<?xml version="1.0" encoding="latin1" ?>
<!DOCTYPE erlref SYSTEM "erlref.dtd">
<erlref>
<header>
<copyright>
<year>1996</year><year>2009</year>
<holder>Ericsson AB. All Rights Reserved.</holder>
</copyright>
<legalnotice>
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.
</legalnotice>
<title>yecc</title>
<prepared>Carl Wilhelm Welin</prepared>
<responsible>Carl Wilhelm Welin</responsible>
<docno></docno>
<approved>Bjarne Däcker</approved>
<checked></checked>
<date>1997-01-27</date>
<rev>B</rev>
<file>yecc.sgml</file>
</header>
<module>yecc</module>
<modulesummary>LALR-1 Parser Generator</modulesummary>
<description>
<p>An LALR-1 parser generator for Erlang, similar to <c>yacc</c>.
Takes a BNF grammar definition as input, and produces Erlang code
for a parser. </p>
<p>To understand this text, you also have to
look at the <c>yacc</c> documentation in the UNIX(TM) manual. This
is most probably necessary in order to understand the idea of a
parser generator, and the principle and problems of LALR parsing
with finite look-ahead.</p>
</description>
<funcs>
<func>
<name>file(Grammarfile [, Options]) -> YeccRet</name>
<fsummary>Give information about resolved and unresolved parse action conflicts.</fsummary>
<type>
<v>Grammarfile = filename()</v>
<v>Options = Option | [Option]</v>
<v>Option = - see below -</v>
<v>YeccRet = {ok, Parserfile} | {ok, Parserfile, Warnings} | error | {error, Warnings, Errors}</v>
<v>Parserfile = filename()</v>
<v>Warnings = Errors = [{filename(), [ErrorInfo]}]</v>
<v>ErrorInfo = {ErrorLine, module(), Reason}</v>
<v>ErrorLine = integer()</v>
<v>Reason = - formatable by format_error/1 -</v>
</type>
<desc>
<p><c>Grammarfile</c> is the file of declarations and grammar
rules. Returns <c>ok</c> upon success, or <c>error</c> if
there are errors. An Erlang file containing the parser is
created if there are no errors. The options are:
</p>
<taglist>
<tag><c>{parserfile, Parserfile}</c>.</tag>
<item><c>Parserfile</c> is the name of the file that will
contain the Erlang parser code that is generated. The
default (<c>""</c>) is to add the extension <c>.erl</c>
to <c>Grammarfile</c> stripped of the <c>.yrl</c>
extension.
</item>
<tag><c>{includefile, Includefile}</c>.</tag>
<item>Indicates a customized prologue file which the user
may want to use instead of the default file
<c>lib/parsetools/include/yeccpre.hrl</c> which is
otherwise included at the beginning of the resulting
parser file. <em>N.B.</em> The <c>Includefile</c> is
included 'as is' in the parser file, so it must not have a
module declaration of its own, and it should not be
compiled. It must, however, contain the necessary export
declarations. The default is indicated by <c>""</c>.
</item>
<tag><c>{report_errors, bool()}</c>.</tag>
<item>Causes errors to be printed as they occur. Default is
<c>true</c>.
</item>
<tag><c>{report_warnings, bool()}</c>.</tag>
<item>Causes warnings to be printed as they occur. Default is
<c>true</c>.
</item>
<tag><c>{report, bool()}</c>.</tag>
<item>This is a short form for both <c>report_errors</c> and
<c>report_warnings</c>.
</item>
<tag><c>warnings_as_errors</c></tag>
<item>
<p>Causes warnings to be treated as errors.</p>
</item>
<tag><c>{return_errors, bool()}</c>.</tag>
<item>If this flag is set, <c>{error, Errors, Warnings}</c>
is returned when there are errors. Default is
<c>false</c>.
</item>
<tag><c>{return_warnings, bool()}</c>.</tag>
<item>If this flag is set, an extra field containing
<c>Warnings</c> is added to the tuple returned upon
success. Default is <c>false</c>.
</item>
<tag><c>{return, bool()}</c>.</tag>
<item>This is a short form for both <c>return_errors</c> and
<c>return_warnings</c>.
</item>
<tag><c>{verbose, bool()}</c>. </tag>
<item>Determines whether the parser generator should give
full information about resolved and unresolved parse
action conflicts (<c>true</c>), or only about those
conflicts that prevent a parser from being generated from
the input grammar (<c>false</c>, the default).
</item>
</taglist>
<p>Any of the Boolean options can be set to <c>true</c> by
stating the name of the option. For example, <c>verbose</c>
is equivalent to <c>{verbose, true}</c>.
</p>
<p>The value of the <c>Parserfile</c> option stripped of the
<c>.erl</c> extension is used by Yecc as the module name of
the generated parser file.</p>
<p>Yecc will add the extension <c>.yrl</c> to the
<c>Grammarfile</c> name, the extension <c>.hrl</c> to the
<c>Includefile</c> name, and the extension <c>.erl</c> to
the <c>Parserfile</c> name, unless the extension is already
there.</p>
</desc>
</func>
<func>
<name>format_error(Reason) -> Chars</name>
<fsummary>Return an English description of a an error tuple.</fsummary>
<type>
<v>Reason = - as returned by yecc:file/1,2 -</v>
<v>Chars = [char() | Chars]</v>
</type>
<desc>
<p>Returns a descriptive string in English of an error tuple
returned by <c>yecc:file/1,2</c>. This function is mainly
used by the compiler invoking Yecc.</p>
</desc>
</func>
</funcs>
<section>
<title>Pre-Processing</title>
<p>A <c>scanner</c> to pre-process the text (program, etc.) to be
parsed is not provided in the <c>yecc</c> module. The scanner
serves as a kind of lexicon look-up routine. It is possible to
write a grammar that uses only character tokens as terminal
symbols, thereby eliminating the need for a scanner, but this
would make the parser larger and slower.</p>
<p>The user should implement a scanner that segments the input
text, and turns it into one or more lists of tokens. Each token
should be a tuple containing information about syntactic
category, position in the text (e.g. line number), and the
actual terminal symbol found in the text: <c>{Category, LineNumber, Symbol}</c>.</p>
<p>If a terminal symbol is the only member of a category, and the
symbol name is identical to the category name, the token format
may be <c>{Symbol, LineNumber}</c>.</p>
<p>A list of tokens produced by the scanner should end with a
special <c>end_of_input</c> tuple which the parser is looking
for. The format of this tuple should be <c>{Endsymbol, LastLineNumber}</c>, where <c>Endsymbol</c> is an identifier
that is distinguished from all the terminal and non-terminal
categories of the syntax rules. The <c>Endsymbol</c> may be
declared in the grammar file (see below).</p>
<p>The simplest case is to segment the input string into a list of
identifiers (atoms) and use those atoms both as categories and
values of the tokens. For example, the input string <c>aaa bbb 777, X</c> may be scanned (tokenized) as:</p>
<code type="none">
[{aaa, 1}, {bbb, 1}, {777, 1}, {',' , 1}, {'X', 1},
{'$end', 1}]. </code>
<p>This assumes that this is the first line of the input text, and
that <c>'$end'</c> is the distinguished <c>end_of_input</c>
symbol.</p>
<p>The Erlang scanner in the <c>io</c> module can be used as a
starting point when writing a new scanner. Study
<c>yeccscan.erl</c> in order to see how a filter can be added on
top of <c>io:scan_erl_form/3</c> to provide a scanner for
Yecc that tokenizes grammar files before parsing them
with the Yecc parser. A more general approach to scanner
implementation is to use a scanner generator. A scanner
generator in Erlang called <c>leex</c> is under development.</p>
</section>
<section>
<title>Grammar Definition Format</title>
<p>Erlang style <c>comments</c>, starting with a <c>'%'</c>, are
allowed in grammar files.</p>
<p>Each <c>declaration</c> or <c>rule</c> ends with a dot (the
character <c>'.'</c>).</p>
<p>The grammar starts with an optional <c>header</c> section. The
header is put first in the generated file, before the module
declaration. The purpose of the header is to provide a means to
make the documentation generated by <c>EDoc</c> look nicer. Each
header line should be enclosed in double quotes, and newlines
will be inserted between the lines. For example:</p>
<code>
Header "%% Copyright (C)"
"%% @private"
"%% @Author John"</code>
<p>Next comes a declaration of the <c>nonterminal categories</c>
to be used in the rules. For example:</p>
<code type="none">
Nonterminals sentence nounphrase verbphrase. </code>
<p>A non-terminal category can be used at the left hand side (=
<c>lhs</c>, or <c>head</c>) of a grammar rule. It can also
appear at the right hand side of rules.</p>
<p>Next comes a declaration of the <c>terminal categories</c>,
which are the categories of tokens produced by the scanner. For
example:</p>
<code type="none">
Terminals article adjective noun verb. </code>
<p>Terminal categories may only appear in the right hand sides (=
<c>rhs</c>) of grammar rules.</p>
<p>Next comes a declaration of the <c>rootsymbol</c>, or start
category of the grammar. For example:</p>
<code type="none">
Rootsymbol sentence. </code>
<p>This symbol should appear in the lhs of at least one grammar
rule. This is the most general syntactic category which the
parser ultimately will parse every input string into.</p>
<p>After the rootsymbol declaration comes an optional declaration
of the <c>end_of_input</c> symbol that your scanner is expected
to use. For example:</p>
<code type="none">
Endsymbol '$end'. </code>
<p>Next comes one or more declarations of <c>operator precedences</c>, if needed. These are used to resolve
shift/reduce conflicts (see <c>yacc</c> documentation).</p>
<p>Examples of operator declarations:</p>
<code type="none">
Right 100 '='.
Nonassoc 200 '==' '=/='.
Left 300 '+'.
Left 400 '*'.
Unary 500 '-'. </code>
<p>These declarations mean that <c>'='</c> is defined as a
<c>right associative binary</c> operator with precedence 100,
<c>'=='</c> and <c>'=/='</c> are operators with <c>no associativity</c>, <c>'+'</c> and <c>'*'</c> are <c>left associative binary</c> operators, where <c>'*'</c> takes
precedence over <c>'+'</c> (the normal case), and <c>'-'</c> is
a <c>unary</c> operator of higher precedence than <c>'*'</c>.
The fact that '==' has no associativity means that an expression
like <c>a == b == c</c> is considered a syntax error.</p>
<p>Certain rules are assigned precedence: each rule gets its
precedence from the last terminal symbol mentioned in the right
hand side of the rule. It is also possible to declare precedence
for non-terminals, "one level up". This is practical when an
operator is overloaded (see also example 3 below).</p>
<p>Next come the <c>grammar rules</c>. Each rule has the general
form</p>
<code type="none">
Left_hand_side -> Right_hand_side : Associated_code. </code>
<p>The left hand side is a non-terminal category. The right hand
side is a sequence of one or more non-terminal or terminal
symbols with spaces between. The associated code is a sequence
of zero or more Erlang expressions (with commas <c>','</c> as
separators). If the associated code is empty, the separating
colon <c>':'</c> is also omitted. A final dot marks the end of
the rule.</p>
<p>Symbols such as <c>'{'</c>, <c>'.'</c>, etc., have to be
enclosed in single quotes when used as terminal or non-terminal
symbols in grammar rules. The use of the symbols
<c>'$empty'</c>, <c>'$end'</c>, and <c>'$undefined'</c> should
be avoided.</p>
<p>The last part of the grammar file is an optional section with
Erlang code (= function definitions) which is included 'as is'
in the resulting parser file. This section must start with the
pseudo declaration, or key words</p>
<code type="none">
Erlang code. </code>
<p>No syntax rule definitions or other declarations may follow
this section. To avoid conflicts with internal variables, do not
use variable names beginning with two underscore characters
('__') in the Erlang code in this section, or in the code
associated with the individual syntax rules.</p>
<p>The optional <c>expect</c> declaration can be placed anywhere
before the last optional section with Erlang code. It is used
for suppressing the warning about conflicts that is ordinarily
given if the grammar is ambiguous. An example:</p>
<code type="none">
Expect 2. </code>
<p>The warning is given if the number of shift/reduce conflicts
differs from 2, or if there are reduce/reduce conflicts.
</p>
</section>
<section>
<title>Examples</title>
<p>A grammar to parse list expressions (with empty associated
code):</p>
<code type="none">
Nonterminals list elements element.
Terminals atom '(' ')'.
Rootsymbol list.
list -> '(' ')'.
list -> '(' elements ')'.
elements -> element.
elements -> element elements.
element -> atom.
element -> list. </code>
<p>This grammar can be used to generate a parser which parses list
expressions, such as <c>(), (a), (peter charles), (a (b c) d (())), ...</c> provided that your scanner tokenizes, for
example, the input <c>(peter charles)</c> as follows:</p>
<code type="none">
[{'(', 1} , {atom, 1, peter}, {atom, 1, charles}, {')', 1},
{'$end', 1}] </code>
<p>When a grammar rule is used by the parser to parse (part of)
the input string as a grammatical phrase, the associated code is
evaluated, and the value of the last expression becomes the
value of the parsed phrase. This value may be used by the parser
later to build structures that are values of higher phrases of
which the current phrase is a part. The values initially
associated with terminal category phrases, i.e. input tokens,
are the token tuples themselves.</p>
<p>Below is an example of the grammar above with structure
building code added:</p>
<code type="none">
list -> '(' ')' : nil.
list -> '(' elements ')' : '$2'.
elements -> element : {cons, '$1', nil}.
elements -> element elements : {cons, '$1', '$2'}.
element -> atom : '$1'.
element -> list : '$1'. </code>
<p>With this code added to the grammar rules, the parser produces
the following value (structure) when parsing the input string
<c>(a b c).</c>. This still assumes that this was the first
input line that the scanner tokenized:</p>
<code type="none">
{cons, {atom, 1, a,} {cons, {atom, 1, b},
{cons, {atom, 1, c}, nil}}} </code>
<p>The associated code contains <c>pseudo variables</c><c>'$1'</c>, <c>'$2'</c>, <c>'$3'</c>, etc. which refer to (are
bound to) the values associated previously by the parser with
the symbols of the right hand side of the rule. When these
symbols are terminal categories, the values are token tuples of
the input string (see above).</p>
<p>The associated code may not only be used to build structures
associated with phrases, but may also be used for syntactic and
semantic tests, printout actions (for example for tracing), etc.
during the parsing process. Since tokens contain positional
(line number) information, it is possible to produce error
messages which contain line numbers. If there is no associated
code after the right hand side of the rule, the value
<c>'$undefined'</c> is associated with the phrase.</p>
<p>The right hand side of a grammar rule may be empty. This is
indicated by using the special symbol <c>'$empty'</c> as rhs.
Then the list grammar above may be simplified to:</p>
<code type="none">
list -> '(' elements ')' : '$2'.
elements -> element elements : {cons, '$1', '$2'}.
elements -> '$empty' : nil.
element -> atom : '$1'.
element -> list : '$1'. </code>
</section>
<section>
<title>Generating a Parser</title>
<p>To call the parser generator, use the following command:</p>
<code type="none">
yecc:file(Grammarfile). </code>
<p>An error message from Yecc will be shown if the grammar
is not of the LALR type (for example too ambiguous).
Shift/reduce conflicts are resolved in favor of shifting if
there are no operator precedence declarations. Refer to the
<c>yacc</c> documentation on the use of operator precedence.</p>
<p>The output file contains Erlang source code for a parser module
with module name equal to the <c>Parserfile</c> parameter. After
compilation, the parser can be called as follows (the module
name is assumed to be <c>myparser</c>):</p>
<code type="none">
myparser:parse(myscanner:scan(Inport)) </code>
<p>The call format may be different if a customized prologue file
has been included when generating the parser instead of the
default file <c>lib/parsetools/include/yeccpre.hrl</c>.</p>
<p>With the standard prologue, this call will return either
<c>{ok, Result}</c>, where <c>Result</c> is a structure that the
Erlang code of the grammar file has built, or <c>{error, {Line_number, Module, Message}}</c> if there was a syntax error
in the input.</p>
<p><c>Message</c> is something which may be converted into a
string by calling <c>Module:format_error(Message)</c>
and printed with <c>io:format/3</c>.</p>
<note>
<p>By default, the parser that was generated will not print out
error messages to the screen. The user will have to do this
either by printing the returned error messages, or by
inserting tests and print instructions in the Erlang code
associated with the syntax rules of the grammar file.</p>
</note>
<p>It is also possible to make the parser ask for more input
tokens when needed if the following call format is used:</p>
<code type="none">
myparser:parse_and_scan({Function, Args})
myparser:parse_and_scan({Mod, Tokenizer, Args}) </code>
<p>The tokenizer <c>Function</c> is either a fun or a tuple
<c>{Mod, Tokenizer}</c>. The call <c>apply(Function, Args)</c>
or <c>apply({Mod, Tokenizer}, Args)</c> is executed whenever a
new token is needed. This, for example, makes it possible to
parse from a file, token by token.</p>
<p>The tokenizer used above has to be implemented so as to return
one of the following:</p>
<code type="none">
{ok, Tokens, Endline}
{eof, Endline}
{error, Error_description, Endline} </code>
<p>This conforms to the format used by the scanner in the Erlang
<c>io</c> library module.</p>
<p>If <c>{eof, Endline}</c> is returned immediately, the call to
<c>parse_and_scan/1</c> returns <c>{ok, eof}</c>. If <c>{eof, Endline}</c> is returned before the parser expects end of input,
<c>parse_and_scan/1</c> will, of course, return an error message
(see above). Otherwise <c>{ok, Result}</c> is returned.</p>
</section>
<section>
<title>More Examples</title>
<p>1. A grammar for parsing infix arithmetic expressions into
prefix notation, without operator precedence:</p>
<code type="none">
Nonterminals E T F.
Terminals '+' '*' '(' ')' number.
Rootsymbol E.
E -> E '+' T: ['$2', '$1', '$3'].
E -> T : '$1'.
T -> T '*' F: ['$2', '$1', '$3'].
T -> F : '$1'.
F -> '(' E ')' : '$2'.
F -> number : '$1'. </code>
<p>2. The same with operator precedence becomes simpler:</p>
<code type="none">
Nonterminals E.
Terminals '+' '*' '(' ')' number.
Rootsymbol E.
Left 100 '+'.
Left 200 '*'.
E -> E '+' E : ['$2', '$1', '$3'].
E -> E '*' E : ['$2', '$1', '$3'].
E -> '(' E ')' : '$2'.
E -> number : '$1'. </code>
<p>3. An overloaded minus operator:</p>
<code type="none">
Nonterminals E uminus.
Terminals '*' '-' number.
Rootsymbol E.
Left 100 '-'.
Left 200 '*'.
Unary 300 uminus.
E -> E '-' E.
E -> E '*' E.
E -> uminus.
E -> number.
uminus -> '-' E. </code>
<p>4. The Yecc grammar that is used for parsing grammar
files, including itself:</p>
<code type="none">
Nonterminals
grammar declaration rule head symbol symbols attached_code
token tokens.
Terminals
atom float integer reserved_symbol reserved_word string char var
'->' ':' dot.
Rootsymbol grammar.
Endsymbol '$end'.
grammar -> declaration : '$1'.
grammar -> rule : '$1'.
declaration -> symbol symbols dot: {'$1', '$2'}.
rule -> head '->' symbols attached_code dot: {rule, ['$1' | '$3'],
'$4'}.
head -> symbol : '$1'.
symbols -> symbol : ['$1'].
symbols -> symbol symbols : ['$1' | '$2'].
attached_code -> ':' tokens : {erlang_code, '$2'}.
attached_code -> '$empty' : {erlang_code,
[{atom, 0, '$undefined'}]}.
tokens -> token : ['$1'].
tokens -> token tokens : ['$1' | '$2'].
symbol -> var : value_of('$1').
symbol -> atom : value_of('$1').
symbol -> integer : value_of('$1').
symbol -> reserved_word : value_of('$1').
token -> var : '$1'.
token -> atom : '$1'.
token -> float : '$1'.
token -> integer : '$1'.
token -> string : '$1'.
token -> char : '$1'.
token -> reserved_symbol : {value_of('$1'), line_of('$1')}.
token -> reserved_word : {value_of('$1'), line_of('$1')}.
token -> '->' : {'->', line_of('$1')}.
token -> ':' : {':', line_of('$1')}.
Erlang code.
value_of(Token) ->
element(3, Token).
line_of(Token) ->
element(2, Token). </code>
<note>
<p>The symbols <c>'->'</c>, and <c>':'</c> have to be treated in
a special way, as they are meta symbols of the grammar
notation, as well as terminal symbols of the Yecc
grammar.</p>
</note>
<p>5. The file <c>erl_parse.yrl</c> in the <c>lib/stdlib/src</c>
directory contains the grammar for Erlang.</p>
<note>
<p>Syntactic tests are used in the code associated with some
rules, and an error is thrown (and caught by the generated
parser to produce an error message) when a test fails. The
same effect can be achieved with a call to
<c>return_error(Error_line, Message_string)</c>, which is
defined in the <c>yeccpre.hrl</c> default header file.</p>
</note>
</section>
<section>
<title>Files</title>
<code type="none">
lib/parsetools/include/yeccpre.hrl </code>
</section>
<section>
<title>See Also</title>
<p>Aho & Johnson: 'LR Parsing', ACM Computing Surveys, vol. 6:2, 1974.</p>
</section>
</erlref>
