X-Git-Url: https://git.draconx.ca/gitweb/cdecl99.git/blobdiff_plain/0d567b471af91e2627c94ab698a2a6d9597ba868..21bc2db0ec83b235caec38d42f4d0812f473d766:/src/parse.y diff --git a/src/parse.y b/src/parse.y index f7ef80c..ba9bf22 100644 --- a/src/parse.y +++ b/src/parse.y @@ -1,6 +1,7 @@ +%code top { /* * Parser for C declarations. - * Copyright © 2011 Nick Bowler + * Copyright © 2011-2012, 2021, 2023 Nick Bowler * * This program is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -15,33 +16,50 @@ * You should have received a copy of the GNU General Public License * along with this program. If not, see . */ +} +%name-prefix "cdecl__yy" +%parse-param {void *scanner} %parse-param {struct cdecl **out} +%lex-param {yyscan_t scanner} %define api.pure %error-verbose %locations %{ +#include #include +#include #include "scan.h" #include "cdecl.h" +#include "cdecl-internal.h" +#include "errmsg.h" #define FAIL(msg) do { \ - yyerror(&yylloc, NULL, msg); \ + yyerror(&yylloc, NULL, NULL, msg); \ YYERROR; \ } while (0) #define ALLOC(ptr, size) do { \ (ptr) = malloc(size); \ - if (!(ptr)) \ - FAIL("failed to allocate memory"); \ + if (!(ptr)) { \ + cdecl__errmsg(CDECL__ENOMEM); \ + YYERROR; \ + } \ } while (0) #define ALLOC_STRUCT(ptr, type, ...) do { \ ALLOC(ptr, sizeof (type)); \ *(ptr) = (type) { __VA_ARGS__ }; \ } while (0) + +/* + * With the postprocessing performed by fix-yytname.awk, all the symbol + * name strings can be used directly in error messages and there is no + * need for any string processing. + */ +#define yytnamerr(a, b) cdecl__strlcpy(a, b, (a) ? INT_MAX : 0) %} %code requires { @@ -49,25 +67,30 @@ } %code provides { -void yyerror(YYLTYPE *, struct cdecl **, const char *); -int yyparse(struct cdecl **out); +void cdecl__free(struct cdecl *); +int cdecl__yyparse(void *scanner, struct cdecl **out); +const char *cdecl__token_name(unsigned token); } %union { uintmax_t uintval; - char *strval; + unsigned spectype; + _Bool boolval; struct cdecl_declspec *declspec; struct cdecl_declarator *declarator; struct cdecl *decl; + struct parse_item *item; } %{ +static void yyerror(YYLTYPE *, yyscan_t, struct cdecl **, const char *); +static void free_decl(struct cdecl *); + static void free_declspec(struct cdecl_declspec *x) { struct cdecl_declspec *p; while (x) { p = x->next; - free(x->ident); free(x); x = p; } @@ -76,51 +99,100 @@ static void free_declspec(struct cdecl_declspec *x) static void free_declarator(struct cdecl_declarator *x) { struct cdecl_declarator *p; + while (x) { - p = x->next; + p = x->child; + switch (x->type) { + case CDECL_DECL_NULL: + x = NULL; case CDECL_DECL_IDENT: - free(x->u.ident); + case CDECL_DECL_ARRAY: break; case CDECL_DECL_POINTER: free_declspec(x->u.pointer.qualifiers); - free_declarator(x->u.pointer.declarator); break; - case CDECL_DECL_ARRAY: - free(x->u.array.vla); - free_declarator(x->u.array.declarator); + case CDECL_DECL_FUNCTION: + free_decl(x->u.function.parameters); break; default: assert(0); } + free(x); x = p; } } -static void free_decl(struct cdecl *decl) +static void free_decl(struct cdecl *x) { - free_declspec(decl->specifiers); - free_declarator(decl->declarators); - free(decl); + struct cdecl *p; + + while (x) { + p = x->next; + + /* The specifiers may be shared by an entire chain. */ + if (!p || p->specifiers != x->specifiers) + free_declspec(x->specifiers); + + free_declarator(x->declarators); + free(x); + x = p; + } } -void cdecl_free(struct cdecl *decl) +void cdecl__free(struct cdecl *decl) { - if (decl) - free_decl(decl); + free_decl(decl); } + +/* + * Join two declaration specifier lists into a single list, with "a" being the + * head of the new list. + * + * The list "a" is assumed to be nonempty. + */ +static void join_specs(struct cdecl_declspec *a, struct cdecl_declspec *b) +{ + while (a->next) + a = a->next; + a->next = b; +} + +/* + * Alter an abstract declarator (type name) to declare an identifier instead, + * used by the English parser rules to reduce "identifier as type" sequences. + */ +static struct cdecl *insert_identifier(struct cdecl *decl, struct parse_item *ident) +{ + struct cdecl_declarator *d, **p = &decl->declarators; + + while ((d = *p)->child) + p = &d->child; + + *p = &ident->u.declarator; + return decl; +} + +static struct cdecl_declarator *nulldecl(void) +{ + static const struct cdecl_declarator nulldecl = {0}; + return (void *)&nulldecl; +} +#define NULLDECL (nulldecl()) + %} -%destructor { free($$); } +%destructor { free($$); } %destructor { free_declspec($$); } %destructor { free_declarator($$); } %destructor { free_decl($$); } +/* Magic tokens */ %token T_LEX_ERROR -%token T_IDENT "identifier" -%token T_UINT "integer constant" +%token T_IDENT "identifier" +%token T_UINT "integer constant" %token T_SEMICOLON ";" %token T_ASTERISK "*" @@ -129,55 +201,107 @@ void cdecl_free(struct cdecl *decl) %token T_LBRACKET "[" %token T_RBRACKET "]" %token T_COMMA "," +%token T_ELLIPSIS "..." + +%token T_TYPEDEF "typedef" +%token T_EXTERN "extern" +%token T_STATIC "static" +%token T_AUTO "auto" +%token T_REGISTER "register" + +%token T_INLINE "inline" + +%token T_RESTRICT "restrict" +%token T_VOLATILE "volatile" +%token T_CONST "const" + +%token T_VOID "void" +%token T_CHAR "char" +%token T_SHORT "short" +%token T_INT "int" +%token T_LONG "long" +%token T_FLOAT "float" +%token T_DOUBLE "double" +%token T_SIGNED "signed" +%token T_UNSIGNED "unsigned" +%token T_BOOL "_Bool" +%token T_COMPLEX "_Complex" +%token T_IMAGINARY "_Imaginary" + +%token T_STRUCT "struct" +%token T_UNION "union" +%token T_ENUM "enum" -%token T_TYPEDEF "typedef" -%token T_EXTERN "extern" -%token T_STATIC "static" -%token T_AUTO "auto" -%token T_REGISTER "register" - -%token T_INLINE "inline" - -%token T_RESTRICT "restrict" -%token T_VOLATILE "volatile" -%token T_CONST "const" - -%token T_VOID "void" -%token T_CHAR "char" -%token T_SHORT "short" -%token T_INT "int" -%token T_LONG "long" -%token T_FLOAT "float" -%token T_DOUBLE "double" -%token T_SIGNED "signed" -%token T_UNSIGNED "unsigned" -%token T_BOOL "_Bool" -%token T_COMPLEX "_Complex" - -%token T_STRUCT "struct" -%token T_UNION "union" -%token T_ENUM "enum" - -%type vla_ident -%type declspec_simple typespec_simple qualifier_simple +/* + * English keywords. + */ +%token T_TYPE "type" +%token T_DECLARE "declare" +%token T_POINTER "pointer" +%token T_FUNCTION "function" +%token T_RETURNING "returning" +%token T_ARRAY "array" +%token T_TO "to" +%token T_OF "of" +%token T_AS "as" +%token T_VLA "variable-length" + +%type vla_ident +%type array_length +%type varargs +%type declspec_simple qualifier_simple +%type typespec_simple typespec_tagged %type declspec_notype declspec_noid typespec_noid typespec -%type qualifier qualifiers pointer +%type qualifier qualifiers %type declspecs declspecs_noid -%type direct_declarator declarator declarators array -%type declaration +%type direct_declarator declarator pointer array parens postfix +%type direct_declarator_ish declarator_ish parameter_type_list +%type declaration declarators declarator_wrap +%type parameter + +%type english_vla +%type storage_func_specs post_specs +%type type_qual_spec type_qual_specs typedef_name_qual +%type english_declarator english_array english_function +%type english_parameter_list null_decl +%type english english_declaration +%type english_parameter + +/* Precedence declaration to avoid conflict in english_parameter; see below. */ +%right T_TYPE +%right T_IDENT %% -input: declaration { +input: english { + *out = $1; +} | declaration { *out = $1; }; -declaration: declspecs declarators T_SEMICOLON { - ALLOC_STRUCT($$, struct cdecl, - .specifiers = $1, - .declarators = $2); +semi: | T_SEMICOLON + +declaration: declspecs declarators semi { + $$ = $2; + $$->specifiers = $1; }; +/* + * We support parsing declarations using arbitrary identifiers as type + * specifiers (a la C typedef). To avoid confusion with identifiers that + * may also be used as declarators, note the following: + * + * (a) Every valid C declaration must have at least one type specifier, and + * (b) Valid declarations with typedef names have exactly one type specifier. + * + * So the rule applied when parsing specifiers is: an identifier is a type + * specifier only if we have not yet seen any type specifiers whatsoever + * (within one declaration specifier list). + * + * Treating identifiers as type specifiers by default can lead to strange and + * unexpected parses; libcdecl applies a simplification step to the resulting + * parse tree afterwards. + */ declspecs: declspec_notype declspecs { $$ = $1; $$->next = $2; @@ -196,33 +320,40 @@ qualifiers: { $$ = NULL; } | qualifiers qualifier { $$->next = $1; } -declarators: declarator | declarator T_COMMA declarators { +declarators: declarator_wrap | declarator_wrap T_COMMA declarators { $$ = $1; $$->next = $3; -}; +} -declspec_simple: T_AUTO { $$ = CDECL_STOR_AUTO; } - | T_TYPEDEF { $$ = CDECL_STOR_TYPEDEF; } - | T_EXTERN { $$ = CDECL_STOR_EXTERN; } - | T_STATIC { $$ = CDECL_STOR_STATIC; } - | T_REGISTER { $$ = CDECL_STOR_REGISTER; } - | T_INLINE { $$ = CDECL_FUNC_INLINE; } - -typespec_simple: T_VOID { $$ = CDECL_TYPE_VOID; } - | T_CHAR { $$ = CDECL_TYPE_CHAR; } - | T_SHORT { $$ = CDECL_TYPE_SHORT; } - | T_INT { $$ = CDECL_TYPE_INT; } - | T_LONG { $$ = CDECL_TYPE_LONG; } - | T_FLOAT { $$ = CDECL_TYPE_FLOAT; } - | T_DOUBLE { $$ = CDECL_TYPE_DOUBLE; } - | T_SIGNED { $$ = CDECL_TYPE_SIGNED; } - | T_UNSIGNED { $$ = CDECL_TYPE_UNSIGNED; } - | T_BOOL { $$ = CDECL_TYPE_BOOL; } - | T_COMPLEX { $$ = CDECL_TYPE_COMPLEX; } - -qualifier_simple: T_CONST { $$ = CDECL_QUAL_CONST; } - | T_RESTRICT { $$ = CDECL_QUAL_RESTRICT; } - | T_VOLATILE { $$ = CDECL_QUAL_VOLATILE; } +declarator_wrap: declarator { + ALLOC_STRUCT($$, struct cdecl, .declarators = $1); +} + +declspec_simple: T_AUTO + | T_TYPEDEF + | T_EXTERN + | T_STATIC + | T_REGISTER + | T_INLINE + +typespec_simple: T_VOID + | T_CHAR + | T_SHORT + | T_INT + | T_LONG + | T_FLOAT + | T_DOUBLE + | T_SIGNED + | T_UNSIGNED + | T_BOOL + | T_COMPLEX + | T_IMAGINARY + +typespec_tagged: T_STRUCT | T_UNION | T_ENUM | { $$ = CDECL_TYPE_IDENT; } + +qualifier_simple: T_CONST + | T_RESTRICT + | T_VOLATILE declspec_notype: qualifier | declspec_simple { ALLOC_STRUCT($$, struct cdecl_declspec, .type = $1); @@ -236,76 +367,247 @@ qualifier: qualifier_simple { ALLOC_STRUCT($$, struct cdecl_declspec, .type = $1); } -typespec: typespec_noid | T_STRUCT T_IDENT { - ALLOC_STRUCT($$, struct cdecl_declspec, - .type = CDECL_TYPE_STRUCT, - .ident = $2); -} | T_UNION T_IDENT { - ALLOC_STRUCT($$, struct cdecl_declspec, - .type = CDECL_TYPE_UNION, - .ident = $2); -} | T_ENUM T_IDENT { - ALLOC_STRUCT($$, struct cdecl_declspec, - .type = CDECL_TYPE_ENUM, - .ident = $2); -} | T_IDENT { - ALLOC_STRUCT($$, struct cdecl_declspec, - .type = CDECL_TYPE_IDENT, - .ident = $1); +typespec: typespec_noid | typespec_tagged T_IDENT { + /* Compiler should be able to elide this assignment. */ + $2->u.declspec.ident = $2->u.declarator.u.ident; + + $$ = &$2->u.declspec; + $$->type = $1; } declspec_noid: declspec_notype | typespec_noid -pointer: T_ASTERISK qualifiers { $$ = $2; } - vla_ident: T_IDENT | T_ASTERISK { - ALLOC($$, sizeof ""); - strcpy($$, ""); + if (!($$ = cdecl__alloc_item(1))) + YYERROR; + *$$->s = 0; } -array: T_LBRACKET T_UINT T_RBRACKET { - if ($2 == 0) - FAIL("array length must be positive"); - +array: T_LBRACKET array_length T_RBRACKET { ALLOC_STRUCT($$, struct cdecl_declarator, .type = CDECL_DECL_ARRAY, .u.array.length = $2); } | T_LBRACKET vla_ident T_RBRACKET { + $$ = &$2->u.declarator; + $$->type = CDECL_DECL_ARRAY; + $$->u.array.vla = $$->u.ident; + $$->u.array.length = 0; +} + +parameter: declspecs declarator { + ALLOC_STRUCT($$, struct cdecl, + .specifiers = $1, + .declarators = $2); +} + +varargs: { $$ = false; } | T_COMMA T_ELLIPSIS { $$ = true; } + +parameter_type_list: parameter varargs { ALLOC_STRUCT($$, struct cdecl_declarator, - .type = CDECL_DECL_ARRAY, - .u.array.vla = $2); -} | T_LBRACKET T_RBRACKET { + .type = CDECL_DECL_FUNCTION, + .u.function.parameters = $1, + .u.function.variadic = $2); +} | parameter T_COMMA parameter_type_list { + $$ = $3; + $1->next = $$->u.function.parameters; + $$->u.function.parameters = $1; +} + +parens: T_LPAREN parameter_type_list T_RPAREN { + $$ = $2; +} | T_LPAREN declarator_ish T_RPAREN { ALLOC_STRUCT($$, struct cdecl_declarator, - .type = CDECL_DECL_ARRAY); + .type = CDECL_DECL_FUNCTION); + ALLOC_STRUCT($$->u.function.parameters, struct cdecl, + .declarators = $2); } -declarator: direct_declarator | pointer direct_declarator { +pointer: T_ASTERISK qualifiers direct_declarator { + ALLOC_STRUCT($$, struct cdecl_declarator, + .type = CDECL_DECL_POINTER, + .u.pointer.qualifiers = $2, + .child = $3); +} | T_ASTERISK qualifiers pointer { ALLOC_STRUCT($$, struct cdecl_declarator, .type = CDECL_DECL_POINTER, - .u.pointer.qualifiers = $1, - .u.pointer.declarator = $2); + .u.pointer.qualifiers = $2, + .child = $3); +} + +declarator: direct_declarator | pointer +declarator_ish: direct_declarator_ish | pointer +postfix: array | parens + +direct_declarator_ish: { + $$ = NULLDECL; +} | direct_declarator_ish postfix { + $$ = $2; + $$->child = $1; } direct_declarator: { - ALLOC_STRUCT($$, struct cdecl_declarator, - .type = CDECL_DECL_IDENT, - .u.ident = NULL); + $$ = NULLDECL; } | T_IDENT { + $$ = &$1->u.declarator; +} | direct_declarator postfix { + $$ = $2; + $$->child = $1; +} + +english: T_DECLARE T_IDENT T_AS english_declaration { + $$ = insert_identifier($4, $2); +} | T_TYPE english_declaration { + $$ = $2; +} + +/* + * We use a precedence declaration to prefer shifting an identifier + * over reducing this empty rule; see below. + */ +storage_func_specs: %prec T_TYPE { $$ = NULL; } +storage_func_specs: declspec_simple storage_func_specs { + ALLOC_STRUCT($$, struct cdecl_declspec, + .type = $1, + .next = $2); +} + +type_qual_spec: typespec_noid | qualifier + +type_qual_specs: { $$ = NULL; } | type_qual_spec type_qual_specs { + $$ = $1; + $$->next = $2; +} + +/* + * The "qualifiers" nonterminal needs to be used here to avoid shift/reduce + * conflicts with pointer declarators. So we end up needing to stitch + * together three different specifiers lists. + */ +post_specs: qualifiers typespec type_qual_specs { + $2->next = $3; + join_specs($2, $1); + $$ = $2; +} + +english_declaration: storage_func_specs english_declarator post_specs { + join_specs($3, $1); + ALLOC_STRUCT($$, struct cdecl, + .specifiers = $3, + .declarators = $2); +} + +english_declarator: { + $$ = NULLDECL; +} | english_declarator qualifiers T_POINTER T_TO { ALLOC_STRUCT($$, struct cdecl_declarator, - .type = CDECL_DECL_IDENT, - .u.ident = $1); -} | direct_declarator array { + .type = CDECL_DECL_POINTER, + .child = $1, + .u.pointer.qualifiers = $2); +} | english_declarator english_array { $$ = $2; - $$->u.array.declarator = $1; -} | T_LPAREN declarator T_RPAREN { + $$->child = $1; +} | english_declarator english_function { $$ = $2; -}; + $$->child = $1; +} + +english_function: T_FUNCTION T_RETURNING { + ALLOC_STRUCT($$, struct cdecl_declarator, + .type = CDECL_DECL_FUNCTION, + .u.function.parameters = NULL); +} | T_FUNCTION T_LPAREN english_parameter_list T_RPAREN T_RETURNING { + $$ = $3; +} + +english_parameter_list: english_parameter varargs { + ALLOC_STRUCT($$, struct cdecl_declarator, + .type = CDECL_DECL_FUNCTION, + .u.function.parameters = $1, + .u.function.variadic = $2); +} | english_parameter T_COMMA english_parameter_list { + $$ = $3; + $1->next = $$->u.function.parameters; + $$->u.function.parameters = $1; +} + +typedef_name_qual: T_IDENT qualifiers { + /* Compiler should be able to elide this assignment. */ + $1->u.declspec.ident = $1->u.declarator.u.ident; + + $$ = &$1->u.declspec; + $$->type = CDECL_TYPE_IDENT; + $$->next = $2; +} + +null_decl: { + $$ = NULLDECL; +} + +/* + * There is a shift/reduce conflict here when an identifier appears as the + * first token. The conflict is between shifting T_IDENT, or reducing the + * empty production for storage_func_specs (cf. english_declaration). + * + * - In either case, if we reduce, we won't match T_IDENT T_AS since the + * stack now has the extra storage_func_specs nonterminal symbol. + * - And if we shift, we won't match english_declaration since it is + * too late to add storage_func_specs to the stack. + * + * The only valid input affected by the conflict is a simple type names, + * possibly followed by qualifiers. So the conflict is adequately resolved + * by shifting, so long as we have a special-case reduction to handle this. + */ +english_parameter: english_declaration | typedef_name_qual null_decl { + ALLOC_STRUCT($$, struct cdecl, + .specifiers = $1, + .declarators = $2); +} | T_IDENT T_AS english_declaration { + $$ = insert_identifier($3, $1); +} + +english_array: T_VLA T_ARRAY english_vla T_OF { + $$ = &$3->u.declarator; + $$->type = CDECL_DECL_ARRAY; + $$->u.array.vla = $$->u.ident; + $$->u.array.length = 0; +} | T_ARRAY array_length T_OF { + ALLOC_STRUCT($$, struct cdecl_declarator, + .type = CDECL_DECL_ARRAY, + .u.array.length = $2); +} + +array_length: { $$ = 0; } +array_length: T_UINT { + if (!($$ = $1)) + FAIL(_("array length must be positive")); +} + +english_vla: T_IDENT | { + if (!($$ = cdecl__alloc_item(1))) + YYERROR; + *$$->s = 0; +} %% -void yyerror(YYLTYPE *loc, struct cdecl **out, const char *err) + +/* + * Expose the token string table to the rest of the library, in order to + * produce strings that match parser keywords. + * + * In order for this to work properly, the Bison output must be postprocessed + * by fix-yytname.awk to remove pointless quotation marks from the keyword + * strings. + */ +const char *cdecl__token_name(unsigned token) +{ + return yytname[YYTRANSLATE(token)]; +} + +static void +yyerror(YYLTYPE *loc, yyscan_t scanner, struct cdecl **out, const char *err) { - if (strstr(err, "T_LEX_ERROR")) + if (strstr(err, yytname[YYTRANSLATE(T_LEX_ERROR)])) return; - fprintf(stderr, "%s\n", err); + cdecl__err(CDECL_ENOPARSE, "%s", err); }