/*
* Parse and validate C declarations.
- * Copyright © 2011-2012, 2020-2021, 2023 Nick Bowler
+ * Copyright © 2011-2012, 2020-2021, 2023-2024 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
#include "scan.h"
#include "errmsg.h"
+static struct cdecl *fake_function_param(struct cdecl_declarator *);
+
+/*
+ * Allocate a "parse item", which is a union of several parse tree
+ * structure types, together with a string buffer. The s_sz argument
+ * specifies the size of the string (including its terminator), which
+ * may be zero.
+ *
+ * The union's declarator member is pre-initialized to a valid "identifier"
+ * declarator, which shares several interesting offsets with the "declspec"
+ * structure for an "identifier" type specifier.
+ */
+struct parse_item *cdecl__alloc_item(size_t s_sz)
+{
+ struct parse_item *ret;
+
+ ret = malloc(offsetof(struct parse_item, s) + s_sz);
+ if (!ret) {
+ cdecl__errmsg(CDECL__ENOMEM);
+ return NULL;
+ }
+
+ ret->u.declarator.child = NULL;
+ ret->u.declarator.type = CDECL_DECL_IDENT;
+ ret->u.declarator.u.ident = ret->s;
+
+ return ret;
+}
+
/*
* We can represent type specifiers as a bitmap, which gives us a finite
* list of acceptable bitmap values according to the C standard. However,
*/
static int valid_typespec(struct cdecl_declspec *s)
{
+ struct cdecl_declspec *c;
unsigned long map = 0;
- for (struct cdecl_declspec *c = s; c; c = c->next) {
+ for (c = s; c; c = c->next) {
unsigned long bit;
if (cdecl_spec_kind(c) != CDECL_SPEC_TYPE)
*/
static bool valid_declspecs(struct cdecl *decl, bool top)
{
- struct cdecl_declspec *specs = decl->specifiers;
+ struct cdecl_declspec *c, *specs = decl->specifiers;
struct cdecl_declarator *d = decl->declarators;
bool abstract = cdecl_is_abstract(d);
unsigned num_storage = 0;
if (!valid_typespec(specs))
return false;
- for (struct cdecl_declspec *c = specs; c; c = c->next) {
+ for (c = specs; c; c = c->next) {
switch (cdecl_spec_kind(c)) {
case CDECL_SPEC_TYPE:
if (c->type == CDECL_TYPE_VOID &&
return true;
}
+/*
+ * Find the tree pointer which leads to the parameter's leaf node.
+ *
+ * Return a null pointer if the traversal locates a syntactic element which
+ * prevents function reduction. This occurs if the leaf node declares an
+ * identifier, or for nontrivial fake function parameters (see below).
+ */
+static struct cdecl_declarator **leaf_pointer(struct cdecl *param)
+{
+ struct cdecl_declarator *d, **p = ¶m->declarators;
+
+ if ((param = fake_function_param(param->declarators))) {
+ if (param->declarators->type != CDECL_DECL_NULL)
+ return NULL; /* e.g. int (x (*)) */
+ }
+
+ while ((d = *p)->child) {
+ p = &d->child;
+
+ if (fake_function_param(d->child))
+ return NULL; /* e.g. int (x (*)[][1]) */
+ }
+
+ if (d->type != CDECL_DECL_NULL)
+ return NULL; /* e.g. int (x y) */
+
+ return p;
+}
+
+
/*
* The C grammar leaves ambiguous some cases where parentheses represent a
* function declarator or just parentheses. The language uses additional
static struct cdecl_declarator *reduce_function(struct cdecl *param)
{
- struct cdecl_declspec *spec = param->specifiers;
- struct cdecl_declarator *decl = param->declarators;
- struct cdecl_declarator *last;
-
- for (last = decl; last && last->type != CDECL_DECL_NULL;)
- last = last->child;
+ struct parse_item *spec = (void *)param->specifiers;
+ struct cdecl_declarator *d, **p;
- if (!last)
+ if (!(p = leaf_pointer(param)))
return NULL;
- last->type = CDECL_DECL_IDENT;
- last->u.ident = spec->ident;
- free(param);
- free(spec);
+ /*
+ * The child and u.ident members of cdecl_declarator are expected
+ * to be located at identical offsets as, respectively, the next
+ * and ident members within cdecl_declspec, so the expectation is
+ * that the compiler can elide both assignments.
+ */
+ spec->u.declarator.child = (void *)spec->u.declspec.next;
+ spec->u.declarator.u.ident = spec->u.declspec.ident;
+ spec->u.declarator.type = CDECL_DECL_IDENT;
+ *p = &spec->u.declarator;
- return decl;
+ d = param->declarators;
+ free(param);
+ return d;
}
static bool function_is_reducible(struct cdecl_declarator *d)
new = reduce_function(d->u.function.parameters);
if (!new)
- return 0; /* e.g. int (foo bar) */
+ return 0;
*p = new;
- free(d->child);
free(d);
- return 0;
+ return 1;
}
/*
- * The parser's bias towards considering things as functions whenever possible
- * makes nested parentheses tricky. (x) is considered to be part of a function
- * declarator until simplify_functions converts it. The problem is that
- * (((x))) is not valid as part of a function declarator, but it *is* valid
- * as an identifier enclosed 3 times in parentheses. This is complicated by
- * the fact that things like (((int))) are not valid anywhere.
+ * The main parser's bias towards considering things as functions whenever
+ * possible makes nested parentheses tricky. "(x)" is considered to be part
+ * of a function declarator until simplify_functions converts it. The problem
+ * is that "(((x)))" is not valid as part of a function declarator, but it _is_
+ * valid as either an identifier enclosed thrice in parentheses, or an abstract
+ * function declarator enclosed twice in parentheses.
+ *
+ * To avoid ambiguities, the main parser actually returns a function declarator
+ * for every pair of parentheses. The ones we need to look at consist of a
+ * single parameter with an empty specifier list (noting that every real
+ * function parameter will have at least one type specifier).
+ *
+ * There are two cases:
+ *
+ * - For (), the parser emits a parameter with a lone null declarator.
+ * This fake parameter simply gets deleted, leaving us with a normal
+ * function declarator with an empty identifier list.
*
- * To avoid ambiguities, the parser actually emits a "function" declarator for
- * every pair of parentheses. The ones that can't reasonably be functions
- * consist of a single "parameter" with no declaration specifiers (note that
- * every valid function parameter will have at least one type specifier).
+ * - Otherwise, the parameter's outermost declarator is not null. The
+ * function itself is deleted, replaced in the parse tree with the
+ * fake parameter's declarator.
*
- * This pass is to remove these fake functions from the parse tree. We take
- * care to avoid turning invalid things like ((int)) into valid things like
- * (int) by observing that the only valid function declarators that appear
- * in these "fake" parentheses are those that have a non-null child declarator
- * (for instance, int ((*)(int)) *or* those that will be eliminated by the
- * simplify_functions pass.
+ * Repeating until there no fake parameters, this reduction transforms, for
+ * example, "(((x)))" into "(x)", an abstract function declarator. The result
+ * is then subject to the function simplification step, which will turn "(x)"
+ * into x (declaring an identifier).
+ *
+ * The whole process is repeated until no more changes are made to the parse
+ * tree, or a syntax error is detected.
*/
-
static struct cdecl *fake_function_param(struct cdecl_declarator *d)
{
struct cdecl *param;
reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
{
struct cdecl *param;
- int fake = 0;
-
- while ((param = fake_function_param(d))) {
- struct cdecl_declarator *decl = param->declarators;
- d->u.function.parameters = NULL;
-
- if (decl->type != CDECL_DECL_NULL) {
- if (d->child->type != CDECL_DECL_NULL) {
- /* Found fake parameter on real function. */
- d->u.function.parameters = param;
- cdecl__errmsg(CDECL__EBADPARAM);
- return -1;
+
+ do {
+ d = *p;
+ while ((param = fake_function_param(d))) {
+ struct cdecl_declarator *decl = param->declarators;
+ d->u.function.parameters = NULL;
+
+ if (decl->type != CDECL_DECL_NULL) {
+ if (d->child->type != CDECL_DECL_NULL) {
+ /* Fake parameter on real function. */
+ d->u.function.parameters = param;
+ cdecl__errmsg(CDECL__EBADPARAM);
+ return -1;
+ }
+
+ param->declarators = d;
+ *p = d = decl;
}
- param->declarators = d;
- *p = d = decl;
- fake = 1;
+ cdecl__free(param);
}
-
- cdecl__free(param);
- }
-
- simplify_functions(p, d);
- if (fake && (*p)->type == CDECL_DECL_FUNCTION) {
- /* Started with a fake function but ended with a real one. */
- cdecl__errmsg(CDECL__EMANYPAREN);
- return -1;
- }
+ } while (simplify_functions(p, d));
return 0;
}
static struct cdecl *do_parse(const char *str, int english_mode)
{
+ struct cdecl *decl = NULL;
YY_BUFFER_STATE state;
yyscan_t scanner;
- struct cdecl *decl;
#if YYDEBUG
extern int cdecl__yydebug;
return NULL;
state = cdecl__yy_scan_string(str, scanner);
- if (cdecl__yyparse(scanner, &decl) != 0)
+ if (cdecl__yyparse(scanner, &decl) != 0) {
+ /*
+ * If the input consists of a complete, valid declaration
+ * followed by some garbage, that parsed declaration will
+ * be output by the parser and we need to free it here.
+ */
+ cdecl__free(decl);
decl = NULL;
+ }
cdecl__yy_delete_buffer(state, scanner);
cdecl__yylex_destroy(scanner);
projects / cdecl99.git / blobdiff