X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/78ab8f674f0679b38ca8a6353f562c86b1c95055..d1a1114f7f257f89887ea5825ba0d896dfaa747b:/src/gram.h?ds=sidebyside diff --git a/src/gram.h b/src/gram.h index f143b4ee..de782f7d 100644 --- a/src/gram.h +++ b/src/gram.h @@ -36,18 +36,18 @@ The rules receive rule numbers 1 to NRULES in the order they are written. More precisely Bison augments the grammar with the - initial rule, `$axiom: START-SYMBOL EOF', which is numbered 1, all - the user rules are 2, 3 etc. Each time a rule number is presented - to the user, we subtract 1, so *displayed* rule numbers are 0, 1, - 2... + initial rule, `$accept: START-SYMBOL $end', 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). + 0, and the token $end). - Actions and guards are accessed via the rule number. + Actions are accessed via the rule number. The rules themselves are described by several arrays: amongst which RITEM, and RULES. @@ -68,6 +68,12 @@ RULES[R].assoc -- the associativity of R. + RULES[R].dprec -- the dynamic precedence level of R (for GLR + parsing). + + RULES[R].merger -- index of merging function for R (for GLR + parsing). + RULES[R].line -- the line where R was defined. RULES[R].useful -- TRUE iff the rule is used (i.e., FALSE if thrown @@ -81,11 +87,9 @@ 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. 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 - item number. + The portions of RITEM come in order of increasing rule number. + NRITEMS is the total length of RITEM. Each 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. @@ -99,42 +103,56 @@ Associativities are recorded similarly in SYMBOLS[I]->assoc. */ -#include "symtab.h" +# include "location.h" +# include "symtab.h" -#define ISTOKEN(s) ((s) < ntokens) -#define ISVAR(s) ((s) >= ntokens) +# define ISTOKEN(s) ((s) < ntokens) +# define ISVAR(s) ((s) >= ntokens) -extern int nrules; extern int nsyms; extern int ntokens; extern int nvars; -#define ITEM_NUMBER_MAX INT_MAX typedef int item_number_t; +# define ITEM_NUMBER_MAX ((item_number_t) INT_MAX) +# define ITEM_NUMBER_MIN ((item_number_t) MIN_MAX) extern item_number_t *ritem; -extern int nritems; +extern unsigned 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. +/* There is weird relationship between OT1H item_number_t and OTOH + symbol_number_t and rule_number_t: we store the latter in + item_number_t. symbol_number_t are stored as are, while + the negation of (rule_number_t + 1) are stored. - Therefore, an token_number_t must be a valid item_number_t, and we + Therefore, an symbol_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)) +# define symbol_number_as_item_number(Tok) ((item_number_t) (Tok)) +# define item_number_as_symbol_number(Ite) ((symbol_number_t) (Ite)) + +extern symbol_number_t start_symbol; + +/* Rules numbers. */ +typedef short rule_number_t; +# define RULE_NUMBER_MAX ((rule_number_t) SHRT_MAX) +extern rule_number_t nrules; +# define int_of_rule_number(RNum) ((int) (RNum)) +# define rule_number_as_item_number(RNum) ((item_number_t) (- RNum - 1)) +# define item_number_as_rule_number(INum) ((rule_number_t) (- INum - 1)) -extern token_number_t start_symbol; +/*--------. +| Rules. | +`--------*/ 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; + rule_number_t user_number; /* The index in RULES. Usually the rule number in the source, except if some rules are useless. */ - short number; + rule_number_t number; symbol_t *lhs; item_number_t *rhs; @@ -142,51 +160,97 @@ typedef struct rule_s /* This symbol provides both the associativity, and the precedence. */ symbol_t *prec; + short dprec; + short merger; + /* This symbol was attached to the rule via %prec. */ symbol_t *precsym; - int line; + location_t location; bool useful; const char *action; - int action_line; - - const char *guard; - int guard_line; + location_t action_location; } rule_t; extern struct rule_s *rules; +/* A function that selects a rule. */ +typedef bool (*rule_filter_t) PARAMS ((rule_t *r)); + +/* Return true IFF the rule has a `number' smaller than NRULES. */ +bool rule_useful_p PARAMS ((rule_t *r)); + +/* Return true IFF the rule has a `number' higher than NRULES. */ +bool rule_useless_p PARAMS ((rule_t *r)); + +/* Return true IFF the rule is not flagged as useful *and* is useful. + In other words, it was discarded because of conflicts. */ +bool rule_never_reduced_p PARAMS ((rule_t *r)); + +/* Print this RULE's number and lhs on OUT. If a PREVIOUS_LHS was + already displayed (by a previous call for another rule), avoid + useless repetitions. */ +void rule_lhs_print PARAMS ((rule_t *rule, symbol_t *previous_lhs, FILE *out)); + +/* Return the length of the RHS. */ +int rule_rhs_length PARAMS ((rule_t *rule)); + +/* Print this RULE's RHS on OUT. */ +void rule_rhs_print PARAMS ((rule_t *rule, FILE *out)); + +/* Print this RULE on OUT. */ +void rule_print PARAMS ((rule_t *rule, FILE *out)); + + + + /* 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 symbol_number_t *token_translations; extern int max_user_token_number; -/* SEMANTIC_PARSER is nonzero if the input file says to use the hairy - parser that provides for semantic error recovery. If it is zero, - the yacc-compatible simplified parser is used. */ -extern int semantic_parser; + +/* GLR_PARSER is nonzero if the input file says to use the GLR + (Generalized LR) parser, and to output some additional + information used by the GLR algorithm. */ + +extern int glr_parser; /* PURE_PARSER is nonzero if should generate a parser that is all pure and reentrant. */ extern int pure_parser; -/* 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)); +/* Print the grammar's rules numbers from BEGIN (inclusive) to END + (exclusive) on OUT under TITLE. */ +void grammar_rules_partial_print PARAMS ((FILE *out, const char *title, + rule_filter_t filter)); + +/* Print the grammar's rules on OUT. */ +void grammar_rules_print PARAMS ((FILE *out)); + /* Dump the grammar. */ void grammar_dump PARAMS ((FILE *out, const char *title)); +/* Report on STDERR the rules that are not flagged USEFUL, using the + MESSAGE (which can be `useless rule' when invoked after grammar + reduction, or `never reduced' after conflicts were taken into + account). */ +void grammar_rules_never_reduced_report PARAMS ((const char *message)); + +/* Free the packed grammar. */ +void grammar_free PARAMS ((void)); + #endif /* !GRAM_H_ */