aboutsummaryrefslogtreecommitdiffstats
path: root/lib/stdlib/test/erl_scan_SUITE.erl
diff options
context:
space:
mode:
authorErlang/OTP <[email protected]>2009-11-20 14:54:40 +0000
committerErlang/OTP <[email protected]>2009-11-20 14:54:40 +0000
commit84adefa331c4159d432d22840663c38f155cd4c1 (patch)
treebff9a9c66adda4df2106dfd0e5c053ab182a12bd /lib/stdlib/test/erl_scan_SUITE.erl
downloadotp-84adefa331c4159d432d22840663c38f155cd4c1.tar.gz
otp-84adefa331c4159d432d22840663c38f155cd4c1.tar.bz2
otp-84adefa331c4159d432d22840663c38f155cd4c1.zip
The R13B03 release.OTP_R13B03
Diffstat (limited to 'lib/stdlib/test/erl_scan_SUITE.erl')
-rw-r--r--lib/stdlib/test/erl_scan_SUITE.erl1214
1 files changed, 1214 insertions, 0 deletions
diff --git a/lib/stdlib/test/erl_scan_SUITE.erl b/lib/stdlib/test/erl_scan_SUITE.erl
new file mode 100644
index 0000000000..32a06d15c7
--- /dev/null
+++ b/lib/stdlib/test/erl_scan_SUITE.erl
@@ -0,0 +1,1214 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 1998-2009. All Rights Reserved.
+%%
+%% The contents of this file are subject to the Erlang Public License,
+%% Version 1.1, (the "License"); you may not use this file except in
+%% compliance with the License. You should have received a copy of the
+%% Erlang Public License along with this software. If not, it can be
+%% retrieved online at http://www.erlang.org/.
+%%
+%% Software distributed under the License is distributed on an "AS IS"
+%% basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
+%% the License for the specific language governing rights and limitations
+%% under the License.
+%%
+%% %CopyrightEnd%
+
+-module(erl_scan_SUITE).
+-export([all/1]).
+
+-export([error/1, error_1/1, error_2/1, iso88591/1, otp_7810/1]).
+
+-import(lists, [nth/2,flatten/1]).
+-import(io_lib, [print/1]).
+
+%%
+%% Define to run outside of test server
+%%
+%-define(STANDALONE,1).
+
+-ifdef(STANDALONE).
+-compile(export_all).
+-define(line, put(line, ?LINE), ).
+-define(config(A,B),config(A,B)).
+-define(t, test_server).
+%% config(priv_dir, _) ->
+%% ".";
+%% config(data_dir, _) ->
+%% ".".
+-else.
+-include("test_server.hrl").
+-export([init_per_testcase/2, fin_per_testcase/2]).
+
+init_per_testcase(_Case, Config) when is_list(Config) ->
+ ?line Dog=test_server:timetrap(test_server:seconds(1200)),
+ [{watchdog, Dog}|Config].
+
+fin_per_testcase(_Case, Config) ->
+ Dog=?config(watchdog, Config),
+ test_server:timetrap_cancel(Dog),
+ ok.
+-endif.
+
+% Default timetrap timeout (set in init_per_testcase).
+-define(default_timeout, ?t:minutes(1)).
+
+all(doc) ->
+ ["Test cases for the 'erl_scan' module."];
+all(suite) ->
+ [error,iso88591,otp_7810].
+
+error(doc) ->
+ ["Error cases"];
+error(suite) ->
+ [error_1, error_2].
+
+error_1(doc) ->
+ ["(OTP-2347)"];
+error_1(suite) ->
+ [];
+error_1(Config) when is_list(Config) ->
+ ?line {error, _, _} = erl_scan:string("'a"),
+ ok.
+
+error_2(doc) ->
+ ["Checks that format_error works on the error cases."];
+error_2(suite) ->
+ [];
+error_2(Config) when is_list(Config) ->
+ ?line lists:foreach(fun check/1, error_cases()),
+ ok.
+
+error_cases() ->
+ ["'a",
+ "\"a",
+ "'\\",
+ "\"\\",
+ "$",
+ "$\\",
+ "2.3e",
+ "2.3e-",
+ "91#9"
+].
+
+assert_type(N, integer) when is_integer(N) ->
+ ok;
+assert_type(N, atom) when is_atom(N) ->
+ ok.
+
+check(String) ->
+ Error = erl_scan:string(String),
+ check_error(Error, erl_scan).
+
+%%% (This should be useful for all format_error functions.)
+check_error({error, Info, EndLine}, Module0) ->
+ ?line {ErrorLine, Module, Desc} = Info,
+ ?line true = (Module == Module0),
+ ?line assert_type(EndLine, integer),
+ ?line assert_type(ErrorLine, integer),
+ ?line true = (ErrorLine =< EndLine),
+ ?line String = lists:flatten(Module0:format_error(Desc)),
+ ?line true = io_lib:printable_list(String).
+
+iso88591(doc) -> ["Tests the support for ISO-8859-1 i.e Latin-1"];
+iso88591(suite) -> [];
+iso88591(Config) when is_list(Config) ->
+ ?line ok =
+ case catch begin
+ %% Some atom and variable names
+ V1s = [$�,$�,$�,$�],
+ V2s = [$N,$�,$r],
+ A1s = [$h,$�,$r],
+ A2s = [$�,$r,$e],
+ %% Test parsing atom and variable characters.
+ {ok,Ts1,_} = erl_scan:string(V1s ++ " " ++ V2s ++
+ "\327" ++
+ A1s ++ " " ++ A2s),
+ V1s = atom_to_list(element(3, nth(1, Ts1))),
+ V2s = atom_to_list(element(3, nth(2, Ts1))),
+ A1s = atom_to_list(element(3, nth(4, Ts1))),
+ A2s = atom_to_list(element(3, nth(5, Ts1))),
+ %% Test printing atoms
+ A1s = flatten(print(element(3, nth(4, Ts1)))),
+ A2s = flatten(print(element(3, nth(5, Ts1)))),
+ %% Test parsing and printing strings.
+ S1 = V1s ++ "\327" ++ A1s ++ "\250" ++ A2s,
+ S1s = "\"" ++ S1 ++ "\"",
+ {ok,Ts2,_} = erl_scan:string(S1s),
+ S1 = element(3, nth(1, Ts2)),
+ S1s = flatten(print(element(3, nth(1, Ts2)))),
+ ok %It all worked
+ end of
+ {'EXIT',R} -> %Something went wrong!
+ {error,R};
+ ok -> ok %Aok
+ end.
+
+otp_7810(doc) ->
+ ["OTP-7810. White spaces, comments, and more.."];
+otp_7810(suite) ->
+ [];
+otp_7810(Config) when is_list(Config) ->
+ ?line ok = reserved_words(),
+ ?line ok = atoms(),
+ ?line ok = punctuations(),
+ ?line ok = comments(),
+ ?line ok = errors(),
+ ?line ok = integers(),
+ ?line ok = base_integers(),
+ ?line ok = floats(),
+ ?line ok = dots(),
+ ?line ok = chars(),
+ ?line ok = variables(),
+ ?line ok = eof(),
+ ?line ok = illegal(),
+ ?line ok = crashes(),
+
+ ?line ok = options(),
+ ?line ok = token_info(),
+ ?line ok = column_errors(),
+ ?line ok = white_spaces(),
+
+ ?line ok = unicode(),
+
+ ?line ok = more_chars(),
+ ?line ok = more_options(),
+ ?line ok = attributes_info(),
+ ?line ok = set_attribute(),
+
+ ok.
+
+reserved_words() ->
+ L = ['after', 'begin', 'case', 'try', 'cond', 'catch',
+ 'andalso', 'orelse', 'end', 'fun', 'if', 'let', 'of',
+ 'query', 'receive', 'when', 'bnot', 'not', 'div',
+ 'rem', 'band', 'and', 'bor', 'bxor', 'bsl', 'bsr',
+ 'or', 'xor', 'spec'] , % 'spec' shouldn't be there...
+ [begin
+ ?line {RW, true} = {RW, erl_scan:reserved_word(RW)},
+ S = atom_to_list(RW),
+ Ts = [{RW,1}],
+ ?line test_string(S, Ts)
+ end || RW <- L],
+ ok.
+
+
+atoms() ->
+ ?line test_string("a
+ b", [{atom,1,a},{atom,2,b}]),
+ ?line test_string("'a b'", [{atom,1,'a b'}]),
+ ?line test_string("a", [{atom,1,a}]),
+ ?line test_string("a@2", [{atom,1,a@2}]),
+ ?line test_string([39,65,200,39], [{atom,1,'A�'}]),
+ ?line test_string("�rlig �sten", [{atom,1,�rlig},{atom,1,�sten}]),
+ ?line {ok,[{atom,_,'$a'}],{1,6}} =
+ erl_scan:string("'$\\a'", {1,1}),
+ ?line test("'$\\a'"),
+ ok.
+
+punctuations() ->
+ L = ["<<", "<-", "<=", "<", ">>", ">=", ">", "->", "--",
+ "-", "++", "+", "=:=", "=/=", "=<", "==", "=", "/=",
+ "/", "||", "|", ":-", "::", ":"],
+ %% One token at a time:
+ [begin
+ W = list_to_atom(S),
+ Ts = [{W,1}],
+ ?line test_string(S, Ts)
+ end || S <- L],
+ Three = ["/=:=", "<=:=", "==:=", ">=:="], % three tokens...
+ No = Three ++ L,
+ SL0 = [{S1++S2,{-length(S1),S1,S2}} ||
+ S1 <- L,
+ S2 <- L,
+ not lists:member(S1++S2, No)],
+ SL = family_list(SL0),
+ %% Two tokens. When there are several answers, the one with
+ %% the longest first token is chosen:
+ %% [the special case "=<<" is among the tested ones]
+ [begin
+ W1 = list_to_atom(S1),
+ W2 = list_to_atom(S2),
+ Ts = [{W1,1},{W2,1}],
+ ?line test_string(S, Ts)
+ end || {S,[{_,S1,S2}|_]} <- SL],
+
+ PTs1 = [{'!',1},{'(',1},{')',1},{',',1},{';',1},{'=',1},{'[',1},
+ {']',1},{'{',1},{'|',1},{'}',1}],
+ ?line test_string("!(),;=[]{|}", PTs1),
+
+ PTs2 = [{'#',1},{'&',1},{'*',1},{'+',1},{'/',1},
+ {':',1},{'<',1},{'>',1},{'?',1},{'@',1},
+ {'\\',1},{'^',1},{'`',1},{'~',1}],
+ ?line test_string("#&*+/:<>?@\\^`~", PTs2),
+
+ ok.
+
+comments() ->
+ ?line test("a %%\n b"),
+ ?line {ok,[],1} = erl_scan:string("%"),
+ ?line test("a %%\n b"),
+ ?line {ok,[{atom,_,a},{atom,_,b}],{2,3}} =
+ erl_scan:string("a %%\n b",{1,1}),
+ ?line {ok,[{atom,_,a},{comment,_,"%%"},{atom,_,b}],{2,3}} =
+ erl_scan:string("a %%\n b",{1,1}, [return_comments]),
+ ?line {ok,[{atom,_,a},
+ {white_space,_," "},
+ {white_space,_,"\n "},
+ {atom,_,b}],
+ {2,3}} =
+ erl_scan:string("a %%\n b",{1,1},[return_white_spaces]),
+ ?line {ok,[{atom,_,a},
+ {white_space,_," "},
+ {comment,_,"%%"},
+ {white_space,_,"\n "},
+ {atom,_,b}],
+ {2,3}} = erl_scan:string("a %%\n b",{1,1},[return]),
+ ok.
+
+errors() ->
+ ?line {error,{1,erl_scan,{string,$',"qa"}},1} = erl_scan:string("'qa"), %'
+ ?line {error,{1,erl_scan,{string,$","str"}},1} = %"
+ erl_scan:string("\"str"), %"
+ ?line {error,{1,erl_scan,char},1} = erl_scan:string("$"),
+ ?line test_string([34,65,200,34], [{string,1,"A�"}]),
+ ?line test_string("\\", [{'\\',1}]),
+ ?line {'EXIT',_} =
+ (catch {foo, erl_scan:string('$\\a', {1,1})}), % type error
+ ?line {'EXIT',_} =
+ (catch {foo, erl_scan:tokens([], '$\\a', {1,1})}), % type error
+
+ ?line "{a,tuple}" = erl_scan:format_error({a,tuple}),
+ ok.
+
+integers() ->
+ [begin
+ I = list_to_integer(S),
+ Ts = [{integer,1,I}],
+ ?line test_string(S, Ts)
+ end || S <- [[N] || N <- lists:seq($0, $9)] ++ ["2323","000"] ],
+ ok.
+
+base_integers() ->
+ [begin
+ B = list_to_integer(BS),
+ I = erlang:list_to_integer(S, B),
+ Ts = [{integer,1,I}],
+ ?line test_string(BS++"#"++S, Ts)
+ end || {BS,S} <- [{"2","11"}, {"5","23234"}, {"12","05a"},
+ {"16","abcdef"}, {"16","ABCDEF"}] ],
+
+ ?line {error,{1,erl_scan,{base,1}},1} = erl_scan:string("1#000"),
+
+ ?line test_string("12#bc", [{integer,1,11},{atom,1,c}]),
+
+ [begin
+ Str = BS ++ "#" ++ S,
+ ?line {error,{1,erl_scan,{illegal,integer}},1} =
+ erl_scan:string(Str)
+ end || {BS,S} <- [{"3","3"},{"15","f"}, {"12","c"}] ],
+
+ ?line {ok,[{integer,1,239},{'@',1}],1} = erl_scan:string("16#ef@"),
+ ?line {ok,[{integer,1,14},{atom,1,g@}],1} = erl_scan:string("16#eg@"),
+
+ ok.
+
+floats() ->
+ [begin
+ F = list_to_float(FS),
+ Ts = [{float,1,F}],
+ ?line test_string(FS, Ts)
+ end || FS <- ["1.0","001.17","3.31200","1.0e0","1.0E17",
+ "34.21E-18", "17.0E+14"]],
+ ?line test_string("1.e2", [{integer,1,1},{'.',1},{atom,1,e2}]),
+
+ ?line {error,{1,erl_scan,{illegal,float}},1} =
+ erl_scan:string("1.0e400"),
+ [begin
+ ?line {error,{1,erl_scan,{illegal,float}},1} = erl_scan:string(S)
+ end || S <- ["1.14Ea"]],
+
+ ok.
+
+dots() ->
+ Dot = [{".", {ok,[{dot,1}],1}},
+ {". ", {ok,[{dot,1}],1}},
+ {".\n", {ok,[{dot,1}],2}},
+ {".%", {ok,[{dot,1}],1}},
+ {".\210",{ok,[{dot,1}],1}},
+ {".% �h",{ok,[{dot,1}],1}},
+ {".%\n", {ok,[{dot,1}],2}},
+ {".$", {error,{1,erl_scan,char},1}},
+ {".$\\", {error,{1,erl_scan,char},1}},
+ {".a", {ok,[{'.',1},{atom,1,a}],1}}
+ ],
+ ?line [R = erl_scan:string(S) || {S, R} <- Dot],
+
+ ?line {ok,[{dot,_}=T1],{1,2}} = erl_scan:string(".", {1,1}, text),
+ ?line [{column,1},{length,1},{line,1},{text,"."}] =
+ erl_scan:token_info(T1, [column, length, line, text]),
+ ?line {ok,[{dot,_}=T2],{1,3}} = erl_scan:string(".%", {1,1}, text),
+ ?line [{column,1},{length,1},{line,1},{text,"."}] =
+ erl_scan:token_info(T2, [column, length, line, text]),
+ ?line {ok,[{dot,_}=T3],{1,6}} =
+ erl_scan:string(".% �h", {1,1}, text),
+ ?line [{column,1},{length,1},{line,1},{text,"."}] =
+ erl_scan:token_info(T3, [column, length, line, text]),
+ ?line {error,{{1,2},erl_scan,char},{1,3}} =
+ erl_scan:string(".$", {1,1}),
+ ?line {error,{{1,2},erl_scan,char},{1,4}} =
+ erl_scan:string(".$\\", {1,1}),
+
+ ?line test(". "),
+ ?line test(". "),
+ ?line test(".\n"),
+ ?line test(".\n\n"),
+ ?line test(".\n\r"),
+ ?line test(".\n\n\n"),
+ ?line test(".\210"),
+ ?line test(".%\n"),
+ ?line test(".a"),
+
+ ?line test("%. \n. "),
+ ?line {more,C} = erl_scan:tokens([], "%. ",{1,1}, return),
+ ?line {done,{ok,[{comment,_,"%. "},
+ {white_space,_,"\n"},
+ {dot,_}],
+ {2,3}}, ""} =
+ erl_scan:tokens(C, "\n. ", {1,1}, return), % any loc, any options
+
+ ?line [test_string(S, R) ||
+ {S, R} <- [{".$\n", [{'.',1},{char,1,$\n}]},
+ {"$\\\n", [{char,1,$\n}]},
+ {"'\\\n'", [{atom,1,'\n'}]},
+ {"$\n", [{char,1,$\n}]}] ],
+ ok.
+
+chars() ->
+ [begin
+ L = lists:flatten(io_lib:format("$\\~.8b", [C])),
+ Ts = [{char,1,C}],
+ ?line test_string(L, Ts)
+ end || C <- lists:seq(0, 255)],
+
+ %% Leading zeroes...
+ [begin
+ L = lists:flatten(io_lib:format("$\\~3.8.0b", [C])),
+ Ts = [{char,1,C}],
+ ?line test_string(L, Ts)
+ end || C <- lists:seq(0, 255)],
+
+ %% $\^\n now increments the line...
+ [begin
+ L = "$\\^" ++ [C],
+ Ts = [{char,1,C band 2#11111}],
+ ?line test_string(L, Ts)
+ end || C <- lists:seq(0, 255)],
+
+ [begin
+ L = "$\\" ++ [C],
+ Ts = [{char,1,V}],
+ ?line test_string(L, Ts)
+ end || {C,V} <- [{$n,$\n}, {$r,$\r}, {$t,$\t}, {$v,$\v},
+ {$b,$\b}, {$f,$\f}, {$e,$\e}, {$s,$\s},
+ {$d,$\d}]],
+
+ EC = [$\n,$\r,$\t,$\v,$\b,$\f,$\e,$\s,$\d],
+ Ds = lists:seq($0, $9),
+ X = [$^,$n,$r,$t,$v,$b,$f,$e,$s,$d],
+ New = [${,$x],
+ No = EC ++ Ds ++ X ++ New,
+ [begin
+ L = "$\\" ++ [C],
+ Ts = [{char,1,C}],
+ ?line test_string(L, Ts)
+ end || C <- lists:seq(0, 255) -- No],
+
+ [begin
+ L = "'$\\" ++ [C] ++ "'",
+ Ts = [{atom,1,list_to_atom("$"++[C])}],
+ ?line test_string(L, Ts)
+ end || C <- lists:seq(0, 255) -- No],
+
+ ?line test_string("\"\\013a\\\n\"", [{string,1,"\va\n"}]),
+
+ ?line test_string("'\n'", [{atom,1,'\n'}]),
+ ?line test_string("\"\n\a\"", [{string,1,"\na"}]),
+
+ %% No escape
+ [begin
+ L = "$" ++ [C],
+ Ts = [{char,1,C}],
+ ?line test_string(L, Ts)
+ end || C <- lists:seq(0, 255) -- (No ++ [$\\])],
+ ?line test_string("$\n", [{char,1,$\n}]),
+
+ ?line {error,{{1,1},erl_scan,char},{1,4}} =
+ erl_scan:string("$\\^",{1,1}),
+ ?line test_string("$\\\n", [{char,1,$\n}]),
+ %% Robert's scanner returns line 1:
+ ?line test_string("$\\\n", [{char,1,$\n}]),
+ ?line test_string("$\n\n", [{char,1,$\n}]),
+ ?line test("$\n\n"),
+ ok.
+
+
+variables() ->
+ ?line test_string(" \237_Aou�eiy��", [{var,1,'_Aou�eiy��'}]),
+ ?line test_string("A_b_c@", [{var,1,'A_b_c@'}]),
+ ?line test_string("V@2", [{var,1,'V@2'}]),
+ ?line test_string("ABD�", [{var,1,'ABD�'}]),
+ ?line test_string("�rlig �sten", [{var,1,'�rlig'},{var,1,'�sten'}]),
+ ok.
+
+eof() ->
+ ?line {done,{eof,1},eof} = erl_scan:tokens([], eof, 1),
+ {more, C1} = erl_scan:tokens([]," \n", 1),
+ ?line {done,{eof,2},eof} = erl_scan:tokens(C1, eof, 1),
+ {more, C2} = erl_scan:tokens([], "abra", 1),
+ %% An error before R13A.
+ %% ?line {done,Err={error,{1,erl_scan,scan},1},eof} =
+ ?line {done,{ok,[{atom,1,abra}],1},eof} =
+ erl_scan:tokens(C2, eof, 1),
+
+ %% With column.
+ ?line {more, C3} = erl_scan:tokens([]," \n",{1,1}),
+ ?line {done,{eof,{2,1}},eof} = erl_scan:tokens(C3, eof, 1),
+ {more, C4} = erl_scan:tokens([], "abra", {1,1}),
+ %% An error before R13A.
+ %% ?line {done,{error,{{1,1},erl_scan,scan},{1,5}},eof} =
+ ?line {done,{ok,[{atom,_,abra}],{1,5}},eof} =
+ erl_scan:tokens(C4, eof, 1),
+
+ %% Robert's scanner returns "" as LeftoverChars;
+ %% the R12B scanner returns eof as LeftoverChars: (eof is correct)
+ ?line {more, C5} = erl_scan:tokens([], "a", 1),
+ %% An error before R13A.
+ %% ?line {done,{error,{1,erl_scan,scan},1},eof} =
+ ?line {done,{ok,[{atom,1,a}],1},eof} =
+ erl_scan:tokens(C5,eof,1),
+
+ %% A dot followed by eof is special:
+ ?line {more, C} = erl_scan:tokens([], "a.", 1),
+ ?line {done,{ok,[{atom,1,a},{dot,1}],1},eof} = erl_scan:tokens(C,eof,1),
+ ?line {ok,[{atom,1,foo},{dot,1}],1} = erl_scan:string("foo."),
+
+ ok.
+
+illegal() ->
+ Atom = lists:duplicate(1000, $a),
+ ?line {error,{1,erl_scan,{illegal,atom}},1} = erl_scan:string(Atom),
+ ?line {done,{error,{1,erl_scan,{illegal,atom}},1},". "} =
+ erl_scan:tokens([], Atom++". ", 1),
+ QAtom = "'" ++ Atom ++ "'",
+ ?line {error,{1,erl_scan,{illegal,atom}},1} = erl_scan:string(QAtom),
+ ?line {done,{error,{1,erl_scan,{illegal,atom}},1},". "} =
+ erl_scan:tokens([], QAtom++". ", 1),
+ Var = lists:duplicate(1000, $A),
+ ?line {error,{1,erl_scan,{illegal,var}},1} = erl_scan:string(Var),
+ ?line {done,{error,{1,erl_scan,{illegal,var}},1},". "} =
+ erl_scan:tokens([], Var++". ", 1),
+ Float = "1" ++ lists:duplicate(400, $0) ++ ".0",
+ ?line {error,{1,erl_scan,{illegal,float}},1} = erl_scan:string(Float),
+ ?line {done,{error,{1,erl_scan,{illegal,float}},1},". "} =
+ erl_scan:tokens([], Float++". ", 1),
+ String = "\"43\\x{aaaaaa}34\"",
+ ?line {error,{1,erl_scan,{illegal,character}},1} = erl_scan:string(String),
+ ?line {done,{error,{1,erl_scan,{illegal,character}},1},"34\". "} =
+ %% Would be nice if `34\"' were skipped...
+ %% Maybe, but then the LeftOverChars would not be the characters
+ %% immediately following the end location of the error.
+ erl_scan:tokens([], String++". ", 1),
+
+ ?line {error,{{1,1},erl_scan,{illegal,atom}},{1,1001}} =
+ erl_scan:string(Atom, {1,1}),
+ ?line {done,{error,{{1,5},erl_scan,{illegal,atom}},{1,1005}},". "} =
+ erl_scan:tokens([], "foo "++Atom++". ", {1,1}),
+ ?line {error,{{1,1},erl_scan,{illegal,atom}},{1,1003}} =
+ erl_scan:string(QAtom, {1,1}),
+ ?line {done,{error,{{1,5},erl_scan,{illegal,atom}},{1,1007}},". "} =
+ erl_scan:tokens([], "foo "++QAtom++". ", {1,1}),
+ ?line {error,{{1,1},erl_scan,{illegal,var}},{1,1001}} =
+ erl_scan:string(Var, {1,1}),
+ ?line {done,{error,{{1,5},erl_scan,{illegal,var}},{1,1005}},". "} =
+ erl_scan:tokens([], "foo "++Var++". ", {1,1}),
+ ?line {error,{{1,1},erl_scan,{illegal,float}},{1,404}} =
+ erl_scan:string(Float, {1,1}),
+ ?line {done,{error,{{1,5},erl_scan,{illegal,float}},{1,408}},". "} =
+ erl_scan:tokens([], "foo "++Float++". ", {1,1}),
+ ?line {error,{{1,4},erl_scan,{illegal,character}},{1,14}} =
+ erl_scan:string(String, {1,1}),
+ ?line {done,{error,{{1,4},erl_scan,{illegal,character}},{1,14}},"34\". "} =
+ erl_scan:tokens([], String++". ", {1,1}),
+ ok.
+
+crashes() ->
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string([-1])}), % type error
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("$"++[-1])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("$\\"++[-1])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("$\\^"++[-1])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string([$",-1,$"],{1,1})}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("\"\\v"++[-1,$"])}), %$"
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string([$",-1,$"])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("% foo"++[-1])}),
+ ?line {'EXIT',_} =
+ (catch {foo, erl_scan:string("% foo"++[-1],{1,1})}),
+
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string([a])}), % type error
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("$"++[a])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("$\\"++[a])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("$\\^"++[a])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string([$",a,$"],{1,1})}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("\"\\v"++[a,$"])}), %$"
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string([$",a,$"])}),
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string("% foo"++[a])}),
+ ?line {'EXIT',_} =
+ (catch {foo, erl_scan:string("% foo"++[a],{1,1})}),
+
+ ?line {'EXIT',_} = (catch {foo, erl_scan:string([3.0])}), % type error
+
+ ok.
+
+options() ->
+ %% line and column are not options, but tested here
+ ?line {ok,[{atom,1,foo},{white_space,1," "},{comment,1,"% bar"}], 1} =
+ erl_scan:string("foo % bar", 1, return),
+ ?line {ok,[{atom,1,foo},{white_space,1," "}],1} =
+ erl_scan:string("foo % bar", 1, return_white_spaces),
+ ?line {ok,[{atom,1,foo},{comment,1,"% bar"}],1} =
+ erl_scan:string("foo % bar", 1, return_comments),
+ ?line {ok,[{atom,17,foo}],17} =
+ erl_scan:string("foo % bar", 17),
+ ?line {'EXIT',{function_clause,_}} =
+ (catch {foo,
+ erl_scan:string("foo % bar", {a,1}, [])}), % type error
+ ?line {ok,[{atom,_,foo}],{17,18}} =
+ erl_scan:string("foo % bar", {17,9}, []),
+ ?line {'EXIT',{function_clause,_}} =
+ (catch {foo,
+ erl_scan:string("foo % bar", {1,0}, [])}), % type error
+ ?line {ok,[{foo,1}],1} =
+ erl_scan:string("foo % bar",1, [{reserved_word_fun,
+ fun(W) -> W =:= foo end}]),
+ ?line {'EXIT',{badarg,_}} =
+ (catch {foo,
+ erl_scan:string("foo % bar",1, % type error
+ [{reserved_word_fun,
+ fun(W,_) -> W =:= foo end}])}),
+ ok.
+
+more_options() ->
+ ?line {ok,[{atom,A1,foo}],{19,20}} =
+ erl_scan:string("foo", {19,17},[]),
+ ?line [{column,17},{line,19}] = erl_scan:attributes_info(A1),
+ ?line {done,{ok,[{atom,A2,foo},{dot,_}],{19,22}},[]} =
+ erl_scan:tokens([], "foo. ", {19,17}, [bad_opt]), % type error
+ ?line [{column,17},{line,19}] = erl_scan:attributes_info(A2),
+ ?line {ok,[{atom,A3,foo}],{19,20}} =
+ erl_scan:string("foo", {19,17},[text]),
+ ?line [{column,17},{length,3},{line,19},{text,"foo"}] =
+ erl_scan:attributes_info(A3),
+
+ ?line {ok,[{atom,A4,foo}],1} = erl_scan:string("foo", 1, [text]),
+ ?line [{length,3},{line,1},{text,"foo"}] = erl_scan:attributes_info(A4),
+
+ ok.
+
+token_info() ->
+ ?line {ok,[T1],_} = erl_scan:string("foo", {1,18}, [text]),
+ {'EXIT',{badarg,_}} =
+ (catch {foo, erl_scan:token_info(T1, foo)}), % type error
+ ?line {line,1} = erl_scan:token_info(T1, line),
+ ?line {column,18} = erl_scan:token_info(T1, column),
+ ?line {length,3} = erl_scan:token_info(T1, length),
+ ?line {text,"foo"} = erl_scan:token_info(T1, text),
+ ?line [{category,atom},{column,18},{length,3},{line,1},
+ {symbol,foo},{text,"foo"}] =
+ erl_scan:token_info(T1),
+ ?line [{length,3},{column,18}] =
+ erl_scan:token_info(T1, [length, column]),
+ ?line [{location,{1,18}}] =
+ erl_scan:token_info(T1, [location]),
+ ?line {category,atom} = erl_scan:token_info(T1, category),
+ ?line [{symbol,foo}] = erl_scan:token_info(T1, [symbol]),
+
+ ?line {ok,[T2],_} = erl_scan:string("foo", 1, []),
+ ?line {line,1} = erl_scan:token_info(T2, line),
+ ?line undefined = erl_scan:token_info(T2, column),
+ ?line undefined = erl_scan:token_info(T2, length),
+ ?line undefined = erl_scan:token_info(T2, text),
+ ?line {location,1} = erl_scan:token_info(T2, location),
+ ?line [{category,atom},{line,1},{symbol,foo}] = erl_scan:token_info(T2),
+ ?line [{line,1}] = erl_scan:token_info(T2, [length, line]),
+
+ ?line {ok,[T3],_} = erl_scan:string("=", 1, []),
+ ?line [{line,1}] = erl_scan:token_info(T3, [column, line]),
+ ?line {category,'='} = erl_scan:token_info(T3, category),
+ ?line [{symbol,'='}] = erl_scan:token_info(T3, [symbol]),
+ ok.
+
+attributes_info() ->
+ ?line {'EXIT',_} =
+ (catch {foo,erl_scan:attributes_info(foo)}), % type error
+ ?line [{line,18}] = erl_scan:attributes_info(18),
+ ?line {location,19} = erl_scan:attributes_info(19, location),
+ ?line {ok,[{atom,A0,foo}],_} = erl_scan:string("foo", 19, [text]),
+ ?line {location,19} = erl_scan:attributes_info(A0, location),
+
+ ?line {ok,[{atom,A3,foo}],_} = erl_scan:string("foo", {1,3}, [text]),
+ ?line {line,1} = erl_scan:attributes_info(A3, line),
+ ?line {column,3} = erl_scan:attributes_info(A3, column),
+ ?line {location,{1,3}} = erl_scan:attributes_info(A3, location),
+ ?line {text,"foo"} = erl_scan:attributes_info(A3, text),
+
+ ?line {ok,[{atom,A4,foo}],_} = erl_scan:string("foo", 2, [text]),
+ ?line {line,2} = erl_scan:attributes_info(A4, line),
+ ?line undefined = erl_scan:attributes_info(A4, column),
+ ?line {location,2} = erl_scan:attributes_info(A4, location),
+ ?line {text,"foo"} = erl_scan:attributes_info(A4, text),
+
+ ?line {ok,[{atom,A5,foo}],_} = erl_scan:string("foo", {1,3}, []),
+ ?line {line,1} = erl_scan:attributes_info(A5, line),
+ ?line {column,3} = erl_scan:attributes_info(A5, column),
+ ?line {location,{1,3}} = erl_scan:attributes_info(A5, location),
+ ?line undefined = erl_scan:attributes_info(A5, text),
+
+ ?line undefined = erl_scan:attributes_info([], line), % type error
+
+ ok.
+
+set_attribute() ->
+ F = fun(Line) -> -Line end,
+ ?line -2 = erl_scan:set_attribute(line, 2, F),
+ ?line {ok,[{atom,A1,foo}],_} = erl_scan:string("foo", {9,17}),
+ ?line A2 = erl_scan:set_attribute(line, A1, F),
+ ?line {line,-9} = erl_scan:attributes_info(A2, line),
+ ?line {location,{-9,17}} = erl_scan:attributes_info(A2, location),
+ ?line [{line,-9},{column,17}] =
+ erl_scan:attributes_info(A2, [line,column,text]),
+
+ F2 = fun(Line) -> {17,Line} end,
+ ?line Attr1 = erl_scan:set_attribute(line, 2, F2),
+ ?line {line,{17,2}} = erl_scan:attributes_info(Attr1, line),
+ ?line undefined = erl_scan:attributes_info(Attr1, column),
+ ?line {location,{17,2}} = % a bit mixed up
+ erl_scan:attributes_info(Attr1, location),
+
+ ?line A3 = erl_scan:set_attribute(line, A1, F2),
+ ?line {line,{17,9}} = erl_scan:attributes_info(A3, line),
+ ?line {location,{{17,9},17}} = erl_scan:attributes_info(A3, location),
+ ?line [{line,{17,9}},{column,17}] =
+ erl_scan:attributes_info(A3, [line,column,text]),
+
+ ?line {ok,[{atom,A4,foo}],_} = erl_scan:string("foo", {9,17}, [text]),
+ ?line A5 = erl_scan:set_attribute(line, A4, F),
+ ?line {line,-9} = erl_scan:attributes_info(A5, line),
+ ?line {location,{-9,17}} = erl_scan:attributes_info(A5, location),
+ ?line [{line,-9},{column,17},{text,"foo"}] =
+ erl_scan:attributes_info(A5, [line,column,text]),
+
+ ?line {ok,[{atom,A6,foo}],_} = erl_scan:string("foo", 11, [text]),
+ ?line A7 = erl_scan:set_attribute(line, A6, F2),
+ ?line {line,{17,11}} = erl_scan:attributes_info(A7, line),
+ ?line {location,{17,11}} = % mixed up
+ erl_scan:attributes_info(A7, location),
+ ?line [{line,{17,11}},{text,"foo"}] =
+ erl_scan:attributes_info(A7, [line,column,text]),
+
+ ?line {'EXIT',_} =
+ (catch {foo, erl_scan:set_attribute(line, [], F2)}), % type error
+ ?line {'EXIT',{badarg,_}} =
+ (catch {foo, erl_scan:set_attribute(column, [], F2)}), % type error
+ ok.
+
+column_errors() ->
+ ?line {error,{{1,1},erl_scan,{string,$',""}},{1,3}} = % $'
+ erl_scan:string("'\\",{1,1}),
+ ?line {error,{{1,1},erl_scan,{string,$",""}},{1,3}} = % $"
+ erl_scan:string("\"\\",{1,1}),
+
+ ?line {error,{{1,1},erl_scan,{string,$',""}},{1,2}} = % $'
+ erl_scan:string("'",{1,1}),
+ ?line {error,{{1,1},erl_scan,{string,$",""}},{1,2}} = % $"
+ erl_scan:string("\"",{1,1}),
+
+ ?line {error,{{1,1},erl_scan,char},{1,2}} =
+ erl_scan:string("$",{1,1}),
+
+ ?line {error,{{1,2},erl_scan,{string,$',"1234567890123456"}},{1,20}} = %'
+ erl_scan:string(" '12345678901234567", {1,1}),
+ ?line {error,{{1,2},erl_scan,{string,$',"123456789012345 "}}, {1,20}} = %'
+ erl_scan:string(" '123456789012345\\s", {1,1}),
+ ?line {error,{{1,2},erl_scan,{string,$","1234567890123456"}},{1,20}} = %"
+ erl_scan:string(" \"12345678901234567", {1,1}),
+ ?line {error,{{1,2},erl_scan,{string,$","123456789012345 "}}, {1,20}} = %"
+ erl_scan:string(" \"123456789012345\\s", {1,1}),
+ ?line {error,{{1,2},erl_scan,{string,$',"1234567890123456"}},{2,1}} = %'
+ erl_scan:string(" '12345678901234567\n", {1,1}),
+ ok.
+
+white_spaces() ->
+ ?line {ok,[{white_space,_,"\r"},
+ {white_space,_," "},
+ {atom,_,a},
+ {white_space,_,"\n"}],
+ _} = erl_scan:string("\r a\n", {1,1}, return),
+ ?line test("\r a\n"),
+ L = "{\"a\nb\", \"a\\nb\",\nabc\r,def}.\n\n",
+ ?line {ok,[{'{',_},
+ {string,_,"a\nb"},
+ {',',_},
+ {white_space,_," "},
+ {string,_,"a\nb"},
+ {',',_},
+ {white_space,_,"\n"},
+ {atom,_,abc},
+ {white_space,_,"\r"},
+ {',',_},
+ {atom,_,def},
+ {'}',_},
+ {dot,_},
+ {white_space,_,"\n"}],
+ _} = erl_scan:string(L, {1,1}, return),
+ ?line test(L),
+ ?line test("\"\n\"\n"),
+ ?line test("\n\r\n"),
+ ?line test("\n\r"),
+ ?line test("\r\n"),
+ ?line test("\n\f"),
+ ?line [test(lists:duplicate(N, $\t)) || N <- lists:seq(1, 20)],
+ ?line [test([$\n|lists:duplicate(N, $\t)]) || N <- lists:seq(1, 20)],
+ ?line [test(lists:duplicate(N, $\s)) || N <- lists:seq(1, 20)],
+ ?line [test([$\n|lists:duplicate(N, $\s)]) || N <- lists:seq(1, 20)],
+ ?line test("\v\f\n\v "),
+ ?line test("\n\e\n\b\f\n\da\n"),
+ ok.
+
+unicode() ->
+ ?line {ok,[{char,1,83},{integer,1,45}],1} =
+ erl_scan:string("$\\12345"), % not unicode
+
+ ?line {error,{1,erl_scan,{illegal,character}},1} =
+ erl_scan:string([1089]),
+ ?line {error,{{1,1},erl_scan,{illegal,character}},{1,2}} =
+ erl_scan:string([1089], {1,1}),
+ ?line {error,{1,erl_scan,{illegal,character}},1} =
+ %% ?line {error,{1,erl_scan,{illegal,atom}},1} =
+ erl_scan:string("'a"++[1089]++"b'"),
+ ?line {error,{{1,3},erl_scan,{illegal,character}},{1,4}} =
+ erl_scan:string("'a"++[1089]++"b'", {1,1}),
+ ?line test("\"a"++[1089]++"b\""),
+ ?line {ok,[{char,1,1}],1} = erl_scan:string([$$,$\\,$^,1089]),
+
+ ?line {error,{1,erl_scan,Error},1} = erl_scan:string("\"qa\x{aaa}"),
+ ?line "unterminated string starting with \"qa\\x{AAA}\"" =
+ erl_scan:format_error(Error),
+ ?line {error,{{1,1},erl_scan,_},{1,11}} =
+ erl_scan:string("\"qa\\x{aaa}",{1,1}),
+ ?line {error,{{1,4},erl_scan,{illegal,character}},{1,11}} =
+ erl_scan:string("'qa\\x{aaa}'",{1,1}),
+
+ Tags = [category, column, length, line, symbol, text],
+
+ %% Workaround. No character codes greater than 255! To be changed.
+ %% Note: don't remove these tests, just modify them!
+
+ ?line {ok,[{integer,1,1089}],1} = erl_scan:string([$$,1089]),
+ ?line {ok,[{integer,1,1089}],1} = erl_scan:string([$$,$\\,1089]),
+
+ Qs = "$\\x{aaa}",
+ ?line {ok,[{integer,1,16#aaa}],1} = erl_scan:string(Qs),
+ ?line {ok,[Q2],{1,9}} = erl_scan:string("$\\x{aaa}", {1,1}, text),
+ ?line [{category,integer},{column,1},{length,8},
+ {line,1},{symbol,16#aaa},{text,Qs}] =
+ erl_scan:token_info(Q2),
+
+ U1 = "\"\\x{aaa}\"",
+ ?line {ok,[T1,T2,T3],{1,10}} = erl_scan:string(U1, {1,1}, text),
+ ?line [{category,'['},{column,1},{length,1},{line,1},
+ {symbol,'['},{text,"\""}] = erl_scan:token_info(T1, Tags),
+ ?line [{category,integer},{column,2},{length,7},
+ {line,1},{symbol,16#aaa},{text,"\\x{aaa}"}] =
+ erl_scan:token_info(T2, Tags),
+ ?line [{category,']'},{column,9},{length,1},{line,1},
+ {symbol,']'},{text,"\""}] = erl_scan:token_info(T3, Tags),
+ ?line {ok,[{'[',1},{integer,1,16#aaa},{']',1}],1} =
+ erl_scan:string(U1, 1),
+
+ U2 = "\"\\x41\\x{fff}\\x42\"",
+ ?line {ok,[{'[',1},{char,1,16#41},{',',1},{integer,1,16#fff},
+ {',',1},{char,1,16#42},{']',1}],1} = erl_scan:string(U2, 1),
+
+ U3 = "\"a\n\\x{fff}\n\"",
+ ?line {ok,[{'[',1},{char,1,$a},{',',1},{char,1,$\n},
+ {',',2},{integer,2,16#fff},{',',2},{char,2,$\n},
+ {']',3}],3} =
+ erl_scan:string(U3, 1),
+
+ U4 = "\"\\^\n\\x{aaa}\\^\n\"",
+ ?line {ok,[{'[',1},{char,1,$\n},{',',2},{integer,2,16#aaa},
+ {',',2},{char,2,$\n},{']',3}],3} = erl_scan:string(U4, 1),
+
+ %% Keep these tests:
+ ?line test(Qs),
+ ?line test(U1),
+ ?line test(U2),
+ ?line test(U3),
+ ?line test(U4),
+
+ Str1 = "\"ab" ++ [1089] ++ "cd\"",
+ ?line {ok,[{'[',1},{char,1,$a},{',',1},{char,1,$b},{',',1},
+ {integer,1,1089},{',',1},{char,1,$c},{',',1},
+ {char,1,$d},{']',1}],1} = erl_scan:string(Str1),
+ ?line {ok,[{'[',_},{char,_,$a},{',',_},{char,_,$b},{',',_},
+ {integer,_,1089},{',',_},{char,_,$c},{',',_},
+ {char,_,$d},{']',_}],{1,8}} = erl_scan:string(Str1, {1,1}),
+ ?line test(Str1),
+ Comment = "%% "++[1089],
+ ?line {ok,[{comment,1,[$%,$%,$\s,1089]}],1} =
+ erl_scan:string(Comment, 1, return),
+ ?line {ok,[{comment,_,[$%,$%,$\s,1089]}],{1,5}} =
+ erl_scan:string(Comment, {1,1}, return),
+ ok.
+
+more_chars() ->
+ %% Due to unicode, the syntax has been incompatibly augmented:
+ %% $\x{...}, $\xHH
+
+ %% All kinds of tests...
+ ?line {ok,[{char,_,123}],{1,4}} =
+ erl_scan:string("$\\{",{1,1}),
+ ?line {more, C1} = erl_scan:tokens([], "$\\{", {1,1}),
+ ?line {done,{ok,[{char,_,123}],{1,4}},eof} =
+ erl_scan:tokens(C1, eof, 1),
+ ?line {ok,[{char,1,123},{atom,1,a},{'}',1}],1} =
+ erl_scan:string("$\\{a}"),
+
+ ?line {error,{{1,1},erl_scan,char},{1,4}} =
+ erl_scan:string("$\\x", {1,1}),
+ ?line {error,{{1,1},erl_scan,char},{1,5}} =
+ erl_scan:string("$\\x{",{1,1}),
+ ?line {more, C3} = erl_scan:tokens([], "$\\x", {1,1}),
+ ?line {done,{error,{{1,1},erl_scan,char},{1,4}},eof} =
+ erl_scan:tokens(C3, eof, 1),
+ ?line {error,{{1,1},erl_scan,char},{1,5}} =
+ erl_scan:string("$\\x{",{1,1}),
+ ?line {more, C2} = erl_scan:tokens([], "$\\x{", {1,1}),
+ ?line {done,{error,{{1,1},erl_scan,char},{1,5}},eof} =
+ erl_scan:tokens(C2, eof, 1),
+ ?line {error,{1,erl_scan,{illegal,character}},1} =
+ erl_scan:string("$\\x{g}"),
+ ?line {error,{{1,1},erl_scan,{illegal,character}},{1,5}} =
+ erl_scan:string("$\\x{g}", {1,1}),
+ ?line {error,{{1,1},erl_scan,{illegal,character}},{1,6}} =
+ erl_scan:string("$\\x{}",{1,1}),
+
+ ?line test("\"\\{0}\""),
+ ?line test("\"\\x{0}\""),
+ ?line test("\'\\{0}\'"),
+ ?line test("\'\\x{0}\'"),
+
+ ?line {error,{{2,3},erl_scan,{illegal,character}},{2,6}} =
+ erl_scan:string("\"ab \n $\\x{g}\"",{1,1}),
+ ?line {error,{{2,3},erl_scan,{illegal,character}},{2,6}} =
+ erl_scan:string("\'ab \n $\\x{g}\'",{1,1}),
+
+ ?line test("$\\{34}"),
+ ?line test("$\\x{34}"),
+ ?line test("$\\{377}"),
+ ?line test("$\\x{FF}"),
+ ?line test("$\\{400}"),
+ ?line test("$\\x{100}"),
+ ?line test("$\\x{10FFFF}"),
+ ?line test("$\\x{10ffff}"),
+ ?line test("\"$\n \\{1}\""),
+ ?line {error,{1,erl_scan,{illegal,character}},1} =
+ erl_scan:string("$\\x{110000}"),
+ ?line {error,{{1,1},erl_scan,{illegal,character}},{1,12}} =
+ erl_scan:string("$\\x{110000}", {1,1}),
+
+ ?line {error,{{1,1},erl_scan,{illegal,character}},{1,4}} =
+ erl_scan:string("$\\xfg", {1,1}),
+
+ ?line test("$\\xffg"),
+
+ ?line {error,{{1,1},erl_scan,{illegal,character}},{1,4}} =
+ erl_scan:string("$\\xg", {1,1}),
+ ok.
+
+test_string(String, Expected) ->
+ {ok, Expected, _End} = erl_scan:string(String),
+ test(String).
+
+%% test_string(String, Expected, StartLocation, Options) ->
+%% {ok, Expected, _End} = erl_scan:string(String, StartLocation, Options),
+%% test(String).
+
+%% There are no checks of the tags...
+test(String) ->
+ %% io:format("Testing `~ts'~n", [String]),
+ [{Tokens, End},
+ {Wtokens, Wend},
+ {Ctokens, Cend},
+ {CWtokens, CWend},
+ {CWtokens2, _}] =
+ [scan_string_with_column(String, X) ||
+ X <- [[],
+ [return_white_spaces],
+ [return_comments],
+ [return],
+ [return]]], % for white space compaction test
+
+ {end1,End,Wend} = {end1,Wend,End},
+ {end2,Wend,Cend} = {end2,Cend,Wend},
+ {end3,Cend,CWend} = {end3,CWend,Cend},
+
+ %% Test that the tokens that are common to two token lists are identical.
+ {none,Tokens} = {none, filter_tokens(CWtokens, [white_space,comment])},
+ {comments,Ctokens} =
+ {comments,filter_tokens(CWtokens, [white_space])},
+ {white_spaces,Wtokens} =
+ {white_spaces,filter_tokens(CWtokens, [comment])},
+
+ %% Use token attributes to extract parts from the original string,
+ %% and check that the parts are identical to the token strings.
+ {Line,Column} = test_decorated_tokens(String, CWtokens),
+ {deco,{Line,Column},End} = {deco,End,{Line,Column}},
+
+ %% Almost the same again: concat texts to get the original:
+ Text = get_text(CWtokens),
+ {text,Text,String} = {text,String,Text},
+
+ %% Test that white spaces occupy less heap than the worst case.
+ ok = test_white_space_compaction(CWtokens, CWtokens2),
+
+ %% Test that white newlines are always first in text:
+ WhiteTokens = select_tokens(CWtokens, [white_space]),
+ ok = newlines_first(WhiteTokens),
+
+ %% Line attribute only:
+ [Simple,Wsimple,Csimple,WCsimple] = Simples =
+ [element(2, erl_scan:string(String, 1, Opts)) ||
+ Opts <- [[],
+ [return_white_spaces],
+ [return_comments],
+ [return]]],
+ {consistent,true} = {consistent,consistent_attributes(Simples)},
+ {simple_wc,WCsimple} = {simple_wc,simplify(CWtokens)},
+ {simple,Simple} = {simple,filter_tokens(WCsimple, [white_space,comment])},
+ {simple_c,Csimple} = {simple_c,filter_tokens(WCsimple, [white_space])},
+ {simple_w,Wsimple} = {simple_w,filter_tokens(WCsimple, [comment])},
+
+ %% Line attribute only, with text:
+ [SimpleTxt,WsimpleTxt,CsimpleTxt,WCsimpleTxt] = SimplesTxt =
+ [element(2, erl_scan:string(String, 1, [text|Opts])) ||
+ Opts <- [[],
+ [return_white_spaces],
+ [return_comments],
+ [return]]],
+ TextTxt = get_text(WCsimpleTxt),
+ {text_txt,TextTxt,String} = {text_txt,String,TextTxt},
+ {consistent_txt,true} =
+ {consistent_txt,consistent_attributes(SimplesTxt)},
+ {simple_txt,SimpleTxt} =
+ {simple_txt,filter_tokens(WCsimpleTxt, [white_space,comment])},
+ {simple_c_txt,CsimpleTxt} =
+ {simple_c_txt,filter_tokens(WCsimpleTxt, [white_space])},
+ {simple_w_txt,WsimpleTxt} =
+ {simple_w_txt,filter_tokens(WCsimpleTxt, [comment])},
+
+ ok.
+
+test_white_space_compaction(Tokens, Tokens2) when Tokens =:= Tokens2 ->
+ [WS, WS2] = [select_tokens(Ts, [white_space]) || Ts <- [Tokens, Tokens2]],
+ test_wsc(WS, WS2).
+
+test_wsc([], []) ->
+ ok;
+test_wsc([Token|Tokens], [Token2|Tokens2]) ->
+ [Text, Text2] = [Text ||
+ {text, Text} <-
+ [erl_scan:token_info(T, text) ||
+ T <- [Token, Token2]]],
+ Sz = erts_debug:size(Text),
+ Sz2 = erts_debug:size({Text, Text2}),
+ IsCompacted = Sz2 < 2*Sz+erts_debug:size({a,a}),
+ ToBeCompacted = is_compacted(Text),
+ if
+ IsCompacted =:= ToBeCompacted ->
+ test_wsc(Tokens, Tokens2);
+ true ->
+ {compaction_error, Token}
+ end.
+
+is_compacted("\r") ->
+ true;
+is_compacted("\n\r") ->
+ true;
+is_compacted("\n\f") ->
+ true;
+is_compacted([$\n|String]) ->
+ all_spaces(String)
+ orelse
+ all_tabs(String);
+is_compacted(String) ->
+ all_spaces(String)
+ orelse
+ all_tabs(String).
+
+all_spaces(L) ->
+ all_same(L, $\s).
+
+all_tabs(L) ->
+ all_same(L, $\t).
+
+all_same(L, Char) ->
+ lists:all(fun(C) -> C =:= Char end, L).
+
+newlines_first([]) ->
+ ok;
+newlines_first([Token|Tokens]) ->
+ {text,Text} = erl_scan:token_info(Token, text),
+ Nnls = length([C || C <- Text, C =:= $\n]),
+ OK = case Text of
+ [$\n|_] ->
+ Nnls =:= 1;
+ _ ->
+ Nnls =:= 0
+ end,
+ if
+ OK -> newlines_first(Tokens);
+ true -> OK
+ end.
+
+filter_tokens(Tokens, Tags) ->
+ lists:filter(fun(T) -> not lists:member(element(1, T), Tags) end, Tokens).
+
+select_tokens(Tokens, Tags) ->
+ lists:filter(fun(T) -> lists:member(element(1, T), Tags) end, Tokens).
+
+simplify([Token|Tokens]) ->
+ {line,Line} = erl_scan:token_info(Token, line),
+ [setelement(2, Token, Line) | simplify(Tokens)];
+simplify([]) ->
+ [].
+
+get_text(Tokens) ->
+ lists:flatten(
+ [T ||
+ Token <- Tokens,
+ ({text,T} = erl_scan:token_info(Token, text)) =/= []]).
+
+test_decorated_tokens(String, Tokens) ->
+ ToksAttrs = token_attrs(Tokens),
+ test_strings(ToksAttrs, String, 1, 1).
+
+token_attrs(Tokens) ->
+ [{L,C,Len,T} ||
+ Token <- Tokens,
+ ([{line,L},{column,C},{length,Len},{text,T}] =
+ erl_scan:token_info(Token, [line,column,length,text])) =/= []].
+
+test_strings([], _S, Line, Column) ->
+ {Line,Column};
+test_strings([{L,C,Len,T}=Attr|Attrs], String0, Line0, Column0) ->
+ {String1, Column1} = skip_newlines(String0, L, Line0, Column0),
+ String = skip_chars(String1, C-Column1),
+ {Str,Rest} = lists:split(Len, String),
+ if
+ Str =:= T ->
+ {Line,Column} = string_newlines(T, L, C),
+ test_strings(Attrs, Rest, Line, Column);
+ true ->
+ {token_error, Attr, Str}
+ end.
+
+skip_newlines(String, Line, Line, Column) ->
+ {String, Column};
+skip_newlines([$\n|String], L, Line, _Column) ->
+ skip_newlines(String, L, Line+1, 1);
+skip_newlines([_|String], L, Line, Column) ->
+ skip_newlines(String, L, Line, Column+1).
+
+skip_chars(String, 0) ->
+ String;
+skip_chars([_|String], N) ->
+ skip_chars(String, N-1).
+
+string_newlines([$\n|String], Line, _Column) ->
+ string_newlines(String, Line+1, 1);
+string_newlines([], Line, Column) ->
+ {Line, Column};
+string_newlines([_|String], Line, Column) ->
+ string_newlines(String, Line, Column+1).
+
+scan_string_with_column(String, Options0) ->
+ Options = [text | Options0],
+ StartLoc = {1, 1},
+ {ok, Ts1, End1} = erl_scan:string(String, StartLoc, Options),
+ TString = String ++ ". ",
+ {ok,Ts2,End2} = scan_tokens(TString, Options, [], StartLoc),
+ {ok, Ts3, End3} =
+ scan_tokens_1({more, []}, TString, Options, [], StartLoc),
+ {end_2,End2,End3} = {end_2,End3,End2},
+ {EndLine1,EndColumn1} = End1,
+ End2 = {EndLine1,EndColumn1+2},
+ {ts_1,Ts2,Ts3} = {ts_1,Ts3,Ts2},
+ Ts2 = Ts1 ++ [lists:last(Ts2)],
+
+ %% Attributes are keylists, but have no text.
+ {ok, Ts7, End7} = erl_scan:string(String, {1,1}, Options),
+ {ok, Ts8, End8} = scan_tokens(TString, Options, [], {1,1}),
+ {end1, End1} = {end1, End7},
+ {end2, End2} = {end2, End8},
+ Ts8 = Ts7 ++ [lists:last(Ts8)],
+ {cons,true} = {cons,consistent_attributes([Ts1,Ts2,Ts3,Ts7,Ts8])},
+
+ {Ts1, End1}.
+
+scan_tokens(String, Options, Rs, Location) ->
+ case erl_scan:tokens([], String, Location, Options) of
+ {done, {ok,Ts,End}, ""} ->
+ {ok, lists:append(lists:reverse([Ts|Rs])), End};
+ {done, {ok,Ts,End}, Rest} ->
+ scan_tokens(Rest, Options, [Ts|Rs], End)
+ end.
+
+scan_tokens_1({done, {ok,Ts,End}, ""}, "", _Options, Rs, _Location) ->
+ {ok,lists:append(lists:reverse([Ts|Rs])),End};
+scan_tokens_1({done, {ok,Ts,End}, Rest}, Cs, Options, Rs, _Location) ->
+ scan_tokens_1({more,[]}, Rest++Cs, Options, [Ts|Rs], End);
+scan_tokens_1({more, Cont}, [C | Cs], Options, Rs, Loc) ->
+ R = erl_scan:tokens(Cont, [C], Loc, Options),
+ scan_tokens_1(R, Cs, Options, Rs, Loc).
+
+consistent_attributes([]) ->
+ true;
+consistent_attributes([Ts | TsL]) ->
+ L = [T || T <- Ts, is_integer(element(2, T))],
+ case L of
+ [] ->
+ TagsL = [[Tag || {Tag,_} <-
+ erl_scan:attributes_info(element(2, T))] ||
+ T <- Ts],
+ case lists:usort(TagsL) of
+ [_] ->
+ consistent_attributes(TsL);
+ [] when Ts =:= [] ->
+ consistent_attributes(TsL);
+ _ ->
+ Ts
+ end;
+ Ts ->
+ consistent_attributes(TsL);
+ _ ->
+ Ts
+ end.
+
+family_list(L) ->
+ sofs:to_external(family(L)).
+
+family(L) ->
+ sofs:relation_to_family(sofs:relation(L)).