cpp5.y

用于lex和yacc的c++及c语言的文法,可用来构造C++语言的编译器

            foo : sizeof(foo + 1);
    */

bit_field_identifier_declarator:
                                   ':' constant_expression
        | identifier_declarator {} ':' constant_expression
        ;

enum_name_elaboration:
        global_opt_scope_opt_enum_key '{' enumerator_list '}'
        | enum_name                   '{' enumerator_list '}'
        ;


    /* As with structures, the distinction between "elaborating"  and 
    "non-elaborating"  enum  types  is  maintained.  In actuality, it 
    probably does not cause much in the way of conflicts, since a ':' 
    is not allowed.  For symmetry, we maintain the distinction.   The 
    {}  actions are intended to allow the symbol table to be updated.  
    These updates are significant to code such as:

        enum A { first=sizeof(A)};
    */

enum_name:
        global_opt_scope_opt_enum_key tag_name
        ;

global_opt_scope_opt_enum_key:
        ENUM
        | global_or_scope ENUM
        ;

enumerator_list:
        enumerator_list_no_trailing_comma
        | enumerator_list_no_trailing_comma ',' /* C++, not ANSI C */
        ;


    /* Note that we do not need to rush to add an enumerator  to  the 
    symbol  table  until  *AFTER* the enumerator_value_opt is parsed. 
    The enumerated value is only in scope  AFTER  its  definition  is 
    complete.   Hence the following is legal: "enum {a, b=a+10};" but 
    the following is (assuming no external matching of names) is  not 
    legal:  "enum {c, d=sizeof(d)};" ("d" not defined when sizeof was 
    applied.) This is  notably  contrasted  with  declarators,  which 
    enter scope as soon as the declarator is complete. */

enumerator_list_no_trailing_comma:
        enumerator_name enumerator_value_opt
        | enumerator_list_no_trailing_comma ',' enumerator_name enumerator_value_opt
        ;

enumerator_name:
        IDENTIFIER
        | TYPEDEFname
        ;

enumerator_value_opt:
        /* Nothing */
        | '=' constant_expression
        ;


    /*  We special case the lone type_name which has no storage class 
    (even though it should be an example of  a  parameter_type_list). 
    This helped to disambiguate type-names in parenthetical casts.*/

parameter_type_list:
        '(' ')'                             type_qualifier_list_opt
        | '(' type_name ')'                 type_qualifier_list_opt
        | '(' type_name initializer ')'     type_qualifier_list_opt /* C++, not ANSI C */
        | '(' named_parameter_type_list ')' type_qualifier_list_opt
        ;


    /* The following are used in old style function definitions, when 
    a complex return type includes the "function returning" modifier. 
    Note  the  subtle  distinction  from  parameter_type_list.  These 
    parameters are NOT the parameters for the function being defined, 
    but are simply part of the type definition.  An example would be:

        int(*f(   a  ))(float) long a; {...}

    which is equivalent to the full new style definition:

        int(*f(long a))(float) {...}

    The   type   list    `(float)'    is    an    example    of    an 
    old_parameter_type_list.   The  bizarre point here is that an old 
    function definition declarator can be followed by  a  type  list, 
    which  can  start  with a qualifier `const'.  This conflicts with 
    the new syntactic construct for const member  functions!?!  As  a 
    result,  an  old  style function definition cannot be used in all 
    cases for a member function.  */

old_parameter_type_list:
        '(' ')'
        | '(' type_name ')'
        | '(' type_name initializer ')'  /* C++, not ANSI C */
        | '(' named_parameter_type_list ')'
        ;

named_parameter_type_list:  /* WARNING: excludes lone type_name*/
        parameter_list
        | parameter_list comma_opt_ellipsis
        | type_name comma_opt_ellipsis
        | type_name initializer comma_opt_ellipsis  /* C++, not ANSI C */
        | ELLIPSIS /* C++, not ANSI C */
        ;

comma_opt_ellipsis:
        ELLIPSIS       /* C++, not ANSI C */
        | ',' ELLIPSIS
        ;

parameter_list:
        non_casting_parameter_declaration
        | non_casting_parameter_declaration initializer /* C++, not ANSI C */
        | type_name             ',' parameter_declaration
        | type_name initializer ',' parameter_declaration  /* C++, not ANSI C */
        | parameter_list        ',' parameter_declaration
        ;


    /* There is some very subtle disambiguation going  on  here.   Do 
    not be tempted to make further use of the following production in 
    parameter_list,  or else the conflict count will grow noticeably. 
    Specifically, the next set  of  rules  has  already  been  inline 
    expanded for the first parameter in a parameter_list to support a 
    deferred disambiguation. The subtle disambiguation has to do with 
    contexts where parameter type lists look like old-style-casts. */

parameter_declaration:
        type_name
        | type_name                         initializer  /* C++, not ANSI C */
        | non_casting_parameter_declaration
        | non_casting_parameter_declaration initializer /* C++, not ANSI C */
        ;


    /* There is an LR ambiguity between old-style parenthesized casts 
    and parameter-type-lists.  This tends to happen in contexts where 
    either  an  expression or a parameter-type-list is possible.  For 
    example, assume that T is an  externally  declared  type  in  the 
    code:

           int (T ((int

    it might continue:

           int (T ((int)0));

    which would make it:

           (int) (T) (int)0 ;

    which  is  an  expression,  consisting  of  a  series  of  casts.  
    Alternatively, it could be:

           int (T ((int a)));

    which would make it the redeclaration of T, equivalent to:

           int T (dummy_name (int a));

    if we see a type that either has a named variable (in  the  above 
    case "a"), or a storage class like:

           int (T ((int register

    then  we  know  it  can't  be  a cast, and it is "forced" to be a 
    parameter_list.

    It is not yet clear that the ANSI C++ committee would  decide  to 
    place this disambiguation into the syntax, rather than leaving it 
    as  a  constraint check (i.e., a valid parser would have to parse 
    everything as though it were  a  parameter  list  (in  these  odd 
    contexts),  and  then  give an error if is to a following context 
    (like "0" above) that invalidated this syntax evaluation. */

    /* One big thing implemented here is that a TYPEDEFname CANNOT be 
    redeclared when we don't have declaration_specifiers! Notice that 
    when we do use a TYPEDEFname based declarator, only the "special" 
    (non-ambiguous  in  this  context)  typedef_declarator  is  used. 
    Everything else that is "missing" shows up as a type_name. */

non_casting_parameter_declaration: /*have names or storage classes */
        declaration_specifier
        | declaration_specifier abstract_declarator
        | declaration_specifier identifier_declarator
        | declaration_specifier parameter_typedef_declarator

        | declaration_qualifier_list
        | declaration_qualifier_list abstract_declarator
        | declaration_qualifier_list identifier_declarator

        | type_specifier identifier_declarator
        | type_specifier parameter_typedef_declarator

        | basic_type_name identifier_declarator
        | basic_type_name parameter_typedef_declarator

        | TYPEDEFname                   identifier_declarator
        | TYPEDEFname                   parameter_typedef_declarator

        | global_or_scoped_typedefname  identifier_declarator
        | global_or_scoped_typedefname  parameter_typedef_declarator

        | type_qualifier_list identifier_declarator
        ;

type_name:
        type_specifier
        | basic_type_name
        | TYPEDEFname
        | global_or_scoped_typedefname
        | type_qualifier_list

        | type_specifier               abstract_declarator
        | basic_type_name              abstract_declarator
        | TYPEDEFname                  abstract_declarator
        | global_or_scoped_typedefname abstract_declarator
        | type_qualifier_list          abstract_declarator
        ;

initializer_opt:
        /* nothing */
        | initializer
        ;

initializer:
        '=' initializer_group
        ;

initializer_group:
        '{' initializer_list '}'
        | '{' initializer_list ',' '}'
        | assignment_expression
        ;

initializer_list:
        initializer_group
        | initializer_list ',' initializer_group
        ;


/*************************** STATEMENTS *******************************/

statement:
        labeled_statement
        | compound_statement
        | expression_statement
        | selection_statement
        | iteration_statement
        | jump_statement
        | declaration /* C++, not ANSI C */
        ;

labeled_statement:
        label                      ':' statement
        | CASE constant_expression ':' statement
        | DEFAULT                  ':' statement
        ;


    /*  I sneak declarations into statement_list to support C++.  The 
    grammar is a little clumsy this  way,  but  the  violation  of  C 
    syntax is heavily localized */

compound_statement:
        '{' statement_list_opt '}'
        ;

declaration_list:
        declaration
        | declaration_list declaration
        ;

statement_list_opt:
        /* nothing */
        | statement_list_opt statement
        ;

expression_statement:
        comma_expression_opt ';'
        ;

selection_statement:
          IF '(' comma_expression ')' statement
        | IF '(' comma_expression ')' statement ELSE statement
        | SWITCH '(' comma_expression ')' statement
        ;

iteration_statement:
        WHILE '(' comma_expression_opt ')' statement
        | DO statement WHILE '(' comma_expression ')' ';'

        | FOR '(' comma_expression_opt ';' comma_expression_opt ';'
                comma_expression_opt ')' statement

        | FOR '(' declaration        comma_expression_opt ';'
                comma_expression_opt ')' statement  /* C++, not ANSI C */
        ;

jump_statement:
        GOTO label                     ';'
        | CONTINUE                     ';'
        | BREAK                        ';'
        | RETURN comma_expression_opt  ';'
        ;


    /*  The  following  actions should update the symbol table in the 
    "label" name space */

label:
        IDENTIFIER
        | TYPEDEFname
        ;


/***************************** EXTERNAL DEFINITIONS *****************************/

translation_unit:
        /* nothing */
        | translation_unit external_definition
        ;

external_definition:
        function_declaration                         /* C++, not ANSI C*/
        | function_definition
        | declaration
        | linkage_specifier function_declaration     /* C++, not ANSI C*/
        | linkage_specifier function_definition      /* C++, not ANSI C*/
        | linkage_sp