path: root/lib/parsetools/doc/src/yecc.xml

<?xml version="1.0" encoding="latin1" ?>
<!DOCTYPE erlref SYSTEM "erlref.dtd">

<erlref>
  <header>
    <copyright>
      <year>1996</year><year>2011</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 D&auml;cker</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 =&nbsp;-&nbsp;see below&nbsp;-</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 =&nbsp;-&nbsp;formatable by format_error/1&nbsp;-</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 =&nbsp;-&nbsp;as returned by yecc:file/1,2&nbsp;-</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 &amp; Johnson: 'LR Parsing', ACM Computing Surveys, vol. 6:2, 1974.</p>
  </section>
</erlref>