORT_UCP */ - } -#endif /* SUPPORT_UTF8 */ - - /* We use the bit map for all cases when not in UTF-8 mode; else - ranges that lie entirely within 0-127 when there is UCP support; else - for partial ranges without UCP support. */ - - class_charcount += d - c + 1; - class_lastchar = d; + int d; +#endif + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; + zerofirstchar = firstchar; + zerofirstcharflags = firstcharflags; - /* We can save a bit of time by skipping this in the pre-compile. */ + /* For caseless UTF-8 mode when UCP support is available, check + whether this character has more than one other case. If so, generate + a special OP_NOTPROP item instead of OP_NOTI. */ - if (lengthptr == NULL) for (; c <= d; c++) - { - classbits[c/8] |= (1 << (c&7)); - if ((options & PCRE_CASELESS) != 0) +#ifdef SUPPORT_UCP + if (utf && (options & PCRE_CASELESS) != 0 && + (d = UCD_CASESET(c)) != 0) { - int uc = cd->fcc[c]; /* flip case */ - classbits[uc/8] |= (1 << (uc&7)); + *code++ = OP_NOTPROP; + *code++ = PT_CLIST; + *code++ = d; } - } - - continue; /* Go get the next char in the class */ - } - - /* Handle a lone single character - we can get here for a normal - non-escape char, or after \ that introduces a single character or for an - apparent range that isn't. */ - - LONE_SINGLE_CHARACTER: - - /* Handle a character that cannot go in the bit map */ - -#ifdef SUPPORT_UTF8 - if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127))) - { - class_utf8 = TRUE; - *class_utf8data++ = XCL_SINGLE; - class_utf8data += _erts_pcre_ord2utf8(c, class_utf8data); + else +#endif + /* Char has only one other case, or UCP not available */ -#ifdef SUPPORT_UCP - if ((options & PCRE_CASELESS) != 0) - { - unsigned int othercase; - if ((othercase = _erts_pcre_ucp_othercase(c)) != NOTACHAR) { - *class_utf8data++ = XCL_SINGLE; - class_utf8data += _erts_pcre_ord2utf8(othercase, class_utf8data); + *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT; +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR) + code += PRIV(ord2utf)(c, code); + else +#endif + *code++ = c; } + + /* We are finished with this character class */ + + goto END_CLASS; } -#endif /* SUPPORT_UCP */ - } - else -#endif /* SUPPORT_UTF8 */ + /* For a single, positive character, get the value into mcbuffer, and + then we can handle this with the normal one-character code. */ - /* Handle a single-byte character */ - { - classbits[c/8] |= (1 << (c&7)); - if ((options & PCRE_CASELESS) != 0) +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR) + mclength = PRIV(ord2utf)(c, mcbuffer); + else +#endif { - c = cd->fcc[c]; /* flip case */ - classbits[c/8] |= (1 << (c&7)); + mcbuffer[0] = c; + mclength = 1; } - class_charcount++; - class_lastchar = c; - } - } - - /* Loop until ']' reached. This "while" is the end of the "do" above. */ + goto ONE_CHAR; + } /* End of 1-char optimization */ - while ((c = *(++ptr)) != 0 && (c != ']' || inescq)); + /* There is more than one character in the class, or an XCLASS item + has been generated. Add this character to the class. */ - if (c == 0) /* Missing terminating ']' */ - { - *errorcodeptr = ERR6; - goto FAILED; + class_has_8bitchar += + add_to_class(classbits, &class_uchardata, options, cd, c, c); } + /* Loop until ']' reached. This "while" is the end of the "do" far above. + If we are at the end of an internal nested string, revert to the outer + string. */ -/* This code has been disabled because it would mean that \s counts as -an explicit \r or \n reference, and that's not really what is wanted. Now -we set the flag only if there is a literal "\r" or "\n" in the class. */ + while (((c = *(++ptr)) != CHAR_NULL || + (nestptr != NULL && + (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != CHAR_NULL)) && + (c != CHAR_RIGHT_SQUARE_BRACKET || inescq)); -#if 0 - /* Remember whether \r or \n are in this class */ + /* Check for missing terminating ']' */ - if (negate_class) + if (c == CHAR_NULL) { - if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF; - } - else - { - if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF; + *errorcodeptr = ERR6; + goto FAILED; } -#endif - - - /* If class_charcount is 1, we saw precisely one character whose value is - less than 256. As long as there were no characters >= 128 and there was no - use of \p or \P, in other words, no use of any XCLASS features, we can - optimize. - In UTF-8 mode, we can optimize the negative case only if there were no - characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR - operate on single-bytes only. This is an historical hangover. Maybe one day - we can tidy these opcodes to handle multi-byte characters. + /* We will need an XCLASS if data has been placed in class_uchardata. In + the second phase this is a sufficient test. However, in the pre-compile + phase, class_uchardata gets emptied to prevent workspace overflow, so it + only if the very last character in the class needs XCLASS will it contain + anything at this point. For this reason, xclass gets set TRUE above when + uchar_classdata is emptied, and that's why this code is the way it is here + instead of just doing a test on class_uchardata below. */ - The optimization throws away the bit map. We turn the item into a - 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note - that OP_NOT does not support multibyte characters. In the positive case, it - can cause firstbyte to be set. Otherwise, there can be no first char if - this item is first, whatever repeat count may follow. In the case of - reqbyte, save the previous value for reinstating. */ - -#ifdef SUPPORT_UTF8 - if (class_charcount == 1 && !class_utf8 && - (!utf8 || !negate_class || class_lastchar < 128)) -#else - if (class_charcount == 1) +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 + if (class_uchardata > class_uchardata_base) xclass = TRUE; #endif - { - zeroreqbyte = reqbyte; - - /* The OP_NOT opcode works on one-byte characters only. */ - - if (negate_class) - { - if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; - zerofirstbyte = firstbyte; - *code++ = OP_NOT; - *code++ = class_lastchar; - break; - } - - /* For a single, positive character, get the value into mcbuffer, and - then we can handle this with the normal one-character code. */ - -#ifdef SUPPORT_UTF8 - if (utf8 && class_lastchar > 127) - mclength = _erts_pcre_ord2utf8(class_lastchar, mcbuffer); - else -#endif - { - mcbuffer[0] = class_lastchar; - mclength = 1; - } - goto ONE_CHAR; - } /* End of 1-char optimization */ - /* The general case - not the one-char optimization. If this is the first - thing in the branch, there can be no first char setting, whatever the - repeat count. Any reqbyte setting must remain unchanged after any kind of - repeat. */ + /* If this is the first thing in the branch, there can be no first char + setting, whatever the repeat count. Any reqchar setting must remain + unchanged after any kind of repeat. */ - if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; - zerofirstbyte = firstbyte; - zeroreqbyte = reqbyte; + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; + zerofirstchar = firstchar; + zerofirstcharflags = firstcharflags; + zeroreqchar = reqchar; + zeroreqcharflags = reqcharflags; /* If there are characters with values > 255, we have to compile an extended class, with its own opcode, unless there was a negated special - such as \S in the class, because in that case all characters > 255 are in - the class, so any that were explicitly given as well can be ignored. If - (when there are explicit characters > 255 that must be listed) there are no - characters < 256, we can omit the bitmap in the actual compiled code. */ - -#ifdef SUPPORT_UTF8 - if (class_utf8 && !should_flip_negation) + such as \S in the class, and PCRE_UCP is not set, because in that case all + characters > 255 are in the class, so any that were explicitly given as + well can be ignored. If (when there are explicit characters > 255 that must + be listed) there are no characters < 256, we can omit the bitmap in the + actual compiled code. */ + +#ifdef SUPPORT_UTF + if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0)) +#elif !defined COMPILE_PCRE8 + if (xclass && !should_flip_negation) +#endif +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 { - *class_utf8data++ = XCL_END; /* Marks the end of extra data */ + *class_uchardata++ = XCL_END; /* Marks the end of extra data */ *code++ = OP_XCLASS; code += LINK_SIZE; - *code = negate_class? XCL_NOT : 0; + *code = negate_class? XCL_NOT:0; /* If the map is required, move up the extra data to make room for it; otherwise just move the code pointer to the end of the extra data. */ - if (class_charcount > 0) + if (class_has_8bitchar > 0) { *code++ |= XCL_MAP; - memmove(code + 32, code, class_utf8data - code); + memmove(code + (32 / sizeof(pcre_uchar)), code, + IN_UCHARS(class_uchardata - code)); memcpy(code, classbits, 32); - code = class_utf8data + 32; + code = class_uchardata + (32 / sizeof(pcre_uchar)); } - else code = class_utf8data; + else code = class_uchardata; /* Now fill in the complete length of the item */ - PUT(previous, 1, code - previous); + PUT(previous, 1, (int)(code - previous)); break; /* End of class handling */ } #endif - /* If there are no characters > 255, set the opcode to OP_CLASS or - OP_NCLASS, depending on whether the whole class was negated and whether - there were negative specials such as \S in the class. Then copy the 32-byte - map into the code vector, negating it if necessary. */ + /* If there are no characters > 255, or they are all to be included or + excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the + whole class was negated and whether there were negative specials such as \S + (non-UCP) in the class. Then copy the 32-byte map into the code vector, + negating it if necessary. */ *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS; - if (negate_class) - { - if (lengthptr == NULL) /* Save time in the pre-compile phase */ - for (c = 0; c < 32; c++) code[c] = ~classbits[c]; - } - else + if (lengthptr == NULL) /* Save time in the pre-compile phase */ { + if (negate_class) + for (c = 0; c < 32; c++) classbits[c] = ~classbits[c]; memcpy(code, classbits, 32); } - code += 32; + code += 32 / sizeof(pcre_uchar); + + END_CLASS: break; @@ -3442,23 +4709,23 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* Various kinds of repeat; '{' is not necessarily a quantifier, but this has been tested above. */ - case '{': + case CHAR_LEFT_CURLY_BRACKET: if (!is_quantifier) goto NORMAL_CHAR; ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr); if (*errorcodeptr != 0) goto FAILED; goto REPEAT; - case '*': + case CHAR_ASTERISK: repeat_min = 0; repeat_max = -1; goto REPEAT; - case '+': + case CHAR_PLUS: repeat_min = 1; repeat_max = -1; goto REPEAT; - case '?': + case CHAR_QUESTION_MARK: repeat_min = 0; repeat_max = 1; @@ -3471,8 +4738,10 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ if (repeat_min == 0) { - firstbyte = zerofirstbyte; /* Adjust for zero repeat */ - reqbyte = zeroreqbyte; /* Ditto */ + firstchar = zerofirstchar; /* Adjust for zero repeat */ + firstcharflags = zerofirstcharflags; + reqchar = zeroreqchar; /* Ditto */ + reqcharflags = zeroreqcharflags; } /* Remember whether this is a variable length repeat */ @@ -3482,8 +4751,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ op_type = 0; /* Default single-char op codes */ possessive_quantifier = FALSE; /* Default not possessive quantifier */ - /* Save start of previous item, in case we have to move it up to make space - for an inserted OP_ONCE for the additional '+' extension. */ + /* Save start of previous item, in case we have to move it up in order to + insert something before it. */ tempcode = previous; @@ -3493,50 +4762,92 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ but if PCRE_UNGREEDY is set, it works the other way round. We change the repeat type to the non-default. */ - if (ptr[1] == '+') + if (ptr[1] == CHAR_PLUS) { repeat_type = 0; /* Force greedy */ possessive_quantifier = TRUE; ptr++; } - else if (ptr[1] == '?') + else if (ptr[1] == CHAR_QUESTION_MARK) { repeat_type = greedy_non_default; ptr++; } else repeat_type = greedy_default; - /* If previous was a character match, abolish the item and generate a - repeat item instead. If a char item has a minumum of more than one, ensure - that it is set in reqbyte - it might not be if a sequence such as x{3} is - the first thing in a branch because the x will have gone into firstbyte - instead. */ + /* If previous was a recursion call, wrap it in atomic brackets so that + previous becomes the atomic group. All recursions were so wrapped in the + past, but it no longer happens for non-repeated recursions. In fact, the + repeated ones could be re-implemented independently so as not to need this, + but for the moment we rely on the code for repeating groups. */ - if (*previous == OP_CHAR || *previous == OP_CHARNC) + if (*previous == OP_RECURSE) { - /* Deal with UTF-8 characters that take up more than one byte. It's + memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE)); + *previous = OP_ONCE; + PUT(previous, 1, 2 + 2*LINK_SIZE); + previous[2 + 2*LINK_SIZE] = OP_KET; + PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE); + code += 2 + 2 * LINK_SIZE; + length_prevgroup = 3 + 3*LINK_SIZE; + + /* When actually compiling, we need to check whether this was a forward + reference, and if so, adjust the offset. */ + + if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE) + { + int offset = GET(cd->hwm, -LINK_SIZE); + if (offset == previous + 1 - cd->start_code) + PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE); + } + } + + /* Now handle repetition for the different types of item. */ + + /* If previous was a character or negated character match, abolish the item + and generate a repeat item instead. If a char item has a minimum of more + than one, ensure that it is set in reqchar - it might not be if a sequence + such as x{3} is the first thing in a branch because the x will have gone + into firstchar instead. */ + + if (*previous == OP_CHAR || *previous == OP_CHARI + || *previous == OP_NOT || *previous == OP_NOTI) + { + switch (*previous) + { + default: /* Make compiler happy. */ + case OP_CHAR: op_type = OP_STAR - OP_STAR; break; + case OP_CHARI: op_type = OP_STARI - OP_STAR; break; + case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break; + case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break; + } + + /* Deal with UTF characters that take up more than one character. It's easier to write this out separately than try to macrify it. Use c to - hold the length of the character in bytes, plus 0x80 to flag that it's a - length rather than a small character. */ + hold the length of the character in bytes, plus UTF_LENGTH to flag that + it's a length rather than a small character. */ -#ifdef SUPPORT_UTF8 - if (utf8 && (code[-1] & 0x80) != 0) +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf && NOT_FIRSTCHAR(code[-1])) { - uschar *lastchar = code - 1; - while((*lastchar & 0xc0) == 0x80) lastchar--; - c = code - lastchar; /* Length of UTF-8 character */ - memcpy(utf8_char, lastchar, c); /* Save the char */ - c |= 0x80; /* Flag c as a length */ + pcre_uchar *lastchar = code - 1; + BACKCHAR(lastchar); + c = (int)(code - lastchar); /* Length of UTF-8 character */ + memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */ + c |= UTF_LENGTH; /* Flag c as a length */ } else -#endif - - /* Handle the case of a single byte - either with no UTF8 support, or - with UTF-8 disabled, or for a UTF-8 character < 128. */ +#endif /* SUPPORT_UTF */ + /* Handle the case of a single charater - either with no UTF support, or + with UTF disabled, or for a single character UTF character. */ { c = code[-1]; - if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt; + if (*previous <= OP_CHARI && repeat_min > 1) + { + reqchar = c; + reqcharflags = req_caseopt | cd->req_varyopt; + } } /* If the repetition is unlimited, it pays to see if the next thing on @@ -3546,8 +4857,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ if (!possessive_quantifier && repeat_max < 0 && - check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1, - options, cd)) + check_auto_possessive(previous, utf, ptr + 1, options, cd)) { repeat_type = 0; /* Force greedy */ possessive_quantifier = TRUE; @@ -3556,26 +4866,6 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */ } - /* If previous was a single negated character ([^a] or similar), we use - one of the special opcodes, replacing it. The code is shared with single- - character repeats by setting opt_type to add a suitable offset into - repeat_type. We can also test for auto-possessification. OP_NOT is - currently used only for single-byte chars. */ - - else if (*previous == OP_NOT) - { - op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */ - c = previous[1]; - if (!possessive_quantifier && - repeat_max < 0 && - check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd)) - { - repeat_type = 0; /* Force greedy */ - possessive_quantifier = TRUE; - } - goto OUTPUT_SINGLE_REPEAT; - } - /* If previous was a character type match (\d or similar), abolish it and create a suitable repeat item. The code is shared with single-character repeats by setting op_type to add a suitable offset into repeat_type. Note @@ -3585,14 +4875,14 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ else if (*previous < OP_EODN) { - uschar *oldcode; + pcre_uchar *oldcode; int prop_type, prop_value; op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */ c = *previous; if (!possessive_quantifier && repeat_max < 0 && - check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd)) + check_auto_possessive(previous, utf, ptr + 1, options, cd)) { repeat_type = 0; /* Force greedy */ possessive_quantifier = TRUE; @@ -3614,11 +4904,6 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ if (repeat_max == 0) goto END_REPEAT; - /* All real repeats make it impossible to handle partial matching (maybe - one day we will be able to remove this restriction). */ - - if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; - /* Combine the op_type with the repeat_type */ repeat_type += op_type; @@ -3667,14 +4952,14 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ we have to insert the character for the previous code. For a repeated Unicode property match, there are two extra bytes that define the required property. In UTF-8 mode, long characters have their length in - c, with the 0x80 bit as a flag. */ + c, with the UTF_LENGTH bit as a flag. */ if (repeat_max < 0) { -#ifdef SUPPORT_UTF8 - if (utf8 && c >= 128) +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf && (c & UTF_LENGTH) != 0) { - memcpy(code, utf8_char, c & 7); + memcpy(code, utf_chars, IN_UCHARS(c & 7)); code += c & 7; } else @@ -3696,10 +4981,10 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ else if (repeat_max != repeat_min) { -#ifdef SUPPORT_UTF8 - if (utf8 && c >= 128) +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf && (c & UTF_LENGTH) != 0) { - memcpy(code, utf8_char, c & 7); + memcpy(code, utf_chars, IN_UCHARS(c & 7)); code += c & 7; } else @@ -3726,10 +5011,10 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* The character or character type itself comes last in all cases. */ -#ifdef SUPPORT_UTF8 - if (utf8 && c >= 128) +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf && (c & UTF_LENGTH) != 0) { - memcpy(code, utf8_char, c & 7); + memcpy(code, utf_chars, IN_UCHARS(c & 7)); code += c & 7; } else @@ -3753,10 +5038,11 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ else if (*previous == OP_CLASS || *previous == OP_NCLASS || -#ifdef SUPPORT_UTF8 +#if defined SUPPORT_UTF || !defined COMPILE_PCRE8 *previous == OP_XCLASS || #endif - *previous == OP_REF) + *previous == OP_REF || + *previous == OP_REFI) { if (repeat_max == 0) { @@ -3764,11 +5050,6 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ goto END_REPEAT; } - /* All real repeats make it impossible to handle partial matching (maybe - one day we will be able to remove this restriction). */ - - if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; - if (repeat_min == 0 && repeat_max == -1) *code++ = OP_CRSTAR + repeat_type; else if (repeat_min == 1 && repeat_max == -1) @@ -3785,35 +5066,35 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } /* If previous was a bracket group, we may have to replicate it in certain - cases. */ - - else if (*previous == OP_BRA || *previous == OP_CBRA || - *previous == OP_ONCE || *previous == OP_COND) + cases. Note that at this point we can encounter only the "basic" bracket + opcodes such as BRA and CBRA, as this is the place where they get converted + into the more special varieties such as BRAPOS and SBRA. A test for >= + OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK, + ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow + repetition of assertions, but now it does, for Perl compatibility. */ + + else if (*previous >= OP_ASSERT && *previous <= OP_COND) { register int i; - int ketoffset = 0; - int len = code - previous; - uschar *bralink = NULL; + int len = (int)(code - previous); + pcre_uchar *bralink = NULL; + pcre_uchar *brazeroptr = NULL; - /* Repeating a DEFINE group is pointless */ + /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so + we just ignore the repeat. */ if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF) - { - *errorcodeptr = ERR55; - goto FAILED; - } + goto END_REPEAT; - /* If the maximum repeat count is unlimited, find the end of the bracket - by scanning through from the start, and compute the offset back to it - from the current code pointer. There may be an OP_OPT setting following - the final KET, so we can't find the end just by going back from the code - pointer. */ + /* There is no sense in actually repeating assertions. The only potential + use of repetition is in cases when the assertion is optional. Therefore, + if the minimum is greater than zero, just ignore the repeat. If the + maximum is not not zero or one, set it to 1. */ - if (repeat_max == -1) + if (*previous < OP_ONCE) /* Assertion */ { - register uschar *ket = previous; - do ket += GET(ket, 1); while (*ket != OP_KET); - ketoffset = code - ket; + if (repeat_min > 0) goto END_REPEAT; + if (repeat_max < 0 || repeat_max > 1) repeat_max = 1; } /* The case of a zero minimum is special because of the need to stick @@ -3825,28 +5106,40 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ if (repeat_min == 0) { - /* If the maximum is also zero, we just omit the group from the output - altogether. */ - - if (repeat_max == 0) - { - code = previous; - goto END_REPEAT; - } - - /* If the maximum is 1 or unlimited, we just have to stick in the - BRAZERO and do no more at this point. However, we do need to adjust - any OP_RECURSE calls inside the group that refer to the group itself or - any internal or forward referenced group, because the offset is from - the start of the whole regex. Temporarily terminate the pattern while - doing this. */ - - if (repeat_max <= 1) + /* If the maximum is also zero, we used to just omit the group from the + output altogether, like this: + + ** if (repeat_max == 0) + ** { + ** code = previous; + ** goto END_REPEAT; + ** } + + However, that fails when a group or a subgroup within it is referenced + as a subroutine from elsewhere in the pattern, so now we stick in + OP_SKIPZERO in front of it so that it is skipped on execution. As we + don't have a list of which groups are referenced, we cannot do this + selectively. + + If the maximum is 1 or unlimited, we just have to stick in the BRAZERO + and do no more at this point. However, we do need to adjust any + OP_RECURSE calls inside the group that refer to the group itself or any + internal or forward referenced group, because the offset is from the + start of the whole regex. Temporarily terminate the pattern while doing + this. */ + + if (repeat_max <= 1) /* Covers 0, 1, and unlimited */ { *code = OP_END; - adjust_recurse(previous, 1, utf8, cd, save_hwm); - memmove(previous+1, previous, len); + adjust_recurse(previous, 1, utf, cd, save_hwm); + memmove(previous + 1, previous, IN_UCHARS(len)); code++; + if (repeat_max == 0) + { + *previous++ = OP_SKIPZERO; + goto END_REPEAT; + } + brazeroptr = previous; /* Save for possessive optimizing */ *previous++ = OP_BRAZERO + repeat_type; } @@ -3862,8 +5155,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ { int offset; *code = OP_END; - adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm); - memmove(previous + 2 + LINK_SIZE, previous, len); + adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm); + memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len)); code += 2 + LINK_SIZE; *previous++ = OP_BRAZERO + repeat_type; *previous++ = OP_BRA; @@ -3871,7 +5164,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* We chain together the bracket offset fields that have to be filled in later when the ends of the brackets are reached. */ - offset = (bralink == NULL)? 0 : previous - bralink; + offset = (bralink == NULL)? 0 : (int)(previous - bralink); bralink = previous; PUTINC(previous, 0, offset); } @@ -3892,13 +5185,15 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ { /* In the pre-compile phase, we don't actually do the replication. We just adjust the length as if we had. Do some paranoid checks for - potential integer overflow. */ + potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit + integer type when available, otherwise double. */ if (lengthptr != NULL) { int delta = (repeat_min - 1)*length_prevgroup; - if ((double)(repeat_min - 1)*(double)length_prevgroup > - (double)INT_MAX || + if ((INT64_OR_DOUBLE)(repeat_min - 1)* + (INT64_OR_DOUBLE)length_prevgroup > + (INT64_OR_DOUBLE)INT_MAX || OFLOW_MAX - *lengthptr < delta) { *errorcodeptr = ERR20; @@ -3907,16 +5202,36 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ *lengthptr += delta; } - /* This is compiling for real */ + /* This is compiling for real. If there is a set first byte for + the group, and we have not yet set a "required byte", set it. Make + sure there is enough workspace for copying forward references before + doing the copy. */ else { - if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte; + if (groupsetfirstchar && reqcharflags < 0) + { + reqchar = firstchar; + reqcharflags = firstcharflags; + } + for (i = 1; i < repeat_min; i++) { - uschar *hc; - uschar *this_hwm = cd->hwm; - memcpy(code, previous, len); + pcre_uchar *hc; + pcre_uchar *this_hwm = cd->hwm; + memcpy(code, previous, IN_UCHARS(len)); + + while (cd->hwm > cd->start_workspace + cd->workspace_size - + WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm)) + { + int save_offset = save_hwm - cd->start_workspace; + int this_offset = this_hwm - cd->start_workspace; + *errorcodeptr = expand_workspace(cd); + if (*errorcodeptr != 0) goto FAILED; + save_hwm = (pcre_uchar *)cd->start_workspace + save_offset; + this_hwm = (pcre_uchar *)cd->start_workspace + this_offset; + } + for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE) { PUT(cd->hwm, 0, GET(hc, 0) + len); @@ -3944,15 +5259,16 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ just adjust the length as if we had. For each repetition we must add 1 to the length for BRAZERO and for all but the last repetition we must add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some - paranoid checks to avoid integer overflow. */ + paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is + a 64-bit integer type when available, otherwise double. */ if (lengthptr != NULL && repeat_max > 0) { int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) - 2 - 2*LINK_SIZE; /* Last one doesn't nest */ - if ((double)repeat_max * - (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE) - > (double)INT_MAX || + if ((INT64_OR_DOUBLE)repeat_max * + (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE) + > (INT64_OR_DOUBLE)INT_MAX || OFLOW_MAX - *lengthptr < delta) { *errorcodeptr = ERR20; @@ -3965,8 +5281,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ else for (i = repeat_max - 1; i >= 0; i--) { - uschar *hc; - uschar *this_hwm = cd->hwm; + pcre_uchar *hc; + pcre_uchar *this_hwm = cd->hwm; *code++ = OP_BRAZERO + repeat_type; @@ -3977,12 +5293,27 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ { int offset; *code++ = OP_BRA; - offset = (bralink == NULL)? 0 : code - bralink; + offset = (bralink == NULL)? 0 : (int)(code - bralink); bralink = code; PUTINC(code, 0, offset); } - memcpy(code, previous, len); + memcpy(code, previous, IN_UCHARS(len)); + + /* Ensure there is enough workspace for forward references before + copying them. */ + + while (cd->hwm > cd->start_workspace + cd->workspace_size - + WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm)) + { + int save_offset = save_hwm - cd->start_workspace; + int this_offset = this_hwm - cd->start_workspace; + *errorcodeptr = expand_workspace(cd); + if (*errorcodeptr != 0) goto FAILED; + save_hwm = (pcre_uchar *)cd->start_workspace + save_offset; + this_hwm = (pcre_uchar *)cd->start_workspace + this_offset; + } + for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE) { PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1)); @@ -3998,8 +5329,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ while (bralink != NULL) { int oldlinkoffset; - int offset = code - bralink + 1; - uschar *bra = code - offset; + int offset = (int)(code - bralink + 1); + pcre_uchar *bra = code - offset; oldlinkoffset = GET(bra, 1); bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset; *code++ = OP_KET; @@ -4008,39 +5339,121 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } } - /* If the maximum is unlimited, set a repeater in the final copy. We - can't just offset backwards from the current code point, because we - don't know if there's been an options resetting after the ket. The - correct offset was computed above. + /* If the maximum is unlimited, set a repeater in the final copy. For + ONCE brackets, that's all we need to do. However, possessively repeated + ONCE brackets can be converted into non-capturing brackets, as the + behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to + deal with possessive ONCEs specially. - Then, when we are doing the actual compile phase, check to see whether - this group is a non-atomic one that could match an empty string. If so, + Otherwise, when we are doing the actual compile phase, check to see + whether this group is one that could match an empty string. If so, convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so - that runtime checking can be done. [This check is also applied to - atomic groups at runtime, but in a different way.] */ + that runtime checking can be done. [This check is also applied to ONCE + groups at runtime, but in a different way.] + + Then, if the quantifier was possessive and the bracket is not a + conditional, we convert the BRA code to the POS form, and the KET code to + KETRPOS. (It turns out to be convenient at runtime to detect this kind of + subpattern at both the start and at the end.) The use of special opcodes + makes it possible to reduce greatly the stack usage in pcre_exec(). If + the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO. + + Then, if the minimum number of matches is 1 or 0, cancel the possessive + flag so that the default action below, of wrapping everything inside + atomic brackets, does not happen. When the minimum is greater than 1, + there will be earlier copies of the group, and so we still have to wrap + the whole thing. */ else { - uschar *ketcode = code - ketoffset; - uschar *bracode = ketcode - GET(ketcode, 1); - *ketcode = OP_KETRMAX + repeat_type; - if (lengthptr == NULL && *bracode != OP_ONCE) + pcre_uchar *ketcode = code - 1 - LINK_SIZE; + pcre_uchar *bracode = ketcode - GET(ketcode, 1); + + /* Convert possessive ONCE brackets to non-capturing */ + + if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) && + possessive_quantifier) *bracode = OP_BRA; + + /* For non-possessive ONCE brackets, all we need to do is to + set the KET. */ + + if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC) + *ketcode = OP_KETRMAX + repeat_type; + + /* Handle non-ONCE brackets and possessive ONCEs (which have been + converted to non-capturing above). */ + + else { - uschar *scode = bracode; - do + /* In the compile phase, check for empty string matching. */ + + if (lengthptr == NULL) { - if (could_be_empty_branch(scode, ketcode, utf8)) + pcre_uchar *scode = bracode; + do { - *bracode += OP_SBRA - OP_BRA; - break; + if (could_be_empty_branch(scode, ketcode, utf, cd)) + { + *bracode += OP_SBRA - OP_BRA; + break; + } + scode += GET(scode, 1); + } + while (*scode == OP_ALT); + } + + /* Handle possessive quantifiers. */ + + if (possessive_quantifier) + { + /* For COND brackets, we wrap the whole thing in a possessively + repeated non-capturing bracket, because we have not invented POS + versions of the COND opcodes. Because we are moving code along, we + must ensure that any pending recursive references are updated. */ + + if (*bracode == OP_COND || *bracode == OP_SCOND) + { + int nlen = (int)(code - bracode); + *code = OP_END; + adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm); + memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen)); + code += 1 + LINK_SIZE; + nlen += 1 + LINK_SIZE; + *bracode = OP_BRAPOS; + *code++ = OP_KETRPOS; + PUTINC(code, 0, nlen); + PUT(bracode, 1, nlen); + } + + /* For non-COND brackets, we modify the BRA code and use KETRPOS. */ + + else + { + *bracode += 1; /* Switch to xxxPOS opcodes */ + *ketcode = OP_KETRPOS; } - scode += GET(scode, 1); + + /* If the minimum is zero, mark it as possessive, then unset the + possessive flag when the minimum is 0 or 1. */ + + if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO; + if (repeat_min < 2) possessive_quantifier = FALSE; } - while (*scode == OP_ALT); + + /* Non-possessive quantifier */ + + else *ketcode = OP_KETRMAX + repeat_type; } } } + /* If previous is OP_FAIL, it was generated by an empty class [] in + JavaScript mode. The other ways in which OP_FAIL can be generated, that is + by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat" + error above. We can just ignore the repeat in JS case. */ + + else if (*previous == OP_FAIL) goto END_REPEAT; + /* Else there's some kind of shambles */ else @@ -4050,13 +5463,18 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } /* If the character following a repeat is '+', or if certain optimization - tests above succeeded, possessive_quantifier is TRUE. For some of the - simpler opcodes, there is an special alternative opcode for this. For - anything else, we wrap the entire repeated item inside OP_ONCE brackets. - The '+' notation is just syntactic sugar, taken from Sun's Java package, - but the special opcodes can optimize it a bit. The repeated item starts at - tempcode, not at previous, which might be the first part of a string whose - (former) last char we repeated. + tests above succeeded, possessive_quantifier is TRUE. For some opcodes, + there are special alternative opcodes for this case. For anything else, we + wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+' + notation is just syntactic sugar, taken from Sun's Java package, but the + special opcodes can optimize it. + + Some (but not all) possessively repeated subpatterns have already been + completely handled in the code just above. For them, possessive_quantifier + is always FALSE at this stage. + + Note that the repeated item starts at tempcode, not at previous, which + might be the first part of a string whose (former) last char we repeated. Possessifying an 'exact' quantifier has no effect, so we can ignore it. But an 'upto' may follow. We skip over an 'exact' item, and then test the @@ -4065,12 +5483,22 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ if (possessive_quantifier) { int len; - if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT || - *tempcode == OP_NOTEXACT) - tempcode += _erts_pcre_OP_lengths[*tempcode] + - ((*tempcode == OP_TYPEEXACT && - (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0); - len = code - tempcode; + + if (*tempcode == OP_TYPEEXACT) + tempcode += PRIV(OP_lengths)[*tempcode] + + ((tempcode[1 + IMM2_SIZE] == OP_PROP + || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0); + + else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT) + { + tempcode += PRIV(OP_lengths)[*tempcode]; +#ifdef SUPPORT_UTF + if (utf && HAS_EXTRALEN(tempcode[-1])) + tempcode += GET_EXTRALEN(tempcode[-1]); +#endif + } + + len = (int)(code - tempcode); if (len > 0) switch (*tempcode) { case OP_STAR: *tempcode = OP_POSSTAR; break; @@ -4078,18 +5506,33 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ case OP_QUERY: *tempcode = OP_POSQUERY; break; case OP_UPTO: *tempcode = OP_POSUPTO; break; - case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break; - case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break; - case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break; - case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break; + case OP_STARI: *tempcode = OP_POSSTARI; break; + case OP_PLUSI: *tempcode = OP_POSPLUSI; break; + case OP_QUERYI: *tempcode = OP_POSQUERYI; break; + case OP_UPTOI: *tempcode = OP_POSUPTOI; break; case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break; case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break; case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break; case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break; + case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break; + case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break; + case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break; + case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break; + + case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break; + case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break; + case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break; + case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break; + + /* Because we are moving code along, we must ensure that any + pending recursive references are updated. */ + default: - memmove(tempcode + 1+LINK_SIZE, tempcode, len); + *code = OP_END; + adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm); + memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len)); code += 1 + LINK_SIZE; len += 1 + LINK_SIZE; tempcode[0] = OP_ONCE; @@ -4101,7 +5544,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } /* In all case we no longer have a previous item. We also set the - "follows varying string" flag for subsequently encountered reqbytes if + "follows varying string" flag for subsequently encountered reqchars if it isn't already set and we have just passed a varying length item. */ END_REPEAT: @@ -4115,7 +5558,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ lookbehind or option setting or condition or all the other extended parenthesis forms. */ - case '(': + case CHAR_LEFT_PARENTHESIS: newoptions = options; skipbytes = 0; bravalue = OP_CBRA; @@ -4124,56 +5567,143 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* First deal with various "verbs" that can be introduced by '*'. */ - if (*(++ptr) == '*' && (cd->ctypes[ptr[1]] & ctype_letter) != 0) + ptr++; + if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':' + || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0)))) { int i, namelen; + int arglen = 0; const char *vn = verbnames; - const uschar *name = ++ptr; + const pcre_uchar *name = ptr + 1; + const pcre_uchar *arg = NULL; previous = NULL; - while ((cd->ctypes[*++ptr] & ctype_letter) != 0); - if (*ptr == ':') + ptr++; + while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++; + namelen = (int)(ptr - name); + + /* It appears that Perl allows any characters whatsoever, other than + a closing parenthesis, to appear in arguments, so we no longer insist on + letters, digits, and underscores. */ + + if (*ptr == CHAR_COLON) { - *errorcodeptr = ERR59; /* Not supported */ - goto FAILED; + arg = ++ptr; + while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++; + arglen = (int)(ptr - arg); + if ((unsigned int)arglen > MAX_MARK) + { + *errorcodeptr = ERR75; + goto FAILED; + } } - if (*ptr != ')') + + if (*ptr != CHAR_RIGHT_PARENTHESIS) { *errorcodeptr = ERR60; goto FAILED; } - namelen = ptr - name; + + /* Scan the table of verb names */ + for (i = 0; i < verbcount; i++) { if (namelen == verbs[i].len && - strncmp((char *)name, vn, namelen) == 0) + STRNCMP_UC_C8(name, vn, namelen) == 0) { - *code = verbs[i].op; - if (*code++ == OP_ACCEPT) cd->had_accept = TRUE; - break; + int setverb; + + /* Check for open captures before ACCEPT and convert it to + ASSERT_ACCEPT if in an assertion. */ + + if (verbs[i].op == OP_ACCEPT) + { + open_capitem *oc; + if (arglen != 0) + { + *errorcodeptr = ERR59; + goto FAILED; + } + cd->had_accept = TRUE; + for (oc = cd->open_caps; oc != NULL; oc = oc->next) + { + *code++ = OP_CLOSE; + PUT2INC(code, 0, oc->number); + } + setverb = *code++ = + (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT; + + /* Do not set firstchar after *ACCEPT */ + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; + } + + /* Handle other cases with/without an argument */ + + else if (arglen == 0) + { + if (verbs[i].op < 0) /* Argument is mandatory */ + { + *errorcodeptr = ERR66; + goto FAILED; + } + setverb = *code++ = verbs[i].op; + } + + else + { + if (verbs[i].op_arg < 0) /* Argument is forbidden */ + { + *errorcodeptr = ERR59; + goto FAILED; + } + setverb = *code++ = verbs[i].op_arg; + *code++ = arglen; + memcpy(code, arg, IN_UCHARS(arglen)); + code += arglen; + *code++ = 0; + } + + switch (setverb) + { + case OP_THEN: + case OP_THEN_ARG: + cd->external_flags |= PCRE_HASTHEN; + break; + + case OP_PRUNE: + case OP_PRUNE_ARG: + case OP_SKIP: + case OP_SKIP_ARG: + cd->had_pruneorskip = TRUE; + break; + } + + break; /* Found verb, exit loop */ } + vn += verbs[i].len + 1; } - if (i < verbcount) continue; - *errorcodeptr = ERR60; + + if (i < verbcount) continue; /* Successfully handled a verb */ + *errorcodeptr = ERR60; /* Verb not recognized */ goto FAILED; } /* Deal with the extended parentheses; all are introduced by '?', and the appearance of any of them means that this is not a capturing group. */ - else if (*ptr == '?') + else if (*ptr == CHAR_QUESTION_MARK) { int i, set, unset, namelen; int *optset; - const uschar *name; - uschar *slot; + const pcre_uchar *name; + pcre_uchar *slot; switch (*(++ptr)) { - case '#': /* Comment; skip to ket */ + case CHAR_NUMBER_SIGN: /* Comment; skip to ket */ ptr++; - while (*ptr != 0 && *ptr != ')') ptr++; - if (*ptr == 0) + while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++; + if (*ptr == CHAR_NULL) { *errorcodeptr = ERR18; goto FAILED; @@ -4182,20 +5712,21 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* ------------------------------------------------------------ */ - case '|': /* Reset capture count for each branch */ + case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */ reset_bracount = TRUE; /* Fall through */ /* ------------------------------------------------------------ */ - case ':': /* Non-capturing bracket */ + case CHAR_COLON: /* Non-capturing bracket */ bravalue = OP_BRA; ptr++; break; /* ------------------------------------------------------------ */ - case '(': + case CHAR_LEFT_PARENTHESIS: bravalue = OP_COND; /* Conditional group */ + tempptr = ptr; /* A condition can be an assertion, a number (referring to a numbered group), a name (referring to a named group), or 'R', referring to @@ -4208,25 +5739,40 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ be the recursive thing or the name 'R' (and similarly for 'R' followed by digits), and (b) a number could be a name that consists of digits. In both cases, we look for a name first; if not found, we try the other - cases. */ + cases. + + For compatibility with auto-callouts, we allow a callout to be + specified before a condition that is an assertion. First, check for the + syntax of a callout; if found, adjust the temporary pointer that is + used to check for an assertion condition. That's all that is needed! */ + + if (ptr[1] == CHAR_QUESTION_MARK && ptr[2] == CHAR_C) + { + for (i = 3;; i++) if (!IS_DIGIT(ptr[i])) break; + if (ptr[i] == CHAR_RIGHT_PARENTHESIS) + tempptr += i + 1; + } /* For conditions that are assertions, check the syntax, and then exit the switch. This will take control down to where bracketed groups, including assertions, are processed. */ - if (ptr[1] == '?' && (ptr[2] == '=' || ptr[2] == '!' || ptr[2] == '<')) + if (tempptr[1] == CHAR_QUESTION_MARK && + (tempptr[2] == CHAR_EQUALS_SIGN || + tempptr[2] == CHAR_EXCLAMATION_MARK || + tempptr[2] == CHAR_LESS_THAN_SIGN)) break; /* Most other conditions use OP_CREF (a couple change to OP_RREF - below), and all need to skip 3 bytes at the start of the group. */ + below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */ code[1+LINK_SIZE] = OP_CREF; - skipbytes = 3; + skipbytes = 1+IMM2_SIZE; refsign = -1; /* Check for a test for recursion in a named group. */ - if (ptr[1] == 'R' && ptr[2] == '&') + if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND) { terminator = -1; ptr += 2; @@ -4236,25 +5782,25 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* Check for a test for a named group's having been set, using the Perl syntax (?(<name>) or (?('name') */ - else if (ptr[1] == '<') + else if (ptr[1] == CHAR_LESS_THAN_SIGN) { - terminator = '>'; + terminator = CHAR_GREATER_THAN_SIGN; ptr++; } - else if (ptr[1] == '\'') + else if (ptr[1] == CHAR_APOSTROPHE) { - terminator = '\''; + terminator = CHAR_APOSTROPHE; ptr++; } else { - terminator = 0; - if (ptr[1] == '-' || ptr[1] == '+') refsign = *(++ptr); + terminator = CHAR_NULL; + if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr); } /* We now expect to read a name; any thing else is an error */ - if ((cd->ctypes[ptr[1]] & ctype_word) == 0) + if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0) { ptr += 1; /* To get the right offset */ *errorcodeptr = ERR28; @@ -4265,16 +5811,16 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ recno = 0; name = ++ptr; - while ((cd->ctypes[*ptr] & ctype_word) != 0) + while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) { if (recno >= 0) - recno = ((digitab[*ptr] & ctype_digit) != 0)? - recno * 10 + *ptr - '0' : -1; + recno = (IS_DIGIT(*ptr))? recno * 10 + (int)(*ptr - CHAR_0) : -1; ptr++; } - namelen = ptr - name; + namelen = (int)(ptr - name); - if ((terminator > 0 && *ptr++ != terminator) || *ptr++ != ')') + if ((terminator > 0 && *ptr++ != (pcre_uchar)terminator) || + *ptr++ != CHAR_RIGHT_PARENTHESIS) { ptr--; /* Error offset */ *errorcodeptr = ERR26; @@ -4296,7 +5842,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ *errorcodeptr = ERR58; goto FAILED; } - recno = (refsign == '-')? + recno = (refsign == CHAR_MINUS)? cd->bracount - recno + 1 : recno +cd->bracount; if (recno <= 0 || recno > cd->final_bracount) { @@ -4308,12 +5854,15 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } /* Otherwise (did not start with "+" or "-"), start by looking for the - name. */ + name. If we find a name, add one to the opcode to change OP_CREF or + OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same, + except they record that the reference was originally to a name. The + information is used to check duplicate names. */ slot = cd->name_table; for (i = 0; i < cd->names_found; i++) { - if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break; + if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break; slot += cd->name_entry_size; } @@ -4323,23 +5872,25 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ { recno = GET2(slot, 0); PUT2(code, 2+LINK_SIZE, recno); + code[1+LINK_SIZE]++; } /* Search the pattern for a forward reference */ - else if ((i = find_parens(ptr, cd->bracount, name, namelen, - (options & PCRE_EXTENDED) != 0)) > 0) + else if ((i = find_parens(cd, name, namelen, + (options & PCRE_EXTENDED) != 0, utf)) > 0) { PUT2(code, 2+LINK_SIZE, i); + code[1+LINK_SIZE]++; } - /* If terminator == 0 it means that the name followed directly after - the opening parenthesis [e.g. (?(abc)...] and in this case there are - some further alternatives to try. For the cases where terminator != 0 - [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have + /* If terminator == CHAR_NULL it means that the name followed directly + after the opening parenthesis [e.g. (?(abc)...] and in this case there + are some further alternatives to try. For the cases where terminator != + 0 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have now checked all the possibilities, so give an error. */ - else if (terminator != 0) + else if (terminator != CHAR_NULL) { *errorcodeptr = ERR15; goto FAILED; @@ -4348,17 +5899,17 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* Check for (?(R) for recursion. Allow digits after R to specify a specific group number. */ - else if (*name == 'R') + else if (*name == CHAR_R) { recno = 0; for (i = 1; i < namelen; i++) { - if ((digitab[name[i]] & ctype_digit) == 0) + if (!IS_DIGIT(name[i])) { *errorcodeptr = ERR15; goto FAILED; } - recno = recno * 10 + name[i] - '0'; + recno = recno * 10 + name[i] - CHAR_0; } if (recno == 0) recno = RREF_ANY; code[1+LINK_SIZE] = OP_RREF; /* Change test type */ @@ -4368,7 +5919,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* Similarly, check for the (?(DEFINE) "condition", which is always false. */ - else if (namelen == 6 && strncmp((char *)name, "DEFINE", 6) == 0) + else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0) { code[1+LINK_SIZE] = OP_DEF; skipbytes = 1; @@ -4393,41 +5944,46 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* ------------------------------------------------------------ */ - case '=': /* Positive lookahead */ + case CHAR_EQUALS_SIGN: /* Positive lookahead */ bravalue = OP_ASSERT; + cd->assert_depth += 1; ptr++; break; /* ------------------------------------------------------------ */ - case '!': /* Negative lookahead */ + case CHAR_EXCLAMATION_MARK: /* Negative lookahead */ ptr++; - if (*ptr == ')') /* Optimize (?!) */ + if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */ { *code++ = OP_FAIL; previous = NULL; continue; } bravalue = OP_ASSERT_NOT; + cd->assert_depth += 1; break; /* ------------------------------------------------------------ */ - case '<': /* Lookbehind or named define */ + case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */ switch (ptr[1]) { - case '=': /* Positive lookbehind */ + case CHAR_EQUALS_SIGN: /* Positive lookbehind */ bravalue = OP_ASSERTBACK; + cd->assert_depth += 1; ptr += 2; break; - case '!': /* Negative lookbehind */ + case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */ bravalue = OP_ASSERTBACK_NOT; + cd->assert_depth += 1; ptr += 2; break; default: /* Could be name define, else bad */ - if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME; + if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0) + goto DEFINE_NAME; ptr++; /* Correct offset for error */ *errorcodeptr = ERR24; goto FAILED; @@ -4436,22 +5992,23 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* ------------------------------------------------------------ */ - case '>': /* One-time brackets */ + case CHAR_GREATER_THAN_SIGN: /* One-time brackets */ bravalue = OP_ONCE; ptr++; break; /* ------------------------------------------------------------ */ - case 'C': /* Callout - may be followed by digits; */ - previous_callout = code; /* Save for later completion */ - after_manual_callout = 1; /* Skip one item before completing */ + case CHAR_C: /* Callout - may be followed by digits; */ + previous_callout = code; /* Save for later completion */ + after_manual_callout = 1; /* Skip one item before completing */ *code++ = OP_CALLOUT; { int n = 0; - while ((digitab[*(++ptr)] & ctype_digit) != 0) - n = n * 10 + *ptr - '0'; - if (*ptr != ')') + ptr++; + while(IS_DIGIT(*ptr)) + n = n * 10 + *ptr++ - CHAR_0; + if (*ptr != CHAR_RIGHT_PARENTHESIS) { *errorcodeptr = ERR39; goto FAILED; @@ -4462,8 +6019,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ goto FAILED; } *code++ = n; - PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */ - PUT(code, LINK_SIZE, 0); /* Default length */ + PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */ + PUT(code, LINK_SIZE, 0); /* Default length */ code += 2 * LINK_SIZE; } previous = NULL; @@ -4471,14 +6028,15 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* ------------------------------------------------------------ */ - case 'P': /* Python-style named subpattern handling */ - if (*(++ptr) == '=' || *ptr == '>') /* Reference or recursion */ + case CHAR_P: /* Python-style named subpattern handling */ + if (*(++ptr) == CHAR_EQUALS_SIGN || + *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */ { - is_recurse = *ptr == '>'; - terminator = ')'; + is_recurse = *ptr == CHAR_GREATER_THAN_SIGN; + terminator = CHAR_RIGHT_PARENTHESIS; goto NAMED_REF_OR_RECURSE; } - else if (*ptr != '<') /* Test for Python-style definition */ + else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */ { *errorcodeptr = ERR41; goto FAILED; @@ -4488,19 +6046,20 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* ------------------------------------------------------------ */ DEFINE_NAME: /* Come here from (?< handling */ - case '\'': + case CHAR_APOSTROPHE: { - terminator = (*ptr == '<')? '>' : '\''; + terminator = (*ptr == CHAR_LESS_THAN_SIGN)? + CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE; name = ++ptr; - while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++; - namelen = ptr - name; + while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++; + namelen = (int)(ptr - name); /* In the pre-compile phase, just do a syntax check. */ if (lengthptr != NULL) { - if (*ptr != terminator) + if (*ptr != (pcre_uchar)terminator) { *errorcodeptr = ERR42; goto FAILED; @@ -4510,9 +6069,9 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ *errorcodeptr = ERR49; goto FAILED; } - if (namelen + 3 > cd->name_entry_size) + if (namelen + IMM2_SIZE + 1 > cd->name_entry_size) { - cd->name_entry_size = namelen + 3; + cd->name_entry_size = namelen + IMM2_SIZE + 1; if (namelen > MAX_NAME_SIZE) { *errorcodeptr = ERR48; @@ -4521,51 +6080,97 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } } - /* In the real compile, create the entry in the table */ + /* In the real compile, create the entry in the table, maintaining + alphabetical order. Duplicate names for different numbers are + permitted only if PCRE_DUPNAMES is set. Duplicate names for the same + number are always OK. (An existing number can be re-used if (?| + appears in the pattern.) In either event, a duplicate name results in + a duplicate entry in the table, even if the number is the same. This + is because the number of names, and hence the table size, is computed + in the pre-compile, and it affects various numbers and pointers which + would all have to be modified, and the compiled code moved down, if + duplicates with the same number were omitted from the table. This + doesn't seem worth the hassle. However, *different* names for the + same number are not permitted. */ else { + BOOL dupname = FALSE; slot = cd->name_table; + for (i = 0; i < cd->names_found; i++) { - int crc = memcmp(name, slot+2, namelen); + int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen)); if (crc == 0) { - if (slot[2+namelen] == 0) + if (slot[IMM2_SIZE+namelen] == 0) { - if ((options & PCRE_DUPNAMES) == 0) + if (GET2(slot, 0) != cd->bracount + 1 && + (options & PCRE_DUPNAMES) == 0) { *errorcodeptr = ERR43; goto FAILED; } + else dupname = TRUE; } - else crc = -1; /* Current name is substring */ + else crc = -1; /* Current name is a substring */ } + + /* Make space in the table and break the loop for an earlier + name. For a duplicate or later name, carry on. We do this for + duplicates so that in the simple case (when ?(| is not used) they + are in order of their numbers. */ + if (crc < 0) { memmove(slot + cd->name_entry_size, slot, - (cd->names_found - i) * cd->name_entry_size); + IN_UCHARS((cd->names_found - i) * cd->name_entry_size)); break; } + + /* Continue the loop for a later or duplicate name */ + slot += cd->name_entry_size; } + /* For non-duplicate names, check for a duplicate number before + adding the new name. */ + + if (!dupname) + { + pcre_uchar *cslot = cd->name_table; + for (i = 0; i < cd->names_found; i++) + { + if (cslot != slot) + { + if (GET2(cslot, 0) == cd->bracount + 1) + { + *errorcodeptr = ERR65; + goto FAILED; + } + } + else i--; + cslot += cd->name_entry_size; + } + } + PUT2(slot, 0, cd->bracount + 1); - memcpy(slot + 2, name, namelen); - slot[2+namelen] = 0; + memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen)); + slot[IMM2_SIZE + namelen] = 0; } } - /* In both cases, count the number of names we've encountered. */ + /* In both pre-compile and compile, count the number of names we've + encountered. */ - ptr++; /* Move past > or ' */ cd->names_found++; + ptr++; /* Move past > or ' */ goto NUMBERED_GROUP; /* ------------------------------------------------------------ */ - case '&': /* Perl recursion/subroutine syntax */ - terminator = ')'; + case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */ + terminator = CHAR_RIGHT_PARENTHESIS; is_recurse = TRUE; /* Fall through */ @@ -4573,24 +6178,30 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ references (?P=name) and recursion (?P>name), as well as falling through from the Perl recursion syntax (?&name). We also come here from the Perl \k<name> or \k'name' back reference syntax and the \k{name} - .NET syntax. */ + .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */ NAMED_REF_OR_RECURSE: name = ++ptr; - while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++; - namelen = ptr - name; + while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++; + namelen = (int)(ptr - name); - /* In the pre-compile phase, do a syntax check and set a dummy - reference number. */ + /* In the pre-compile phase, do a syntax check. We used to just set + a dummy reference number, because it was not used in the first pass. + However, with the change of recursive back references to be atomic, + we have to look for the number so that this state can be identified, as + otherwise the incorrect length is computed. If it's not a backwards + reference, the dummy number will do. */ if (lengthptr != NULL) { + const pcre_uchar *temp; + if (namelen == 0) { *errorcodeptr = ERR62; goto FAILED; } - if (*ptr != terminator) + if (*ptr != (pcre_uchar)terminator) { *errorcodeptr = ERR42; goto FAILED; @@ -4600,7 +6211,22 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ *errorcodeptr = ERR48; goto FAILED; } - recno = 0; + + /* The name table does not exist in the first pass, so we cannot + do a simple search as in the code below. Instead, we have to scan the + pattern to find the number. It is important that we scan it only as + far as we have got because the syntax of named subpatterns has not + been checked for the rest of the pattern, and find_parens() assumes + correct syntax. In any case, it's a waste of resources to scan + further. We stop the scan at the current point by temporarily + adjusting the value of cd->endpattern. */ + + temp = cd->end_pattern; + cd->end_pattern = ptr; + recno = find_parens(cd, name, namelen, + (options & PCRE_EXTENDED) != 0, utf); + cd->end_pattern = temp; + if (recno < 0) recno = 0; /* Forward ref; set dummy number */ } /* In the real compile, seek the name in the table. We check the name @@ -4613,8 +6239,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ slot = cd->name_table; for (i = 0; i < cd->names_found; i++) { - if (strncmp((char *)name, (char *)slot+2, namelen) == 0 && - slot[2+namelen] == 0) + if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 && + slot[IMM2_SIZE+namelen] == 0) break; slot += cd->name_entry_size; } @@ -4624,8 +6250,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ recno = GET2(slot, 0); } else if ((recno = /* Forward back reference */ - find_parens(ptr, cd->bracount, name, namelen, - (options & PCRE_EXTENDED) != 0)) <= 0) + find_parens(cd, name, namelen, + (options & PCRE_EXTENDED) != 0, utf)) <= 0) { *errorcodeptr = ERR15; goto FAILED; @@ -4640,45 +6266,54 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* ------------------------------------------------------------ */ - case 'R': /* Recursion */ + case CHAR_R: /* Recursion */ ptr++; /* Same as (?0) */ /* Fall through */ /* ------------------------------------------------------------ */ - case '-': case '+': - case '0': case '1': case '2': case '3': case '4': /* Recursion or */ - case '5': case '6': case '7': case '8': case '9': /* subroutine */ + case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */ + case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: + case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9: { - const uschar *called; + const pcre_uchar *called; + terminator = CHAR_RIGHT_PARENTHESIS; + + /* Come here from the \g<...> and \g'...' code (Oniguruma + compatibility). However, the syntax has been checked to ensure that + the ... are a (signed) number, so that neither ERR63 nor ERR29 will + be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY + ever be taken. */ + + HANDLE_NUMERICAL_RECURSION: - if ((refsign = *ptr) == '+') + if ((refsign = *ptr) == CHAR_PLUS) { ptr++; - if ((digitab[*ptr] & ctype_digit) == 0) + if (!IS_DIGIT(*ptr)) { *errorcodeptr = ERR63; goto FAILED; } } - else if (refsign == '-') + else if (refsign == CHAR_MINUS) { - if ((digitab[ptr[1]] & ctype_digit) == 0) + if (!IS_DIGIT(ptr[1])) goto OTHER_CHAR_AFTER_QUERY; ptr++; } recno = 0; - while((digitab[*ptr] & ctype_digit) != 0) - recno = recno * 10 + *ptr++ - '0'; + while(IS_DIGIT(*ptr)) + recno = recno * 10 + *ptr++ - CHAR_0; - if (*ptr != ')') + if (*ptr != (pcre_uchar)terminator) { *errorcodeptr = ERR29; goto FAILED; } - if (refsign == '-') + if (refsign == CHAR_MINUS) { if (recno == 0) { @@ -4692,7 +6327,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ goto FAILED; } } - else if (refsign == '+') + else if (refsign == CHAR_PLUS) { if (recno == 0) { @@ -4719,56 +6354,64 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ if (lengthptr == NULL) { *code = OP_END; - if (recno != 0) called = find_bracket(cd->start_code, utf8, recno); + if (recno != 0) + called = PRIV(find_bracket)(cd->start_code, utf, recno); /* Forward reference */ if (called == NULL) { - if (find_parens(ptr, cd->bracount, NULL, recno, - (options & PCRE_EXTENDED) != 0) < 0) + if (find_parens(cd, NULL, recno, + (options & PCRE_EXTENDED) != 0, utf) < 0) { *errorcodeptr = ERR15; goto FAILED; } + + /* Fudge the value of "called" so that when it is inserted as an + offset below, what it actually inserted is the reference number + of the group. Then remember the forward reference. */ + called = cd->start_code + recno; - PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code); + if (cd->hwm >= cd->start_workspace + cd->workspace_size - + WORK_SIZE_SAFETY_MARGIN) + { + *errorcodeptr = expand_workspace(cd); + if (*errorcodeptr != 0) goto FAILED; + } + PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code)); } /* If not a forward reference, and the subpattern is still open, this is a recursive call. We check to see if this is a left - recursion that could loop for ever, and diagnose that case. */ - - else if (GET(called, 1) == 0 && - could_be_empty(called, code, bcptr, utf8)) + recursion that could loop for ever, and diagnose that case. We + must not, however, do this check if we are in a conditional + subpattern because the condition might be testing for recursion in + a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid. + Forever loops are also detected at runtime, so those that occur in + conditional subpatterns will be picked up then. */ + + else if (GET(called, 1) == 0 && cond_depth <= 0 && + could_be_empty(called, code, bcptr, utf, cd)) { *errorcodeptr = ERR40; goto FAILED; } } - /* Insert the recursion/subroutine item, automatically wrapped inside - "once" brackets. Set up a "previous group" length so that a - subsequent quantifier will work. */ - - *code = OP_ONCE; - PUT(code, 1, 2 + 2*LINK_SIZE); - code += 1 + LINK_SIZE; + /* Insert the recursion/subroutine item. It does not have a set first + character (relevant if it is repeated, because it will then be + wrapped with ONCE brackets). */ *code = OP_RECURSE; - PUT(code, 1, called - cd->start_code); + PUT(code, 1, (int)(called - cd->start_code)); code += 1 + LINK_SIZE; - - *code = OP_KET; - PUT(code, 1, 2 + 2*LINK_SIZE); - code += 1 + LINK_SIZE; - - length_prevgroup = 3 + 3*LINK_SIZE; + groupsetfirstchar = FALSE; } /* Can't determine a first byte now */ - if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; continue; @@ -4778,23 +6421,23 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ set = unset = 0; optset = &set; - while (*ptr != ')' && *ptr != ':') + while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON) { switch (*ptr++) { - case '-': optset = &unset; break; + case CHAR_MINUS: optset = &unset; break; - case 'J': /* Record that it changed in the external options */ + case CHAR_J: /* Record that it changed in the external options */ *optset |= PCRE_DUPNAMES; cd->external_flags |= PCRE_JCHANGED; break; - case 'i': *optset |= PCRE_CASELESS; break; - case 'm': *optset |= PCRE_MULTILINE; break; - case 's': *optset |= PCRE_DOTALL; break; - case 'x': *optset |= PCRE_EXTENDED; break; - case 'U': *optset |= PCRE_UNGREEDY; break; - case 'X': *optset |= PCRE_EXTRA; break; + case CHAR_i: *optset |= PCRE_CASELESS; break; + case CHAR_m: *optset |= PCRE_MULTILINE; break; + case CHAR_s: *optset |= PCRE_DOTALL; break; + case CHAR_x: *optset |= PCRE_EXTENDED; break; + case CHAR_U: *optset |= PCRE_UNGREEDY; break; + case CHAR_X: *optset |= PCRE_EXTRA; break; default: *errorcodeptr = ERR12; ptr--; /* Correct the offset */ @@ -4824,33 +6467,25 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ is necessary to ensure we correctly detect the start of the pattern in both phases. - If we are not at the pattern start, compile code to change the ims - options if this setting actually changes any of them, and reset the - greedy defaults and the case value for firstbyte and reqbyte. */ + If we are not at the pattern start, reset the greedy defaults and the + case value for firstchar and reqchar. */ - if (*ptr == ')') + if (*ptr == CHAR_RIGHT_PARENTHESIS) { if (code == cd->start_code + 1 + LINK_SIZE && (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE)) { cd->external_options = newoptions; } - else + else { - if ((options & PCRE_IMS) != (newoptions & PCRE_IMS)) - { - *code++ = OP_OPT; - *code++ = newoptions & PCRE_IMS; - } greedy_default = ((newoptions & PCRE_UNGREEDY) != 0); greedy_non_default = greedy_default ^ 1; - req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0; + req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0; } /* Change options at this level, and pass them back for use - in subsequent branches. When not at the start of the pattern, this - information is also necessary so that a resetting item can be - compiled at the end of a group (if we are in a group). */ + in subsequent branches. */ *optionsptr = options = newoptions; previous = NULL; /* This item can't be repeated */ @@ -4867,8 +6502,8 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } /* End of switch for character following (? */ } /* End of (? handling */ - /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set, - all unadorned brackets become non-capturing and behave like (?:...) + /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE + is set, all unadorned brackets become non-capturing and behave like (?:...) brackets. */ else if ((options & PCRE_NO_AUTO_CAPTURE) != 0) @@ -4883,53 +6518,64 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ NUMBERED_GROUP: cd->bracount += 1; PUT2(code, 1+LINK_SIZE, cd->bracount); - skipbytes = 2; + skipbytes = IMM2_SIZE; } - /* Process nested bracketed regex. Assertions may not be repeated, but - other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a - non-register variable in order to be able to pass its address because some - compilers complain otherwise. Pass in a new setting for the ims options if - they have changed. */ + /* Process nested bracketed regex. Assertions used not to be repeatable, + but this was changed for Perl compatibility, so all kinds can now be + repeated. We copy code into a non-register variable (tempcode) in order to + be able to pass its address because some compilers complain otherwise. */ - previous = (bravalue >= OP_ONCE)? code : NULL; + previous = code; /* For handling repetition */ *code = bravalue; tempcode = code; - tempreqvary = cd->req_varyopt; /* Save value before bracket */ - length_prevgroup = 0; /* Initialize for pre-compile phase */ + tempreqvary = cd->req_varyopt; /* Save value before bracket */ + tempbracount = cd->bracount; /* Save value before bracket */ + length_prevgroup = 0; /* Initialize for pre-compile phase */ if (!compile_regex( - newoptions, /* The complete new option state */ - options & PCRE_IMS, /* The previous ims option state */ - &tempcode, /* Where to put code (updated) */ - &ptr, /* Input pointer (updated) */ - errorcodeptr, /* Where to put an error message */ + newoptions, /* The complete new option state */ + &tempcode, /* Where to put code (updated) */ + &ptr, /* Input pointer (updated) */ + errorcodeptr, /* Where to put an error message */ (bravalue == OP_ASSERTBACK || bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */ - reset_bracount, /* True if (?| group */ - skipbytes, /* Skip over bracket number */ - &subfirstbyte, /* For possible first char */ - &subreqbyte, /* For possible last char */ - bcptr, /* Current branch chain */ - cd, /* Tables block */ - (lengthptr == NULL)? NULL : /* Actual compile phase */ - &length_prevgroup /* Pre-compile phase */ + reset_bracount, /* True if (?| group */ + skipbytes, /* Skip over bracket number */ + cond_depth + + ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */ + &subfirstchar, /* For possible first char */ + &subfirstcharflags, + &subreqchar, /* For possible last char */ + &subreqcharflags, + bcptr, /* Current branch chain */ + cd, /* Tables block */ + (lengthptr == NULL)? NULL : /* Actual compile phase */ + &length_prevgroup /* Pre-compile phase */ )) goto FAILED; + /* If this was an atomic group and there are no capturing groups within it, + generate OP_ONCE_NC instead of OP_ONCE. */ + + if (bravalue == OP_ONCE && cd->bracount <= tempbracount) + *code = OP_ONCE_NC; + + if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT) + cd->assert_depth -= 1; + /* At the end of compiling, code is still pointing to the start of the - group, while tempcode has been updated to point past the end of the group - and any option resetting that may follow it. The pattern pointer (ptr) - is on the bracket. */ + group, while tempcode has been updated to point past the end of the group. + The pattern pointer (ptr) is on the bracket. - /* If this is a conditional bracket, check that there are no more than + If this is a conditional bracket, check that there are no more than two branches in the group, or just one if it's a DEFINE group. We do this in the real compile phase, not in the pre-pass, where the whole group may not be available. */ if (bravalue == OP_COND && lengthptr == NULL) { - uschar *tc = code; + pcre_uchar *tc = code; int condcount = 0; do { @@ -4952,7 +6598,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ } /* A "normal" conditional group. If there is just one branch, we must not - make use of its firstbyte or reqbyte, because this is equivalent to an + make use of its firstchar or reqchar, because this is equivalent to an empty second branch. */ else @@ -4962,13 +6608,13 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ *errorcodeptr = ERR27; goto FAILED; } - if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE; + if (condcount == 1) subfirstcharflags = subreqcharflags = REQ_NONE; } } /* Error if hit end of pattern */ - if (*ptr != ')') + if (*ptr != CHAR_RIGHT_PARENTHESIS) { *errorcodeptr = ERR14; goto FAILED; @@ -4987,7 +6633,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ goto FAILED; } *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE; - *code++ = OP_BRA; + code++; /* This already contains bravalue */ PUTINC(code, 0, 1 + LINK_SIZE); *code++ = OP_KET; PUTINC(code, 0, 1 + LINK_SIZE); @@ -5006,131 +6652,233 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* Handle updating of the required and first characters for other types of group. Update for normal brackets of all kinds, and conditions with two branches (see code above). If the bracket is followed by a quantifier with - zero repeat, we have to back off. Hence the definition of zeroreqbyte and - zerofirstbyte outside the main loop so that they can be accessed for the + zero repeat, we have to back off. Hence the definition of zeroreqchar and + zerofirstchar outside the main loop so that they can be accessed for the back off. */ - zeroreqbyte = reqbyte; - zerofirstbyte = firstbyte; - groupsetfirstbyte = FALSE; + zeroreqchar = reqchar; + zeroreqcharflags = reqcharflags; + zerofirstchar = firstchar; + zerofirstcharflags = firstcharflags; + groupsetfirstchar = FALSE; if (bravalue >= OP_ONCE) { - /* If we have not yet set a firstbyte in this branch, take it from the + /* If we have not yet set a firstchar in this branch, take it from the subpattern, remembering that it was set here so that a repeat of more - than one can replicate it as reqbyte if necessary. If the subpattern has - no firstbyte, set "none" for the whole branch. In both cases, a zero - repeat forces firstbyte to "none". */ + than one can replicate it as reqchar if necessary. If the subpattern has + no firstchar, set "none" for the whole branch. In both cases, a zero + repeat forces firstchar to "none". */ - if (firstbyte == REQ_UNSET) + if (firstcharflags == REQ_UNSET) { - if (subfirstbyte >= 0) + if (subfirstcharflags >= 0) { - firstbyte = subfirstbyte; - groupsetfirstbyte = TRUE; + firstchar = subfirstchar; + firstcharflags = subfirstcharflags; + groupsetfirstchar = TRUE; } - else firstbyte = REQ_NONE; - zerofirstbyte = REQ_NONE; + else firstcharflags = REQ_NONE; + zerofirstcharflags = REQ_NONE; } - /* If firstbyte was previously set, convert the subpattern's firstbyte - into reqbyte if there wasn't one, using the vary flag that was in + /* If firstchar was previously set, convert the subpattern's firstchar + into reqchar if there wasn't one, using the vary flag that was in existence beforehand. */ - else if (subfirstbyte >= 0 && subreqbyte < 0) - subreqbyte = subfirstbyte | tempreqvary; + else if (subfirstcharflags >= 0 && subreqcharflags < 0) + { + subreqchar = subfirstchar; + subreqcharflags = subfirstcharflags | tempreqvary; + } /* If the subpattern set a required byte (or set a first byte that isn't really the first byte - see above), set it. */ - if (subreqbyte >= 0) reqbyte = subreqbyte; + if (subreqcharflags >= 0) + { + reqchar = subreqchar; + reqcharflags = subreqcharflags; + } } - /* For a forward assertion, we take the reqbyte, if set. This can be + /* For a forward assertion, we take the reqchar, if set. This can be helpful if the pattern that follows the assertion doesn't set a different - char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte + char. For example, it's useful for /(?=abcde).+/. We can't set firstchar for an assertion, however because it leads to incorrect effect for patterns - such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead - of a firstbyte. This is overcome by a scan at the end if there's no - firstbyte, looking for an asserted first char. */ + such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead + of a firstchar. This is overcome by a scan at the end if there's no + firstchar, looking for an asserted first char. */ - else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte; + else if (bravalue == OP_ASSERT && subreqcharflags >= 0) + { + reqchar = subreqchar; + reqcharflags = subreqcharflags; + } break; /* End of processing '(' */ /* ===================================================================*/ /* Handle metasequences introduced by \. For ones like \d, the ESC_ values - are arranged to be the negation of the corresponding OP_values. For the - back references, the values are ESC_REF plus the reference number. Only - back references and those types that consume a character may be repeated. - We can test for values between ESC_b and ESC_Z for the latter; this may - have to change if any new ones are ever created. */ - - case '\\': + are arranged to be the negation of the corresponding OP_values in the + default case when PCRE_UCP is not set. For the back references, the values + are negative the reference number. Only back references and those types + that consume a character may be repeated. We can test for values between + ESC_b and ESC_Z for the latter; this may have to change if any new ones are + ever created. */ + + case CHAR_BACKSLASH: tempptr = ptr; - c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE); + escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, FALSE); if (*errorcodeptr != 0) goto FAILED; - if (c < 0) + if (escape == 0) /* The escape coded a single character */ + c = ec; + else { - if (-c == ESC_Q) /* Handle start of quoted string */ + if (escape == ESC_Q) /* Handle start of quoted string */ { - if (ptr[1] == '\\' && ptr[2] == 'E') ptr += 2; /* avoid empty string */ - else inescq = TRUE; + if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E) + ptr += 2; /* avoid empty string */ + else inescq = TRUE; continue; } - if (-c == ESC_E) continue; /* Perl ignores an orphan \E */ + if (escape == ESC_E) continue; /* Perl ignores an orphan \E */ /* For metasequences that actually match a character, we disable the setting of a first character if it hasn't already been set. */ - if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z) - firstbyte = REQ_NONE; + if (firstcharflags == REQ_UNSET && escape > ESC_b && escape < ESC_Z) + firstcharflags = REQ_NONE; /* Set values to reset to if this is followed by a zero repeat. */ - zerofirstbyte = firstbyte; - zeroreqbyte = reqbyte; + zerofirstchar = firstchar; + zerofirstcharflags = firstcharflags; + zeroreqchar = reqchar; + zeroreqcharflags = reqcharflags; + + /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n' + is a subroutine call by number (Oniguruma syntax). In fact, the value + ESC_g is returned only for these cases. So we don't need to check for < + or ' if the value is ESC_g. For the Perl syntax \g{n} the value is + -n, and for the Perl syntax \g{name} the result is ESC_k (as + that is a synonym for a named back reference). */ + + if (escape == ESC_g) + { + const pcre_uchar *p; + save_hwm = cd->hwm; /* Normally this is set when '(' is read */ + terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)? + CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE; + + /* These two statements stop the compiler for warning about possibly + unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In + fact, because we actually check for a number below, the paths that + would actually be in error are never taken. */ + + skipbytes = 0; + reset_bracount = FALSE; + + /* Test for a name */ + + if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS) + { + BOOL is_a_number = TRUE; + for (p = ptr + 1; *p != CHAR_NULL && *p != (pcre_uchar)terminator; p++) + { + if (!MAX_255(*p)) { is_a_number = FALSE; break; } + if ((cd->ctypes[*p] & ctype_digit) == 0) is_a_number = FALSE; + if ((cd->ctypes[*p] & ctype_word) == 0) break; + } + if (*p != (pcre_uchar)terminator) + { + *errorcodeptr = ERR57; + break; + } + if (is_a_number) + { + ptr++; + goto HANDLE_NUMERICAL_RECURSION; + } + is_recurse = TRUE; + goto NAMED_REF_OR_RECURSE; + } + + /* Test a signed number in angle brackets or quotes. */ + + p = ptr + 2; + while (IS_DIGIT(*p)) p++; + if (*p != (pcre_uchar)terminator) + { + *errorcodeptr = ERR57; + break; + } + ptr++; + goto HANDLE_NUMERICAL_RECURSION; + } /* \k<name> or \k'name' is a back reference by name (Perl syntax). - We also support \k{name} (.NET syntax) */ + We also support \k{name} (.NET syntax). */ - if (-c == ESC_k && (ptr[1] == '<' || ptr[1] == '\'' || ptr[1] == '{')) + if (escape == ESC_k) { + if ((ptr[1] != CHAR_LESS_THAN_SIGN && + ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET)) + { + *errorcodeptr = ERR69; + break; + } is_recurse = FALSE; - terminator = (*(++ptr) == '<')? '>' : (*ptr == '\'')? '\'' : '}'; + terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)? + CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)? + CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET; goto NAMED_REF_OR_RECURSE; } - /* Back references are handled specially; must disable firstbyte if + /* Back references are handled specially; must disable firstchar if not set to cope with cases like (?=(\w+))\1: which would otherwise set ':' later. */ - if (-c >= ESC_REF) + if (escape < 0) { - recno = -c - ESC_REF; + open_capitem *oc; + recno = -escape; HANDLE_REFERENCE: /* Come here from named backref handling */ - if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; + if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE; previous = code; - *code++ = OP_REF; + *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF; PUT2INC(code, 0, recno); cd->backref_map |= (recno < 32)? (1 << recno) : 1; if (recno > cd->top_backref) cd->top_backref = recno; + + /* Check to see if this back reference is recursive, that it, it + is inside the group that it references. A flag is set so that the + group can be made atomic. */ + + for (oc = cd->open_caps; oc != NULL; oc = oc->next) + { + if (oc->number == recno) + { + oc->flag = TRUE; + break; + } + } } /* So are Unicode property matches, if supported. */ #ifdef SUPPORT_UCP - else if (-c == ESC_P || -c == ESC_p) + else if (escape == ESC_P || escape == ESC_p) { BOOL negated; - int pdata; - int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr); - if (ptype < 0) goto FAILED; + unsigned int ptype = 0, pdata = 0; + if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr)) + goto FAILED; previous = code; - *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP; + *code++ = ((escape == ESC_p) != negated)? OP_PROP : OP_NOTPROP; *code++ = ptype; *code++ = pdata; } @@ -5139,7 +6887,7 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ /* If Unicode properties are not supported, \X, \P, and \p are not allowed. */ - else if (-c == ESC_X || -c == ESC_P || -c == ESC_p) + else if (escape == ESC_X || escape == ESC_P || escape == ESC_p) { *errorcodeptr = ERR45; goto FAILED; @@ -5147,12 +6895,31 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ #endif /* For the rest (including \X when Unicode properties are supported), we - can obtain the OP value by negating the escape value. */ + can obtain the OP value by negating the escape value in the default + situation when PCRE_UCP is not set. When it *is* set, we substitute + Unicode property tests. Note that \b and \B do a one-character + lookbehind, and \A also behaves as if it does. */ else { - previous = (-c > ESC_b && -c < ESC_Z)? code : NULL; - *code++ = -c; + if ((escape == ESC_b || escape == ESC_B || escape == ESC_A) && + cd->max_lookbehind == 0) + cd->max_lookbehind = 1; +#ifdef SUPPORT_UCP + if (escape >= ESC_DU && escape <= ESC_wu) + { + nestptr = ptr + 1; /* Where to resume */ + ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */ + } + else +#endif + /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE + so that it works in DFA mode and in lookbehinds. */ + + { + previous = (escape > ESC_b && escape < ESC_Z)? code : NULL; + *code++ = (!utf && escape == ESC_C)? OP_ALLANY : escape; + } } continue; } @@ -5161,9 +6928,9 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ a value > 127. We set its representation in the length/buffer, and then handle it as a data character. */ -#ifdef SUPPORT_UTF8 - if (utf8 && c > 127) - mclength = _erts_pcre_ord2utf8(c, mcbuffer); +#if defined SUPPORT_UTF && !defined COMPILE_PCRE32 + if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR) + mclength = PRIV(ord2utf)(c, mcbuffer); else #endif @@ -5184,12 +6951,9 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ mclength = 1; mcbuffer[0] = c; -#ifdef SUPPORT_UTF8 - if (utf8 && c >= 0xc0) - { - while ((ptr[1] & 0xc0) == 0x80) - mcbuffer[mclength++] = *(++ptr); - } +#ifdef SUPPORT_UTF + if (utf && HAS_EXTRALEN(c)) + ACROSSCHAR(TRUE, ptr[1], mcbuffer[mclength++] = *(++ptr)); #endif /* At this point we have the character's bytes in mcbuffer, and the length @@ -5197,44 +6961,79 @@ we set the flag only if there is a literal "\r" or "\n" in the class. */ ONE_CHAR: previous = code; - *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR; + + /* For caseless UTF-8 mode when UCP support is available, check whether + this character has more than one other case. If so, generate a special + OP_PROP item instead of OP_CHARI. */ + +#ifdef SUPPORT_UCP + if (utf && (options & PCRE_CASELESS) != 0) + { + GETCHAR(c, mcbuffer); + if ((c = UCD_CASESET(c)) != 0) + { + *code++ = OP_PROP; + *code++ = PT_CLIST; + *code++ = c; + if (firstcharflags == REQ_UNSET) firstcharflags = zerofirstcharflags = REQ_NONE; + break; + } + } +#endif + + /* Caseful matches, or not one of the multicase characters. */ + + *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR; for (c = 0; c < mclength; c++) *code++ = mcbuffer[c]; /* Remember if \r or \n were seen */ - if (mcbuffer[0] == '\r' || mcbuffer[0] == '\n') + if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF; /* Set the first and required bytes appropriately. If no previous first byte, set it from this character, but revert to none on a zero repeat. - Otherwise, leave the firstbyte value alone, and don't change it on a zero + Otherwise, leave the firstchar value alone, and don't change it on a zero repeat. */ - if (firstbyte == REQ_UNSET) + if (firstcharflags == REQ_UNSET) { - zerofirstbyte = REQ_NONE; - zeroreqbyte = reqbyte; + zerofirstcharflags = REQ_NONE; + zeroreqchar = reqchar; + zeroreqcharflags = reqcharflags; - /* If the character is more than one byte long, we can set firstbyte + /* If the character is more than one byte long, we can set firstchar only if it is not to be matched caselessly. */ if (mclength == 1 || req_caseopt == 0) { - firstbyte = mcbuffer[0] | req_caseopt; - if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt; + firstchar = mcbuffer[0] | req_caseopt; + firstchar = mcbuffer[0]; + firstcharflags = req_caseopt; + + if (mclength != 1) + { + reqchar = code[-1]; + reqcharflags = cd->req_varyopt; + } } - else firstbyte = reqbyte = REQ_NONE; + else firstcharflags = reqcharflags = REQ_NONE; } - /* firstbyte was previously set; we can set reqbyte only the length is + /* firstchar was previously set; we can set reqchar only if the length is 1 or the matching is caseful. */ else { - zerofirstbyte = firstbyte; - zeroreqbyte = reqbyte; + zerofirstchar = firstchar; + zerofirstcharflags = firstcharflags; + zeroreqchar = reqchar; + zeroreqcharflags = reqcharflags; if (mclength == 1 || req_caseopt == 0) - reqbyte = code[-1] | req_caseopt | cd->req_varyopt; + { + reqchar = code[-1]; + reqcharflags = req_caseopt | cd->req_varyopt; + } } break; /* End of literal character handling */ @@ -5253,7 +7052,6 @@ return FALSE; - /************************************************* * Compile sequence of alternatives * *************************************************/ @@ -5261,26 +7059,23 @@ return FALSE; /* On entry, ptr is pointing past the bracket character, but on return it points to the closing bracket, or vertical bar, or end of string. The code variable is pointing at the byte into which the BRA operator has been stored. -If the ims options are changed at the start (for a (?ims: group) or during any -branch, we need to insert an OP_OPT item at the start of every following branch -to ensure they get set correctly at run time, and also pass the new options -into every subsequent branch compile. - This function is used during the pre-compile phase when we are trying to find out the amount of memory needed, as well as during the real compile phase. The value of lengthptr distinguishes the two phases. Arguments: options option bits, including any changes for this subpattern - oldims previous settings of ims option bits codeptr -> the address of the current code pointer ptrptr -> the address of the current pattern pointer errorcodeptr -> pointer to error code variable lookbehind TRUE if this is a lookbehind assertion reset_bracount TRUE to reset the count for each branch skipbytes skip this many bytes at start (for brackets and OP_COND) - firstbyteptr place to put the first required character, or a negative number - reqbyteptr place to put the last required character, or a negative number + cond_depth depth of nesting for conditional subpatterns + firstcharptr place to put the first required character + firstcharflagsptr place to put the first character flags, or a negative number + reqcharptr place to put the last required character + reqcharflagsptr place to put the last required character flags, or a negative number bcptr pointer to the chain of currently open branches cd points to the data block with tables pointers etc. lengthptr NULL during the real compile phase @@ -5290,27 +7085,34 @@ Returns: TRUE on success */ static BOOL -compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr, +compile_regex(int options, pcre_uchar **codeptr, const pcre_uchar **ptrptr, int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes, - int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd, - int *lengthptr) + int cond_depth, + pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr, + pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr, + branch_chain *bcptr, compile_data *cd, int *lengthptr) { -const uschar *ptr = *ptrptr; -uschar *code = *codeptr; -uschar *last_branch = code; -uschar *start_bracket = code; -uschar *reverse_count = NULL; -int firstbyte, reqbyte; -int branchfirstbyte, branchreqbyte; +const pcre_uchar *ptr = *ptrptr; +pcre_uchar *code = *codeptr; +pcre_uchar *last_branch = code; +pcre_uchar *start_bracket = code; +pcre_uchar *reverse_count = NULL; +open_capitem capitem; +int capnumber = 0; +pcre_uint32 firstchar, reqchar; +pcre_int32 firstcharflags, reqcharflags; +pcre_uint32 branchfirstchar, branchreqchar; +pcre_int32 branchfirstcharflags, branchreqcharflags; int length; -int orig_bracount; -int max_bracount; +unsigned int orig_bracount; +unsigned int max_bracount; branch_chain bc; bc.outer = bcptr; -bc.current = code; +bc.current_branch = code; -firstbyte = reqbyte = REQ_UNSET; +firstchar = reqchar = 0; +firstcharflags = reqcharflags = REQ_UNSET; /* Accumulate the length for use in the pre-compile phase. Start with the length of the BRA and KET and any extra bytes that are required at the @@ -5326,6 +7128,21 @@ the code that abstracts option settings at the start of the pattern and makes them global. It tests the value of length for (2 + 2*LINK_SIZE) in the pre-compile phase to find out whether anything has yet been compiled or not. */ +/* If this is a capturing subpattern, add to the chain of open capturing items +so that we can detect them if (*ACCEPT) is encountered. This is also used to +detect groups that contain recursive back references to themselves. Note that +only OP_CBRA need be tested here; changing this opcode to one of its variants, +e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */ + +if (*code == OP_CBRA) + { + capnumber = GET2(code, 1 + LINK_SIZE); + capitem.number = capnumber; + capitem.next = cd->open_caps; + capitem.flag = FALSE; + cd->open_caps = &capitem; + } + /* Offset is set zero to mark that this bracket is still open */ PUT(code, 1, 0); @@ -5341,15 +7158,6 @@ for (;;) if (reset_bracount) cd->bracount = orig_bracount; - /* Handle a change of ims options at the start of the branch */ - - if ((options & PCRE_IMS) != oldims) - { - *code++ = OP_OPT; - *code++ = options & PCRE_IMS; - length += 2; - } - /* Set up dummy OP_REVERSE if lookbehind assertion */ if (lookbehind) @@ -5363,8 +7171,9 @@ for (;;) /* Now compile the branch; in the pre-compile phase its length gets added into the length. */ - if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte, - &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length)) + if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstchar, + &branchfirstcharflags, &branchreqchar, &branchreqcharflags, &bc, + cond_depth, cd, (lengthptr == NULL)? NULL : &length)) { *ptrptr = ptr; return FALSE; @@ -5379,62 +7188,92 @@ for (;;) if (lengthptr == NULL) { - /* If this is the first branch, the firstbyte and reqbyte values for the + /* If this is the first branch, the firstchar and reqchar values for the branch become the values for the regex. */ if (*last_branch != OP_ALT) { - firstbyte = branchfirstbyte; - reqbyte = branchreqbyte; + firstchar = branchfirstchar; + firstcharflags = branchfirstcharflags; + reqchar = branchreqchar; + reqcharflags = branchreqcharflags; } - /* If this is not the first branch, the first char and reqbyte have to + /* If this is not the first branch, the first char and reqchar have to match the values from all the previous branches, except that if the - previous value for reqbyte didn't have REQ_VARY set, it can still match, + previous value for reqchar didn't have REQ_VARY set, it can still match, and we set REQ_VARY for the regex. */ else { - /* If we previously had a firstbyte, but it doesn't match the new branch, - we have to abandon the firstbyte for the regex, but if there was - previously no reqbyte, it takes on the value of the old firstbyte. */ + /* If we previously had a firstchar, but it doesn't match the new branch, + we have to abandon the firstchar for the regex, but if there was + previously no reqchar, it takes on the value of the old firstchar. */ - if (firstbyte >= 0 && firstbyte != branchfirstbyte) + if (firstcharflags >= 0 && + (firstcharflags != branchfirstcharflags || firstchar != branchfirstchar)) { - if (reqbyte < 0) reqbyte = firstbyte; - firstbyte = REQ_NONE; + if (reqcharflags < 0) + { + reqchar = firstchar; + reqcharflags = firstcharflags; + } + firstcharflags = REQ_NONE; } - /* If we (now or from before) have no firstbyte, a firstbyte from the - branch becomes a reqbyte if there isn't a branch reqbyte. */ + /* If we (now or from before) have no firstchar, a firstchar from the + branch becomes a reqchar if there isn't a branch reqchar. */ - if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0) - branchreqbyte = branchfirstbyte; + if (firstcharflags < 0 && branchfirstcharflags >= 0 && branchreqcharflags < 0) + { + branchreqchar = branchfirstchar; + branchreqcharflags = branchfirstcharflags; + } - /* Now ensure that the reqbytes match */ + /* Now ensure that the reqchars match */ - if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY)) - reqbyte = REQ_NONE; - else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */ + if (((reqcharflags & ~REQ_VARY) != (branchreqcharflags & ~REQ_VARY)) || + reqchar != branchreqchar) + reqcharflags = REQ_NONE; + else + { + reqchar = branchreqchar; + reqcharflags |= branchreqcharflags; /* To "or" REQ_VARY */ + } } /* If lookbehind, check that this branch matches a fixed-length string, and put the length into the OP_REVERSE item. Temporarily mark the end of the - branch with OP_END. */ + branch with OP_END. If the branch contains OP_RECURSE, the result is -3 + because there may be forward references that we can't check here. Set a + flag to cause another lookbehind check at the end. Why not do it all at the + end? Because common, erroneous checks are picked up here and the offset of + the problem can be shown. */ if (lookbehind) { int fixed_length; *code = OP_END; - fixed_length = find_fixedlength(last_branch, options); + fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0, + FALSE, cd); DPRINTF(("fixed length = %d\n", fixed_length)); - if (fixed_length < 0) + if (fixed_length == -3) + { + cd->check_lookbehind = TRUE; + } + else if (fixed_length < 0) { - *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25; + *errorcodeptr = (fixed_length == -2)? ERR36 : + (fixed_length == -4)? ERR70: ERR25; *ptrptr = ptr; return FALSE; } - PUT(reverse_count, 0, fixed_length); + else + { + if (fixed_length > cd->max_lookbehind) + cd->max_lookbehind = fixed_length; + PUT(reverse_count, 0, fixed_length); + } } } @@ -5443,15 +7282,13 @@ for (;;) of offsets, with the field in the BRA item now becoming an offset to the first alternative. If there are no alternatives, it points to the end of the group. The length in the terminating ket is always the length of the whole - bracketed item. If any of the ims options were changed inside the group, - compile a resetting op-code following, except at the very end of the pattern. - Return leaving the pointer at the terminating char. */ + bracketed item. Return leaving the pointer at the terminating char. */ - if (*ptr != '|') + if (*ptr != CHAR_VERTICAL_LINE) { if (lengthptr == NULL) { - int branch_length = code - last_branch; + int branch_length = (int)(code - last_branch); do { int prev_length = GET(last_branch, 1); @@ -5465,16 +7302,28 @@ for (;;) /* Fill in the ket */ *code = OP_KET; - PUT(code, 1, code - start_bracket); + PUT(code, 1, (int)(code - start_bracket)); code += 1 + LINK_SIZE; - /* Resetting option if needed */ + /* If it was a capturing subpattern, check to see if it contained any + recursive back references. If so, we must wrap it in atomic brackets. + In any event, remove the block from the chain. */ - if ((options & PCRE_IMS) != oldims && *ptr == ')') + if (capnumber > 0) { - *code++ = OP_OPT; - *code++ = oldims; - length += 2; + if (cd->open_caps->flag) + { + memmove(start_bracket + 1 + LINK_SIZE, start_bracket, + IN_UCHARS(code - start_bracket)); + *start_bracket = OP_ONCE; + code += 1 + LINK_SIZE; + PUT(start_bracket, 1, (int)(code - start_bracket)); + *code = OP_KET; + PUT(code, 1, (int)(code - start_bracket)); + code += 1 + LINK_SIZE; + length += 2 + 2*LINK_SIZE; + } + cd->open_caps = cd->open_caps->next; } /* Retain the highest bracket number, in case resetting was used. */ @@ -5485,8 +7334,10 @@ for (;;) *codeptr = code; *ptrptr = ptr; - *firstbyteptr = firstbyte; - *reqbyteptr = reqbyte; + *firstcharptr = firstchar; + *firstcharflagsptr = firstcharflags; + *reqcharptr = reqchar; + *reqcharflagsptr = reqcharflags; if (lengthptr != NULL) { if (OFLOW_MAX - *lengthptr < length) @@ -5516,8 +7367,8 @@ for (;;) else { *code = OP_ALT; - PUT(code, 1, code - last_branch); - bc.current = last_branch = code; + PUT(code, 1, (int)(code - last_branch)); + bc.current_branch = last_branch = code; code += 1 + LINK_SIZE; } @@ -5536,8 +7387,8 @@ for (;;) /* Try to find out if this is an anchored regular expression. Consider each alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then -it's anchored. However, if this is a multiline pattern, then only OP_SOD -counts, since OP_CIRC can match in the middle. +it's anchored. However, if this is a multiline pattern, then only OP_SOD will +be found, because ^ generates OP_CIRCM in that mode. We can also consider a regex to be anchored if OP_SOM starts all its branches. This is the code for \G, which means "match at start of match position, taking @@ -5556,64 +7407,78 @@ and the highest back reference was greater than or equal to that level. However, by keeping a bitmap of the first 31 back references, we can catch some of the more common cases more precisely. +... A second exception is when the .* appears inside an atomic group, because +this prevents the number of characters it matches from being adjusted. + Arguments: code points to start of expression (the bracket) - options points to the options setting bracket_map a bitmap of which brackets we are inside while testing; this handles up to substring 31; after that we just have to take the less precise approach - backref_map the back reference bitmap + cd points to the compile data block + atomcount atomic group level Returns: TRUE or FALSE */ static BOOL -is_anchored(register const uschar *code, int *options, unsigned int bracket_map, - unsigned int backref_map) +is_anchored(register const pcre_uchar *code, unsigned int bracket_map, + compile_data *cd, int atomcount) { do { - const uschar *scode = first_significant_code(code + _erts_pcre_OP_lengths[*code], - options, PCRE_MULTILINE, FALSE); + const pcre_uchar *scode = first_significant_code( + code + PRIV(OP_lengths)[*code], FALSE); register int op = *scode; /* Non-capturing brackets */ - if (op == OP_BRA) + if (op == OP_BRA || op == OP_BRAPOS || + op == OP_SBRA || op == OP_SBRAPOS) { - if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE; + if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE; } /* Capturing brackets */ - else if (op == OP_CBRA) + else if (op == OP_CBRA || op == OP_CBRAPOS || + op == OP_SCBRA || op == OP_SCBRAPOS) { int n = GET2(scode, 1+LINK_SIZE); int new_map = bracket_map | ((n < 32)? (1 << n) : 1); - if (!is_anchored(scode, options, new_map, backref_map)) return FALSE; + if (!is_anchored(scode, new_map, cd, atomcount)) return FALSE; } - /* Other brackets */ + /* Positive forward assertions and conditions */ - else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND) + else if (op == OP_ASSERT || op == OP_COND) { - if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE; + if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE; } - /* .* is not anchored unless DOTALL is set and it isn't in brackets that - are or may be referenced. */ + /* Atomic groups */ + + else if (op == OP_ONCE || op == OP_ONCE_NC) + { + if (!is_anchored(scode, bracket_map, cd, atomcount + 1)) + return FALSE; + } + + /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and + it isn't in brackets that are or may be referenced or inside an atomic + group. */ else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR || - op == OP_TYPEPOSSTAR) && - (*options & PCRE_DOTALL) != 0) + op == OP_TYPEPOSSTAR)) { - if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE; + if (scode[1] != OP_ALLANY || (bracket_map & cd->backref_map) != 0 || + atomcount > 0 || cd->had_pruneorskip) + return FALSE; } /* Check for explicit anchoring */ - else if (op != OP_SOD && op != OP_SOM && - ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC)) - return FALSE; + else if (op != OP_SOD && op != OP_SOM && op != OP_CIRC) return FALSE; + code += GET(code, 1); } while (*code == OP_ALT); /* Loop for each alternative */ @@ -5631,59 +7496,109 @@ return TRUE; matching and for non-DOTALL patterns that start with .* (which must start at the beginning or after \n). As in the case of is_anchored() (see above), we have to take account of back references to capturing brackets that contain .* -because in that case we can't make the assumption. +because in that case we can't make the assumption. Also, the appearance of .* +inside atomic brackets or in a pattern that contains *PRUNE or *SKIP does not +count, because once again the assumption no longer holds. Arguments: code points to start of expression (the bracket) bracket_map a bitmap of which brackets we are inside while testing; this handles up to substring 31; after that we just have to take the less precise approach - backref_map the back reference bitmap + cd points to the compile data + atomcount atomic group level Returns: TRUE or FALSE */ static BOOL -is_startline(const uschar *code, unsigned int bracket_map, - unsigned int backref_map) +is_startline(const pcre_uchar *code, unsigned int bracket_map, + compile_data *cd, int atomcount) { do { - const uschar *scode = first_significant_code(code + _erts_pcre_OP_lengths[*code], - NULL, 0, FALSE); + const pcre_uchar *scode = first_significant_code( + code + PRIV(OP_lengths)[*code], FALSE); register int op = *scode; + /* If we are at the start of a conditional assertion group, *both* the + conditional assertion *and* what follows the condition must satisfy the test + for start of line. Other kinds of condition fail. Note that there may be an + auto-callout at the start of a condition. */ + + if (op == OP_COND) + { + scode += 1 + LINK_SIZE; + if (*scode == OP_CALLOUT) scode += PRIV(OP_lengths)[OP_CALLOUT]; + switch (*scode) + { + case OP_CREF: + case OP_NCREF: + case OP_RREF: + case OP_NRREF: + case OP_DEF: + return FALSE; + + default: /* Assertion */ + if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE; + do scode += GET(scode, 1); while (*scode == OP_ALT); + scode += 1 + LINK_SIZE; + break; + } + scode = first_significant_code(scode, FALSE); + op = *scode; + } + /* Non-capturing brackets */ - if (op == OP_BRA) + if (op == OP_BRA || op == OP_BRAPOS || + op == OP_SBRA || op == OP_SBRAPOS) { - if (!is_startline(scode, bracket_map, backref_map)) return FALSE; + if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE; } /* Capturing brackets */ - else if (op == OP_CBRA) + else if (op == OP_CBRA || op == OP_CBRAPOS || + op == OP_SCBRA || op == OP_SCBRAPOS) { int n = GET2(scode, 1+LINK_SIZE); int new_map = bracket_map | ((n < 32)? (1 << n) : 1); - if (!is_startline(scode, new_map, backref_map)) return FALSE; + if (!is_startline(scode, new_map, cd, atomcount)) return FALSE; + } + + /* Positive forward assertions */ + + else if (op == OP_ASSERT) + { + if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE; } - /* Other brackets */ + /* Atomic brackets */ - else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND) - { if (!is_startline(scode, bracket_map, backref_map)) return FALSE; } + else if (op == OP_ONCE || op == OP_ONCE_NC) + { + if (!is_startline(scode, bracket_map, cd, atomcount + 1)) return FALSE; + } - /* .* means "start at start or after \n" if it isn't in brackets that - may be referenced. */ + /* .* means "start at start or after \n" if it isn't in atomic brackets or + brackets that may be referenced, as long as the pattern does not contain + *PRUNE or *SKIP, because these break the feature. Consider, for example, + /.*?a(*PRUNE)b/ with the subject "aab", which matches "ab", i.e. not at the + start of a line. */ else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR) { - if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE; + if (scode[1] != OP_ANY || (bracket_map & cd->backref_map) != 0 || + atomcount > 0 || cd->had_pruneorskip) + return FALSE; } - /* Check for explicit circumflex */ + /* Check for explicit circumflex; anything else gives a FALSE result. Note + in particular that this includes atomic brackets OP_ONCE and OP_ONCE_NC + because the number of characters matched by .* cannot be adjusted inside + them. */ - else if (op != OP_CIRC) return FALSE; + else if (op != OP_CIRC && op != OP_CIRCM) return FALSE; /* Move on to the next alternative */ @@ -5709,58 +7624,82 @@ we return that char, otherwise -1. Arguments: code points to start of expression (the bracket) - options pointer to the options (used to check casing changes) + flags points to the first char flags, or to REQ_NONE inassert TRUE if in an assertion -Returns: -1 or the fixed first char +Returns: the fixed first char, or 0 with REQ_NONE in flags */ -static int -find_firstassertedchar(const uschar *code, int *options, BOOL inassert) +static pcre_uint32 +find_firstassertedchar(const pcre_uchar *code, pcre_int32 *flags, + BOOL inassert) { -register int c = -1; +register pcre_uint32 c = 0; +int cflags = REQ_NONE; + +*flags = REQ_NONE; do { - int d; - const uschar *scode = - first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE); - register int op = *scode; + pcre_uint32 d; + int dflags; + int xl = (*code == OP_CBRA || *code == OP_SCBRA || + *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? IMM2_SIZE:0; + const pcre_uchar *scode = first_significant_code(code + 1+LINK_SIZE + xl, + TRUE); + register pcre_uchar op = *scode; switch(op) { default: - return -1; + return 0; case OP_BRA: + case OP_BRAPOS: case OP_CBRA: + case OP_SCBRA: + case OP_CBRAPOS: + case OP_SCBRAPOS: case OP_ASSERT: case OP_ONCE: + case OP_ONCE_NC: case OP_COND: - if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0) - return -1; - if (c < 0) c = d; else if (c != d) return -1; + d = find_firstassertedchar(scode, &dflags, op == OP_ASSERT); + if (dflags < 0) + return 0; + if (cflags < 0) { c = d; cflags = dflags; } else if (c != d || cflags != dflags) return 0; break; - case OP_EXACT: /* Fall through */ - scode += 2; + case OP_EXACT: + scode += IMM2_SIZE; + /* Fall through */ case OP_CHAR: - case OP_CHARNC: case OP_PLUS: case OP_MINPLUS: case OP_POSPLUS: - if (!inassert) return -1; - if (c < 0) - { - c = scode[1]; - if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS; - } - else if (c != scode[1]) return -1; + if (!inassert) return 0; + if (cflags < 0) { c = scode[1]; cflags = 0; } + else if (c != scode[1]) return 0; + break; + + case OP_EXACTI: + scode += IMM2_SIZE; + /* Fall through */ + + case OP_CHARI: + case OP_PLUSI: + case OP_MINPLUSI: + case OP_POSPLUSI: + if (!inassert) return 0; + if (cflags < 0) { c = scode[1]; cflags = REQ_CASELESS; } + else if (c != scode[1]) return 0; break; } code += GET(code, 1); } while (*code == OP_ALT); + +*flags = cflags; return c; } @@ -5778,7 +7717,7 @@ compatibility. The new function is given a new name. Arguments: pattern the regular expression options various option bits - errorcodeptr pointer to error code variable (erts_pcre_compile2() only) + errorcodeptr pointer to error code variable (pcre_compile2() only) can be NULL if you don't want a code value errorptr pointer to pointer to error text erroroffset ptr offset in pattern where error was detected @@ -5788,30 +7727,59 @@ Returns: pointer to compiled data block, or NULL on error, with errorptr and erroroffset set */ -PCRE_EXP_DEFN pcre * -erts_pcre_compile(const char *pattern, int options, const char **errorptr, +#if defined COMPILE_PCRE8 +PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION +pcre_compile(const char *pattern, int options, const char **errorptr, + int *erroroffset, const unsigned char *tables) +#elif defined COMPILE_PCRE16 +PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION +pcre16_compile(PCRE_SPTR16 pattern, int options, const char **errorptr, int *erroroffset, const unsigned char *tables) +#elif defined COMPILE_PCRE32 +PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION +pcre32_compile(PCRE_SPTR32 pattern, int options, const char **errorptr, + int *erroroffset, const unsigned char *tables) +#endif { -return erts_pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables); +#if defined COMPILE_PCRE8 +return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables); +#elif defined COMPILE_PCRE16 +return pcre16_compile2(pattern, options, NULL, errorptr, erroroffset, tables); +#elif defined COMPILE_PCRE32 +return pcre32_compile2(pattern, options, NULL, errorptr, erroroffset, tables); +#endif } -PCRE_EXP_DEFN pcre * -erts_pcre_compile2(const char *pattern, int options, int *errorcodeptr, +#if defined COMPILE_PCRE8 +PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION +pcre_compile2(const char *pattern, int options, int *errorcodeptr, + const char **errorptr, int *erroroffset, const unsigned char *tables) +#elif defined COMPILE_PCRE16 +PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION +pcre16_compile2(PCRE_SPTR16 pattern, int options, int *errorcodeptr, + const char **errorptr, int *erroroffset, const unsigned char *tables) +#elif defined COMPILE_PCRE32 +PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION +pcre32_compile2(PCRE_SPTR32 pattern, int options, int *errorcodeptr, const char **errorptr, int *erroroffset, const unsigned char *tables) +#endif { -real_pcre *re; +REAL_PCRE *re; int length = 1; /* For final END opcode */ -int firstbyte, reqbyte, newline; +pcre_int32 firstcharflags, reqcharflags; +pcre_uint32 firstchar, reqchar; +pcre_uint32 limit_match = PCRE_UINT32_MAX; +pcre_uint32 limit_recursion = PCRE_UINT32_MAX; +int newline; int errorcode = 0; int skipatstart = 0; -#ifdef SUPPORT_UTF8 -BOOL utf8; -#endif +BOOL utf; +BOOL never_utf = FALSE; size_t size; -uschar *code; -const uschar *codestart; -const uschar *ptr; +pcre_uchar *code; +const pcre_uchar *codestart; +const pcre_uchar *ptr; compile_data compile_block; compile_data *cd = &compile_block; @@ -5819,13 +7787,14 @@ compile_data *cd = &compile_block; computing the amount of memory that is needed. Compiled items are thrown away as soon as possible, so that a fairly large buffer should be sufficient for this purpose. The same space is used in the second phase for remembering where -to fill in forward references to subpatterns. */ +to fill in forward references to subpatterns. That may overflow, in which case +new memory is obtained from malloc(). */ -uschar cworkspace[COMPILE_WORK_SIZE]; +pcre_uchar cworkspace[COMPILE_WORK_SIZE]; /* Set this early so that early errors get offset 0. */ -ptr = (const uschar *)pattern; +ptr = (const pcre_uchar *)pattern; /* We can't pass back an error message if errorptr is NULL; I guess the best we can do is just return NULL, but we can set a code value if there is a code @@ -5850,60 +7819,113 @@ if (erroroffset == NULL) *erroroffset = 0; -/* Can't support UTF8 unless PCRE has been compiled to include the code. */ +/* Set up pointers to the individual character tables */ -#ifdef SUPPORT_UTF8 -utf8 = (options & PCRE_UTF8) != 0; -if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 && - (*erroroffset = _erts_pcre_valid_utf8((uschar *)pattern, -1)) >= 0) - { - errorcode = ERR44; - goto PCRE_EARLY_ERROR_RETURN2; - } -#else -if ((options & PCRE_UTF8) != 0) - { - errorcode = ERR32; - goto PCRE_EARLY_ERROR_RETURN; - } -#endif +if (tables == NULL) tables = PRIV(default_tables); +cd->lcc = tables + lcc_offset; +cd->fcc = tables + fcc_offset; +cd->cbits = tables + cbits_offset; +cd->ctypes = tables + ctypes_offset; + +/* Check that all undefined public option bits are zero */ -if ((options & ~PUBLIC_OPTIONS) != 0) +if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0) { errorcode = ERR17; goto PCRE_EARLY_ERROR_RETURN; } -/* Set up pointers to the individual character tables */ +/* If PCRE_NEVER_UTF is set, remember it. */ -if (tables == NULL) tables = _erts_pcre_default_tables; -cd->lcc = tables + lcc_offset; -cd->fcc = tables + fcc_offset; -cd->cbits = tables + cbits_offset; -cd->ctypes = tables + ctypes_offset; +if ((options & PCRE_NEVER_UTF) != 0) never_utf = TRUE; /* Check for global one-time settings at the start of the pattern, and remember the offset for later. */ -while (ptr[skipatstart] == '(' && ptr[skipatstart+1] == '*') +cd->external_flags = 0; /* Initialize here for LIMIT_MATCH/RECURSION */ + +while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS && + ptr[skipatstart+1] == CHAR_ASTERISK) { int newnl = 0; int newbsr = 0; - if (strncmp((char *)(ptr+skipatstart+2), "CR)", 3) == 0) +/* For completeness and backward compatibility, (*UTFn) is supported in the +relevant libraries, but (*UTF) is generic and always supported. Note that +PCRE_UTF8 == PCRE_UTF16 == PCRE_UTF32. */ + +#ifdef COMPILE_PCRE8 + if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF8_RIGHTPAR, 5) == 0) + { skipatstart += 7; options |= PCRE_UTF8; continue; } +#endif +#ifdef COMPILE_PCRE16 + if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF16_RIGHTPAR, 6) == 0) + { skipatstart += 8; options |= PCRE_UTF16; continue; } +#endif +#ifdef COMPILE_PCRE32 + if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF32_RIGHTPAR, 6) == 0) + { skipatstart += 8; options |= PCRE_UTF32; continue; } +#endif + + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF_RIGHTPAR, 4) == 0) + { skipatstart += 6; options |= PCRE_UTF8; continue; } + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UCP_RIGHTPAR, 4) == 0) + { skipatstart += 6; options |= PCRE_UCP; continue; } + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_START_OPT_RIGHTPAR, 13) == 0) + { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; } + + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_MATCH_EQ, 12) == 0) + { + pcre_uint32 c = 0; + int p = skipatstart + 14; + while (isdigit(ptr[p])) + { + if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow */ + c = c*10 + ptr[p++] - CHAR_0; + } + if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break; + if (c < limit_match) + { + limit_match = c; + cd->external_flags |= PCRE_MLSET; + } + skipatstart = p; + continue; + } + + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_RECURSION_EQ, 16) == 0) + { + pcre_uint32 c = 0; + int p = skipatstart + 18; + while (isdigit(ptr[p])) + { + if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow check */ + c = c*10 + ptr[p++] - CHAR_0; + } + if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break; + if (c < limit_recursion) + { + limit_recursion = c; + cd->external_flags |= PCRE_RLSET; + } + skipatstart = p; + continue; + } + + if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CR_RIGHTPAR, 3) == 0) { skipatstart += 5; newnl = PCRE_NEWLINE_CR; } - else if (strncmp((char *)(ptr+skipatstart+2), "LF)", 3) == 0) + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LF_RIGHTPAR, 3) == 0) { skipatstart += 5; newnl = PCRE_NEWLINE_LF; } - else if (strncmp((char *)(ptr+skipatstart+2), "CRLF)", 5) == 0) + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CRLF_RIGHTPAR, 5) == 0) { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; } - else if (strncmp((char *)(ptr+skipatstart+2), "ANY)", 4) == 0) + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANY_RIGHTPAR, 4) == 0) { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; } - else if (strncmp((char *)(ptr+skipatstart+2), "ANYCRLF)", 8) == 0) + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANYCRLF_RIGHTPAR, 8) == 0) { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; } - else if (strncmp((char *)(ptr+skipatstart+2), "BSR_ANYCRLF)", 12) == 0) + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0) { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; } - else if (strncmp((char *)(ptr+skipatstart+2), "BSR_UNICODE)", 12) == 0) + else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_UNICODE_RIGHTPAR, 12) == 0) { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; } if (newnl != 0) @@ -5913,15 +7935,57 @@ while (ptr[skipatstart] == '(' && ptr[skipatstart+1] == '*') else break; } +/* PCRE_UTF(16|32) have the same value as PCRE_UTF8. */ +utf = (options & PCRE_UTF8) != 0; +if (utf && never_utf) + { + errorcode = ERR78; + goto PCRE_EARLY_ERROR_RETURN2; + } + +/* Can't support UTF unless PCRE has been compiled to include the code. The +return of an error code from PRIV(valid_utf)() is a new feature, introduced in +release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is +not used here. */ + +#ifdef SUPPORT_UTF +if (utf && (options & PCRE_NO_UTF8_CHECK) == 0 && + (errorcode = PRIV(valid_utf)((PCRE_PUCHAR)pattern, -1, erroroffset)) != 0) + { +#if defined COMPILE_PCRE8 + errorcode = ERR44; +#elif defined COMPILE_PCRE16 + errorcode = ERR74; +#elif defined COMPILE_PCRE32 + errorcode = ERR77; +#endif + goto PCRE_EARLY_ERROR_RETURN2; + } +#else +if (utf) + { + errorcode = ERR32; + goto PCRE_EARLY_ERROR_RETURN; + } +#endif + +/* Can't support UCP unless PCRE has been compiled to include the code. */ + +#ifndef SUPPORT_UCP +if ((options & PCRE_UCP) != 0) + { + errorcode = ERR67; + goto PCRE_EARLY_ERROR_RETURN; + } +#endif + /* Check validity of \R options. */ -switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) +if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) == + (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) { - case 0: - case PCRE_BSR_ANYCRLF: - case PCRE_BSR_UNICODE: - break; - default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN; + errorcode = ERR56; + goto PCRE_EARLY_ERROR_RETURN; } /* Handle different types of newline. The three bits give seven cases. The @@ -5931,10 +7995,10 @@ current code allows for fixed one- or two-byte sequences, plus "any" and switch (options & PCRE_NEWLINE_BITS) { case 0: newline = NEWLINE; break; /* Build-time default */ - case PCRE_NEWLINE_CR: newline = '\r'; break; - case PCRE_NEWLINE_LF: newline = '\n'; break; + case PCRE_NEWLINE_CR: newline = CHAR_CR; break; + case PCRE_NEWLINE_LF: newline = CHAR_NL; break; case PCRE_NEWLINE_CR+ - PCRE_NEWLINE_LF: newline = ('\r' << 8) | '\n'; break; + PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break; case PCRE_NEWLINE_ANY: newline = -1; break; case PCRE_NEWLINE_ANYCRLF: newline = -2; break; default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN; @@ -5974,7 +8038,10 @@ cd->backref_map = 0; /* Reflect pattern for debugging output */ DPRINTF(("------------------------------------------------------------------\n")); -DPRINTF(("%s\n", pattern)); +#ifdef PCRE_DEBUG +print_puchar(stdout, (PCRE_PUCHAR)pattern); +#endif +DPRINTF(("\n")); /* Pretend to compile the pattern while actually just accumulating the length of memory required. This behaviour is triggered by passing a non-NULL final @@ -5987,14 +8054,17 @@ cd->bracount = cd->final_bracount = 0; cd->names_found = 0; cd->name_entry_size = 0; cd->name_table = NULL; -cd->start_workspace = cworkspace; cd->start_code = cworkspace; cd->hwm = cworkspace; -cd->start_pattern = (const uschar *)pattern; -cd->end_pattern = (const uschar *)(pattern + strlen(pattern)); +cd->start_workspace = cworkspace; +cd->workspace_size = COMPILE_WORK_SIZE; +cd->start_pattern = (const pcre_uchar *)pattern; +cd->end_pattern = (const pcre_uchar *)(pattern + STRLEN_UC((const pcre_uchar *)pattern)); cd->req_varyopt = 0; +cd->assert_depth = 0; +cd->max_lookbehind = 0; cd->external_options = options; -cd->external_flags = 0; +cd->open_caps = NULL; /* Now do the pre-compile. On error, errorcode will be set non-zero, so we don't need to look at the result of the function here. The initial options have @@ -6005,13 +8075,13 @@ outside can help speed up starting point checks. */ ptr += skipatstart; code = cworkspace; *code = OP_BRA; -(void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS, - &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, - &length); +(void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE, + FALSE, 0, 0, &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL, + cd, &length); if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN; DPRINTF(("end pre-compile: length=%d workspace=%d\n", length, - cd->hwm - cworkspace)); + (int)(cd->hwm - cworkspace))); if (length > MAX_PATTERN_SIZE) { @@ -6024,8 +8094,8 @@ externally provided function. Integer overflow should no longer be possible because nowadays we limit the maximum value of cd->names_found and cd->name_entry_size. */ -size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3); -re = (real_pcre *)(erts_pcre_malloc)(size); +size = sizeof(REAL_PCRE) + (length + cd->names_found * cd->name_entry_size) * sizeof(pcre_uchar); +re = (REAL_PCRE *)(PUBL(malloc))(size); if (re == NULL) { @@ -6040,18 +8110,24 @@ regex compiled on a system with 4-byte pointers is run on another with 8-byte pointers. */ re->magic_number = MAGIC_NUMBER; -re->size = size; +re->size = (int)size; re->options = cd->external_options; re->flags = cd->external_flags; -re->dummy1 = 0; -re->first_byte = 0; -re->req_byte = 0; -re->name_table_offset = sizeof(real_pcre); +re->limit_match = limit_match; +re->limit_recursion = limit_recursion; +re->first_char = 0; +re->req_char = 0; +re->name_table_offset = sizeof(REAL_PCRE) / sizeof(pcre_uchar); re->name_entry_size = cd->name_entry_size; re->name_count = cd->names_found; re->ref_count = 0; -re->tables = (tables == _erts_pcre_default_tables)? NULL : tables; +re->tables = (tables == PRIV(default_tables))? NULL : tables; re->nullpad = NULL; +#ifdef COMPILE_PCRE32 +re->dummy = 0; +#else +re->dummy1 = re->dummy2 = re->dummy3 = 0; +#endif /* The starting points of the name/number translation table and of the code are passed around in the compile data block. The start/end pattern and initial @@ -6061,69 +8137,144 @@ field; this time it's used for remembering forward references to subpatterns. */ cd->final_bracount = cd->bracount; /* Save for checking forward references */ +cd->assert_depth = 0; cd->bracount = 0; +cd->max_lookbehind = 0; cd->names_found = 0; -cd->name_table = (uschar *)re + re->name_table_offset; +cd->name_table = (pcre_uchar *)re + re->name_table_offset; codestart = cd->name_table + re->name_entry_size * re->name_count; cd->start_code = codestart; -cd->hwm = cworkspace; +cd->hwm = (pcre_uchar *)(cd->start_workspace); cd->req_varyopt = 0; cd->had_accept = FALSE; +cd->had_pruneorskip = FALSE; +cd->check_lookbehind = FALSE; +cd->open_caps = NULL; /* Set up a starting, non-extracting bracket, then compile the expression. On error, errorcode will be set non-zero, so we don't need to look at the result of the function here. */ -ptr = (const uschar *)pattern + skipatstart; -code = (uschar *)codestart; +ptr = (const pcre_uchar *)pattern + skipatstart; +code = (pcre_uchar *)codestart; *code = OP_BRA; -(void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr, - &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL); +(void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0, + &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL, cd, NULL); re->top_bracket = cd->bracount; re->top_backref = cd->top_backref; -re->flags = cd->external_flags; +re->max_lookbehind = cd->max_lookbehind; +re->flags = cd->external_flags | PCRE_MODE; -if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */ +if (cd->had_accept) + { + reqchar = 0; /* Must disable after (*ACCEPT) */ + reqcharflags = REQ_NONE; + } /* If not reached end of pattern on success, there's an excess bracket. */ -if (errorcode == 0 && *ptr != 0) errorcode = ERR22; +if (errorcode == 0 && *ptr != CHAR_NULL) errorcode = ERR22; /* Fill in the terminating state and check for disastrous overflow, but if debugging, leave the test till after things are printed out. */ *code++ = OP_END; -#ifndef DEBUG +#ifndef PCRE_DEBUG if (code - codestart > length) errorcode = ERR23; #endif -/* Fill in any forward references that are required. */ +#ifdef SUPPORT_VALGRIND +/* If the estimated length exceeds the really used length, mark the extra +allocated memory as unaddressable, so that any out-of-bound reads can be +detected. */ +VALGRIND_MAKE_MEM_NOACCESS(code, (length - (code - codestart)) * sizeof(pcre_uchar)); +#endif + +/* Fill in any forward references that are required. There may be repeated +references; optimize for them, as searching a large regex takes time. */ -while (errorcode == 0 && cd->hwm > cworkspace) +if (cd->hwm > cd->start_workspace) { - int offset, recno; - const uschar *groupptr; - cd->hwm -= LINK_SIZE; - offset = GET(cd->hwm, 0); - recno = GET(codestart, offset); - groupptr = find_bracket(codestart, (re->options & PCRE_UTF8) != 0, recno); - if (groupptr == NULL) errorcode = ERR53; - else PUT(((uschar *)codestart), offset, groupptr - codestart); + int prev_recno = -1; + const pcre_uchar *groupptr = NULL; + while (errorcode == 0 && cd->hwm > cd->start_workspace) + { + int offset, recno; + cd->hwm -= LINK_SIZE; + offset = GET(cd->hwm, 0); + recno = GET(codestart, offset); + if (recno != prev_recno) + { + groupptr = PRIV(find_bracket)(codestart, utf, recno); + prev_recno = recno; + } + if (groupptr == NULL) errorcode = ERR53; + else PUT(((pcre_uchar *)codestart), offset, (int)(groupptr - codestart)); + } } +/* If the workspace had to be expanded, free the new memory. */ + +if (cd->workspace_size > COMPILE_WORK_SIZE) + (PUBL(free))((void *)cd->start_workspace); + /* Give an error if there's back reference to a non-existent capturing subpattern. */ if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15; +/* If there were any lookbehind assertions that contained OP_RECURSE +(recursions or subroutine calls), a flag is set for them to be checked here, +because they may contain forward references. Actual recursions cannot be fixed +length, but subroutine calls can. It is done like this so that those without +OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The +exceptional ones forgo this. We scan the pattern to check that they are fixed +length, and set their lengths. */ + +if (cd->check_lookbehind) + { + pcre_uchar *cc = (pcre_uchar *)codestart; + + /* Loop, searching for OP_REVERSE items, and process those that do not have + their length set. (Actually, it will also re-process any that have a length + of zero, but that is a pathological case, and it does no harm.) When we find + one, we temporarily terminate the branch it is in while we scan it. */ + + for (cc = (pcre_uchar *)PRIV(find_bracket)(codestart, utf, -1); + cc != NULL; + cc = (pcre_uchar *)PRIV(find_bracket)(cc, utf, -1)) + { + if (GET(cc, 1) == 0) + { + int fixed_length; + pcre_uchar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE); + int end_op = *be; + *be = OP_END; + fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE, + cd); + *be = end_op; + DPRINTF(("fixed length = %d\n", fixed_length)); + if (fixed_length < 0) + { + errorcode = (fixed_length == -2)? ERR36 : + (fixed_length == -4)? ERR70 : ERR25; + break; + } + if (fixed_length > cd->max_lookbehind) cd->max_lookbehind = fixed_length; + PUT(cc, 1, fixed_length); + } + cc += 1 + LINK_SIZE; + } + } + /* Failed to compile, or error while post-processing */ if (errorcode != 0) { - (erts_pcre_free)(re); + (PUBL(free))(re); PCRE_EARLY_ERROR_RETURN: - *erroroffset = ptr - (const uschar *)pattern; + *erroroffset = (int)(ptr - (const pcre_uchar *)pattern); PCRE_EARLY_ERROR_RETURN2: *errorptr = find_error_text(errorcode); if (errorcodeptr != NULL) *errorcodeptr = errorcode; @@ -6131,33 +8282,57 @@ if (errorcode != 0) } /* If the anchored option was not passed, set the flag if we can determine that -the pattern is anchored by virtue of ^ characters or \A or anything else (such -as starting with .* when DOTALL is set). +the pattern is anchored by virtue of ^ characters or \A or anything else, such +as starting with non-atomic .* when DOTALL is set and there are no occurrences +of *PRUNE or *SKIP. Otherwise, if we know what the first byte has to be, save it, because that speeds up unanchored matches no end. If not, see if we can set the PCRE_STARTLINE flag. This is helpful for multiline matches when all branches -start with ^. and also when all branches start with .* for non-DOTALL matches. -*/ +start with ^. and also when all branches start with non-atomic .* for +non-DOTALL matches when *PRUNE and SKIP are not present. */ if ((re->options & PCRE_ANCHORED) == 0) { - int temp_options = re->options; /* May get changed during these scans */ - if (is_anchored(codestart, &temp_options, 0, cd->backref_map)) - re->options |= PCRE_ANCHORED; + if (is_anchored(codestart, 0, cd, 0)) re->options |= PCRE_ANCHORED; else { - if (firstbyte < 0) - firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE); - if (firstbyte >= 0) /* Remove caseless flag for non-caseable chars */ + if (firstcharflags < 0) + firstchar = find_firstassertedchar(codestart, &firstcharflags, FALSE); + if (firstcharflags >= 0) /* Remove caseless flag for non-caseable chars */ { - int ch = firstbyte & 255; - re->first_byte = ((firstbyte & REQ_CASELESS) != 0 && - cd->fcc[ch] == ch)? ch : firstbyte; +#if defined COMPILE_PCRE8 + re->first_char = firstchar & 0xff; +#elif defined COMPILE_PCRE16 + re->first_char = firstchar & 0xffff; +#elif defined COMPILE_PCRE32 + re->first_char = firstchar; +#endif + if ((firstcharflags & REQ_CASELESS) != 0) + { +#if defined SUPPORT_UCP && !(defined COMPILE_PCRE8) + /* We ignore non-ASCII first chars in 8 bit mode. */ + if (utf) + { + if (re->first_char < 128) + { + if (cd->fcc[re->first_char] != re->first_char) + re->flags |= PCRE_FCH_CASELESS; + } + else if (UCD_OTHERCASE(re->first_char) != re->first_char) + re->flags |= PCRE_FCH_CASELESS; + } + else +#endif + if (MAX_255(re->first_char) + && cd->fcc[re->first_char] != re->first_char) + re->flags |= PCRE_FCH_CASELESS; + } + re->flags |= PCRE_FIRSTSET; } - else if (is_startline(codestart, 0, cd->backref_map)) - re->flags |= PCRE_STARTLINE; + + else if (is_startline(codestart, 0, cd, 0)) re->flags |= PCRE_STARTLINE; } } @@ -6165,20 +8340,43 @@ if ((re->options & PCRE_ANCHORED) == 0) variable length item in the regex. Remove the caseless flag for non-caseable bytes. */ -if (reqbyte >= 0 && - ((re->options & PCRE_ANCHORED) == 0 || (reqbyte & REQ_VARY) != 0)) +if (reqcharflags >= 0 && + ((re->options & PCRE_ANCHORED) == 0 || (reqcharflags & REQ_VARY) != 0)) { - int ch = reqbyte & 255; - re->req_byte = ((reqbyte & REQ_CASELESS) != 0 && - cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte; +#if defined COMPILE_PCRE8 + re->req_char = reqchar & 0xff; +#elif defined COMPILE_PCRE16 + re->req_char = reqchar & 0xffff; +#elif defined COMPILE_PCRE32 + re->req_char = reqchar; +#endif + if ((reqcharflags & REQ_CASELESS) != 0) + { +#if defined SUPPORT_UCP && !(defined COMPILE_PCRE8) + /* We ignore non-ASCII first chars in 8 bit mode. */ + if (utf) + { + if (re->req_char < 128) + { + if (cd->fcc[re->req_char] != re->req_char) + re->flags |= PCRE_RCH_CASELESS; + } + else if (UCD_OTHERCASE(re->req_char) != re->req_char) + re->flags |= PCRE_RCH_CASELESS; + } + else +#endif + if (MAX_255(re->req_char) && cd->fcc[re->req_char] != re->req_char) + re->flags |= PCRE_RCH_CASELESS; + } + re->flags |= PCRE_REQCHSET; } /* Print out the compiled data if debugging is enabled. This is never the case when building a production library. */ -#ifdef DEBUG - +#ifdef PCRE_DEBUG printf("Length = %d top_bracket = %d top_backref = %d\n", length, re->top_bracket, re->top_backref); @@ -6186,38 +8384,50 @@ printf("Options=%08x\n", re->options); if ((re->flags & PCRE_FIRSTSET) != 0) { - int ch = re->first_byte & 255; - const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)? - "" : " (caseless)"; - if (isprint(ch)) printf("First char = %c%s\n", ch, caseless); + pcre_uchar ch = re->first_char; + const char *caseless = + ((re->flags & PCRE_FCH_CASELESS) == 0)? "" : " (caseless)"; + if (PRINTABLE(ch)) printf("First char = %c%s\n", ch, caseless); else printf("First char = \\x%02x%s\n", ch, caseless); } if ((re->flags & PCRE_REQCHSET) != 0) { - int ch = re->req_byte & 255; - const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)? - "" : " (caseless)"; - if (isprint(ch)) printf("Req char = %c%s\n", ch, caseless); + pcre_uchar ch = re->req_char; + const char *caseless = + ((re->flags & PCRE_RCH_CASELESS) == 0)? "" : " (caseless)"; + if (PRINTABLE(ch)) printf("Req char = %c%s\n", ch, caseless); else printf("Req char = \\x%02x%s\n", ch, caseless); } -pcre_printint(re, stdout, TRUE); +#if defined COMPILE_PCRE8 +pcre_printint((pcre *)re, stdout, TRUE); +#elif defined COMPILE_PCRE16 +pcre16_printint((pcre *)re, stdout, TRUE); +#elif defined COMPILE_PCRE32 +pcre32_printint((pcre *)re, stdout, TRUE); +#endif /* This check is done here in the debugging case so that the code that was compiled can be seen. */ if (code - codestart > length) { - (erts_pcre_free)(re); + (PUBL(free))(re); *errorptr = find_error_text(ERR23); - *erroroffset = ptr - (uschar *)pattern; + *erroroffset = ptr - (pcre_uchar *)pattern; if (errorcodeptr != NULL) *errorcodeptr = ERR23; return NULL; } -#endif /* DEBUG */ +#endif /* PCRE_DEBUG */ +#if defined COMPILE_PCRE8 return (pcre *)re; +#elif defined COMPILE_PCRE16 +return (pcre16 *)re; +#elif defined COMPILE_PCRE32 +return (pcre32 *)re; +#endif } /* End of pcre_compile.c */ |