/* Data definitions for internal representation of bison's input,
- Copyright (C) 1984, 1986, 1989, 1992 Free Software Foundation, Inc.
+ Copyright (C) 1984, 1986, 1989, 1992, 2001, 2002
+ Free Software Foundation, Inc.
This file is part of Bison, the GNU Compiler Compiler.
#ifndef GRAM_H_
# define GRAM_H_
-/* representation of the grammar rules:
+/* Representation of the grammar rules:
- ntokens is the number of tokens, and nvars is the number of
- variables (nonterminals). nsyms is the total number, ntokens +
+ NTOKENS is the number of tokens, and NVARS is the number of
+ variables (nonterminals). NSYMS is the total number, ntokens +
nvars.
- (the true number of token values assigned is ntokens reduced by one
- for each alias declaration)
-
Each symbol (either token or variable) receives a symbol number.
- Numbers 0 to ntokens-1 are for tokens, and ntokens to nsyms-1 are
- for variables. Symbol number zero is the end-of-input token. This
- token is counted in ntokens.
+ Numbers 0 to NTOKENS - 1 are for tokens, and NTOKENS to NSYMS - 1
+ are for variables. Symbol number zero is the end-of-input token.
+ This token is counted in ntokens. The true number of token values
+ assigned is NTOKENS reduced by one for each alias declaration.
+
+ The rules receive rule numbers 1 to NRULES in the order they are
+ written. More precisely Bison augments the grammar with the
+ initial rule, `$axiom: START-SYMBOL EOF', which is numbered 1, all
+ the user rules are 2, 3 etc. Each time a rule number is presented
+ to the user, we subtract 1, so *displayed* rule numbers are 0, 1,
+ 2...
+
+ Internally, we cannot use the number 0 for a rule because for
+ instance RITEM stores both symbol (the RHS) and rule numbers: the
+ symbols are shorts >= 0, and rule number are stored negative.
+ Therefore 0 cannot be used, since it would be both the rule number
+ 0, and the token EOF).
+
+ Actions and guards are accessed via the rule number.
+
+ The rules themselves are described by several arrays: amongst which
+ RITEM, and RULES.
+
+ RULES is an array of struct rule_s, which members are:
- The rules receive rule numbers 1 to nrules in the order they are
- written. Actions and guards are accessed via the rule number.
+ RULES[R].lhs -- the symbol of the left hand side of rule R.
- The rules themselves are described by three arrays: rrhs, rlhs and
- ritem. rlhs[R] is the symbol number of the left hand side of rule
- R. The right hand side is stored as symbol numbers in a portion of
- ritem. rrhs[R] contains the index in ritem of the beginning of the
- portion for rule R.
+ RULES[R].rhs -- the index in RITEM of the beginning of the portion
+ for rule R.
- If rlhs[R] is -1, the rule has been thrown out by reduce.c and
- should be ignored.
+ RULES[R].prec -- the symbol providing the precedence level of R.
+
+ RULES[R].precsym -- the symbol attached (via %prec) to give its
+ precedence to R. Of course, if set, it is equal to `prec', but we
+ need to distinguish one from the other when reducing: a symbol used
+ in a %prec is not useless.
+
+ RULES[R].assoc -- the associativity of R.
+
+ RULES[R].line -- the line where R was defined.
+
+ RULES[R].useful -- TRUE iff the rule is used (i.e., FALSE if thrown
+ away by reduce).
+
+ The right hand side is stored as symbol numbers in a portion of
+ RITEM.
The length of the portion is one greater than the number of symbols
in the rule's right hand side. The last element in the portion
contains minus R, which identifies it as the end of a portion and
says which rule it is for.
- The portions of ritem come in order of increasing rule number and
- are followed by an element which is zero to mark the end. nitems
+ The portions of RITEM come in order of increasing rule number and
+ are followed by an element which is zero to mark the end. nritems
is the total length of ritem, not counting the final zero. Each
- element of ritem is called an "item" and its index in ritem is an
+ element of RITEM is called an "item" and its index in RITEM is an
item number.
Item numbers are used in the finite state machine to represent
places that parsing can get to.
- Precedence levels are recorded in the vectors sprec and rprec.
- sprec records the precedence level of each symbol, rprec the
- precedence level of each rule. rprecsym is the symbol-number of
- the symbol in %prec for this rule (if any).
+ SYMBOLS[I]->prec records the precedence level of each symbol.
Precedence levels are assigned in increasing order starting with 1
so that numerically higher precedence values mean tighter binding
as they ought to. Zero as a symbol or rule's precedence means none
is assigned.
- Associativities are recorded similarly in rassoc and sassoc. */
+ Associativities are recorded similarly in SYMBOLS[I]->assoc. */
+#include "symtab.h"
#define ISTOKEN(s) ((s) < ntokens)
#define ISVAR(s) ((s) >= ntokens)
-extern int nitems;
extern int nrules;
extern int nsyms;
extern int ntokens;
extern int nvars;
-extern short *ritem;
-extern short *rlhs;
-extern short *rrhs;
-extern short *rprec;
-extern short *rprecsym;
-extern short *sprec;
-extern short *rassoc;
-extern short *sassoc;
-extern short *rline; /* Source line number of each rule */
+#define ITEM_NUMBER_MAX INT_MAX
+typedef int item_number_t;
+extern item_number_t *ritem;
+extern int nritems;
+
+/* There is weird relationship between item_number_t and
+ token_number_t: we store token_number_t in item_number_t, but in
+ the latter we also store, as negative numbers, the rule numbers.
-extern int start_symbol;
+ Therefore, an token_number_t must be a valid item_number_t, and we
+ sometimes have to perform the converse transformation. */
+#define token_number_as_item_number(Tok) ((item_number_t) (Tok))
+#define item_number_as_token_number(Ite) ((token_number_t) (Ite))
+extern token_number_t start_symbol;
-/* associativity values in elements of rassoc, sassoc. */
-#define RIGHT_ASSOC 1
-#define LEFT_ASSOC 2
-#define NON_ASSOC 3
+typedef struct rule_s
+{
+ /* The number of the rule in the source. It is usually the index in
+ RULES too, except if there are useless rules. */
+ short user_number;
-/* token translation table: indexed by a token number as returned by
- the user's yylex routine, it yields the internal token number used
- by the parser and throughout bison. If translations is zero, the
- translation table is not used because the two kinds of token
- numbers are the same. (It is noted in reader.c that "Nowadays
- translations is always set to 1...") */
+ /* The index in RULES. Usually the rule number in the source,
+ except if some rules are useless. */
+ short number;
-extern short *token_translations;
-extern int translations;
+ symbol_t *lhs;
+ item_number_t *rhs;
+
+ /* This symbol provides both the associativity, and the precedence. */
+ symbol_t *prec;
+
+ /* This symbol was attached to the rule via %prec. */
+ symbol_t *precsym;
+
+ int line;
+ bool useful;
+
+ const char *action;
+ int action_line;
+
+ const char *guard;
+ int guard_line;
+} rule_t;
+
+extern struct rule_s *rules;
+
+/* Table of the symbols, indexed by the symbol number. */
+extern symbol_t **symbols;
+
+/* TOKEN_TRANSLATION -- a table indexed by a token number as returned
+ by the user's yylex routine, it yields the internal token number
+ used by the parser and throughout bison. */
+extern token_number_t *token_translations;
extern int max_user_token_number;
/* SEMANTIC_PARSER is nonzero if the input file says to use the hairy
extern int pure_parser;
-/* ERROR_TOKEN_NUMBER is the token number of the error token. */
+/* Report the length of the RHS. */
+int rule_rhs_length PARAMS ((rule_t *rule));
+
+/* Dump RITEM for traces. */
+void ritem_print PARAMS ((FILE *out));
+
+/* Return the size of the longest rule RHS. */
+size_t ritem_longest_rhs PARAMS ((void));
+
+/* Dump the grammar. */
+void grammar_dump PARAMS ((FILE *out, const char *title));
-extern int error_token_number;
#endif /* !GRAM_H_ */
projects / bison.git / blobdiff