X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/340ef489223a5d1f9c40674ea2973435cb8ecc1b..213744b51eda9ec8971203ca01f520148f38acfe:/src/LR0.c diff --git a/src/LR0.c b/src/LR0.c index 57f6459d..aeb597d8 100644 --- a/src/LR0.c +++ b/src/LR0.c @@ -1,5 +1,7 @@ -/* Generate the nondeterministic finite state machine for bison, - Copyright (C) 1984, 1986, 1989, 2000 Free Software Foundation, Inc. +/* Generate the nondeterministic finite state machine for Bison. + + Copyright (C) 1984, 1986, 1989, 2000, 2001, 2002, 2004 Free + Software Foundation, Inc. This file is part of Bison, the GNU Compiler Compiler. @@ -23,73 +25,105 @@ The entry point is generate_states. */ #include "system.h" -#include "alloc.h" + +#include +#include + +#include "LR0.h" +#include "closure.h" +#include "complain.h" +#include "getargs.h" #include "gram.h" +#include "gram.h" +#include "lalr.h" +#include "reader.h" +#include "reduce.h" #include "state.h" -#include "complain.h" -#include "closure.h" -#include "LR0.h" +#include "symtab.h" +typedef struct state_list +{ + struct state_list *next; + state *state; +} state_list; -int nstates; -int final_state; -core *first_state; -shifts *first_shift; -reductions *first_reduction; +static state_list *first_state = NULL; +static state_list *last_state = NULL; -static core *this_state; -static core *last_state; -static shifts *last_shift; -static reductions *last_reduction; -static int nshifts; -static short *shift_symbol; +/*------------------------------------------------------------------. +| A state was just discovered from another state. Queue it for | +| later examination, in order to find its transitions. Return it. | +`------------------------------------------------------------------*/ + +static state * +state_list_append (symbol_number sym, size_t core_size, item_number *core) +{ + state_list *node = xmalloc (sizeof *node); + state *s = state_new (sym, core_size, core); -static short *redset; -static short *shiftset; + if (trace_flag & trace_automaton) + fprintf (stderr, "state_list_append (state = %d, symbol = %d (%s))\n", + nstates, sym, symbols[sym]->tag); -static short **kernel_base; -static short **kernel_end; -static short *kernel_items; + /* If this is the endtoken, and this is not the initial state, then + this is the final state. */ + if (sym == 0 && first_state) + final_state = s; -/* hash table for states, to recognize equivalent ones. */ + node->next = NULL; + node->state = s; -#define STATE_TABLE_SIZE 1009 -static core **state_table; + if (!first_state) + first_state = node; + if (last_state) + last_state->next = node; + last_state = node; + + return s; +} + +static int nshifts; +static symbol_number *shift_symbol; + +static rule **redset; +static state **shiftset; + +static item_number **kernel_base; +static int *kernel_size; +static item_number *kernel_items; static void allocate_itemsets (void) { - short *itemp; - int symbol; - int i; - int count; - short *symbol_count; - - count = 0; - symbol_count = NEW2 (nsyms, short); - - itemp = ritem; - symbol = *itemp++; - while (symbol) - { - if (symbol > 0) - { - count++; - symbol_count[symbol]++; - } - symbol = *itemp++; - } + symbol_number i; + rule_number r; + item_number *rhsp; + + /* Count the number of occurrences of all the symbols in RITEMS. + Note that useless productions (hence useless nonterminals) are + browsed too, hence we need to allocate room for _all_ the + symbols. */ + size_t count = 0; + size_t *symbol_count = xcalloc (nsyms + nuseless_nonterminals, + sizeof *symbol_count); + + for (r = 0; r < nrules; ++r) + for (rhsp = rules[r].rhs; *rhsp >= 0; ++rhsp) + { + count++; + symbol_count[*rhsp]++; + } /* See comments before new_itemsets. All the vectors of items live inside KERNEL_ITEMS. The number of active items after - some symbol cannot be more than the number of times that symbol - appears as an item, which is symbol_count[symbol]. + some symbol S cannot be more than the number of times that S + appears as an item, which is SYMBOL_COUNT[S]. We allocate that much space for each symbol. */ - kernel_base = NEW2 (nsyms, short *); - kernel_items = NEW2 (count, short); + kernel_base = xnmalloc (nsyms, sizeof *kernel_base); + kernel_items = xnmalloc (count, sizeof *kernel_items); count = 0; for (i = 0; i < nsyms; i++) @@ -98,8 +132,8 @@ allocate_itemsets (void) count += symbol_count[i]; } - shift_symbol = symbol_count; - kernel_end = NEW2 (nsyms, short *); + free (symbol_count); + kernel_size = xnmalloc (nsyms, sizeof *kernel_size); } @@ -108,495 +142,127 @@ allocate_storage (void) { allocate_itemsets (); - shiftset = NEW2 (nsyms, short); - redset = NEW2 (nrules + 1, short); - state_table = NEW2 (STATE_TABLE_SIZE, core *); + shiftset = xnmalloc (nsyms, sizeof *shiftset); + redset = xnmalloc (nrules, sizeof *redset); + state_hash_new (); + shift_symbol = xnmalloc (nsyms, sizeof *shift_symbol); } static void free_storage (void) { - FREE (shift_symbol); - FREE (redset); - FREE (shiftset); - FREE (kernel_base); - FREE (kernel_end); - FREE (kernel_items); - FREE (state_table); + free (shift_symbol); + free (redset); + free (shiftset); + free (kernel_base); + free (kernel_size); + free (kernel_items); + state_hash_free (); } -/*----------------------------------------------------------------. -| Find which symbols can be shifted in the current state, and for | -| each one record which items would be active after that shift. | -| Uses the contents of itemset. | -| | -| shift_symbol is set to a vector of the symbols that can be | -| shifted. For each symbol in the grammar, kernel_base[symbol] | -| points to a vector of item numbers activated if that symbol is | -| shifted, and kernel_end[symbol] points after the end of that | -| vector. | -`----------------------------------------------------------------*/ +/*---------------------------------------------------------------. +| Find which symbols can be shifted in S, and for each one | +| record which items would be active after that shift. Uses the | +| contents of itemset. | +| | +| shift_symbol is set to a vector of the symbols that can be | +| shifted. For each symbol in the grammar, kernel_base[symbol] | +| points to a vector of item numbers activated if that symbol is | +| shifted, and kernel_size[symbol] is their numbers. | +`---------------------------------------------------------------*/ static void -new_itemsets (void) +new_itemsets (state *s) { - int i; - int shiftcount; - short *isp; - short *ksp; - int symbol; + size_t i; -#if TRACE - fprintf (stderr, "Entering new_itemsets\n"); -#endif + if (trace_flag & trace_automaton) + fprintf (stderr, "Entering new_itemsets, state = %d\n", s->number); - for (i = 0; i < nsyms; i++) - kernel_end[i] = NULL; - - shiftcount = 0; + memset (kernel_size, 0, nsyms * sizeof *kernel_size); - isp = itemset; + nshifts = 0; - while (isp < itemsetend) - { - i = *isp++; - symbol = ritem[i]; - if (symbol > 0) - { - ksp = kernel_end[symbol]; - - if (!ksp) - { - shift_symbol[shiftcount++] = symbol; - ksp = kernel_base[symbol]; - } - - *ksp++ = i + 1; - kernel_end[symbol] = ksp; - } - } + for (i = 0; i < nritemset; ++i) + if (ritem[itemset[i]] >= 0) + { + symbol_number sym = item_number_as_symbol_number (ritem[itemset[i]]); + if (!kernel_size[sym]) + { + shift_symbol[nshifts] = sym; + nshifts++; + } - nshifts = shiftcount; + kernel_base[sym][kernel_size[sym]] = itemset[i] + 1; + kernel_size[sym]++; + } } -/*-----------------------------------------------------------------. -| Subroutine of get_state. Create a new state for those items, if | -| necessary. | -`-----------------------------------------------------------------*/ - -static core * -new_state (int symbol) -{ - int n; - core *p; - short *isp1; - short *isp2; - short *iend; - -#if TRACE - fprintf (stderr, "Entering new_state, symbol = %d\n", symbol); -#endif - - if (nstates >= MAXSHORT) - fatal (_("too many states (max %d)"), MAXSHORT); - - isp1 = kernel_base[symbol]; - iend = kernel_end[symbol]; - n = iend - isp1; - - p = - (core *) xmalloc ((unsigned) (sizeof (core) + (n - 1) * sizeof (short))); - p->accessing_symbol = symbol; - p->number = nstates; - p->nitems = n; - - isp2 = p->items; - while (isp1 < iend) - *isp2++ = *isp1++; - - last_state->next = p; - last_state = p; - - nstates++; - - return p; -} - - /*--------------------------------------------------------------. -| Find the state number for the state we would get to (from the | -| current state) by shifting symbol. Create a new state if no | -| equivalent one exists already. Used by append_states. | +| Find the state we would get to (from the current state) by | +| shifting SYM. Create a new state if no equivalent one exists | +| already. Used by append_states. | `--------------------------------------------------------------*/ -static int -get_state (int symbol) +static state * +get_state (symbol_number sym, size_t core_size, item_number *core) { - int key; - short *isp1; - short *isp2; - short *iend; - core *sp; - int found; - - int n; - -#if TRACE - fprintf (stderr, "Entering get_state, symbol = %d\n", symbol); -#endif + state *sp; - isp1 = kernel_base[symbol]; - iend = kernel_end[symbol]; - n = iend - isp1; + if (trace_flag & trace_automaton) + fprintf (stderr, "Entering get_state, symbol = %d (%s)\n", + sym, symbols[sym]->tag); - /* add up the target state's active item numbers to get a hash key */ - key = 0; - while (isp1 < iend) - key += *isp1++; + sp = state_hash_lookup (core_size, core); + if (!sp) + sp = state_list_append (sym, core_size, core); - key = key % STATE_TABLE_SIZE; + if (trace_flag & trace_automaton) + fprintf (stderr, "Exiting get_state => %d\n", sp->number); - sp = state_table[key]; - - if (sp) - { - found = 0; - while (!found) - { - if (sp->nitems == n) - { - found = 1; - isp1 = kernel_base[symbol]; - isp2 = sp->items; - - while (found && isp1 < iend) - { - if (*isp1++ != *isp2++) - found = 0; - } - } - - if (!found) - { - if (sp->link) - { - sp = sp->link; - } - else /* bucket exhausted and no match */ - { - sp = sp->link = new_state (symbol); - found = 1; - } - } - } - } - else /* bucket is empty */ - { - state_table[key] = sp = new_state (symbol); - } - - return sp->number; + return sp; } -/*------------------------------------------------------------------. -| Use the information computed by new_itemsets to find the state | -| numbers reached by each shift transition from the current state. | -| | -| shiftset is set up as a vector of state numbers of those states. | -`------------------------------------------------------------------*/ +/*---------------------------------------------------------------. +| Use the information computed by new_itemsets to find the state | +| numbers reached by each shift transition from S. | +| | +| SHIFTSET is set up as a vector of those states. | +`---------------------------------------------------------------*/ static void -append_states (void) +append_states (state *s) { int i; - int j; - int symbol; -#if TRACE - fprintf (stderr, "Entering append_states\n"); -#endif + if (trace_flag & trace_automaton) + fprintf (stderr, "Entering append_states, state = %d\n", s->number); - /* first sort shift_symbol into increasing order */ + /* First sort shift_symbol into increasing order. */ for (i = 1; i < nshifts; i++) { - symbol = shift_symbol[i]; - j = i; - while (j > 0 && shift_symbol[j - 1] > symbol) - { - shift_symbol[j] = shift_symbol[j - 1]; - j--; - } - shift_symbol[j] = symbol; + symbol_number sym = shift_symbol[i]; + int j; + for (j = i; 0 < j && sym < shift_symbol[j - 1]; j--) + shift_symbol[j] = shift_symbol[j - 1]; + shift_symbol[j] = sym; } for (i = 0; i < nshifts; i++) { - symbol = shift_symbol[i]; - shiftset[i] = get_state (symbol); - } -} - - -static void -new_states (void) -{ - core *p; - - p = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - first_state = last_state = this_state = p; - nstates = 1; -} - - -static void -save_shifts (void) -{ - shifts *p; - short *sp1; - short *sp2; - short *send; - - p = (shifts *) xmalloc ((unsigned) (sizeof (shifts) + - (nshifts - 1) * sizeof (short))); - - p->number = this_state->number; - p->nshifts = nshifts; - - sp1 = shiftset; - sp2 = p->shifts; - send = shiftset + nshifts; - - while (sp1 < send) - *sp2++ = *sp1++; - - if (last_shift) - { - last_shift->next = p; - last_shift = p; - } - else - { - first_shift = p; - last_shift = p; + symbol_number sym = shift_symbol[i]; + shiftset[i] = get_state (sym, kernel_size[sym], kernel_base[sym]); } } -/*------------------------------------------------------------------. -| Subroutine of augment_automaton. Create the next-to-final state, | -| to which a shift has already been made in the initial state. | -`------------------------------------------------------------------*/ - -static void -insert_start_shift (void) -{ - core *statep; - shifts *sp; - - statep = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - statep->number = nstates; - statep->accessing_symbol = start_symbol; - - last_state->next = statep; - last_state = statep; - - /* Make a shift from this state to (what will be) the final state. */ - sp = NEW (shifts); - sp->number = nstates++; - sp->nshifts = 1; - sp->shifts[0] = nstates; - - last_shift->next = sp; - last_shift = sp; -} - - -/*------------------------------------------------------------------. -| Make sure that the initial state has a shift that accepts the | -| grammar's start symbol and goes to the next-to-final state, which | -| has a shift going to the final state, which has a shift to the | -| termination state. Create such states and shifts if they don't | -| happen to exist already. | -`------------------------------------------------------------------*/ - -static void -augment_automaton (void) -{ - int i; - int k; - core *statep; - shifts *sp; - shifts *sp2; - shifts *sp1 = NULL; - - sp = first_shift; - - if (sp) - { - if (sp->number == 0) - { - k = sp->nshifts; - statep = first_state->next; - - /* The states reached by shifts from first_state are numbered 1...K. - Look for one reached by start_symbol. */ - while (statep->accessing_symbol < start_symbol - && statep->number < k) - statep = statep->next; - - if (statep->accessing_symbol == start_symbol) - { - /* We already have a next-to-final state. - Make sure it has a shift to what will be the final state. */ - k = statep->number; - - while (sp && sp->number < k) - { - sp1 = sp; - sp = sp->next; - } - - if (sp && sp->number == k) - { - sp2 = (shifts *) xmalloc ((unsigned) (sizeof (shifts) - + - sp->nshifts * - sizeof (short))); - sp2->number = k; - sp2->nshifts = sp->nshifts + 1; - sp2->shifts[0] = nstates; - for (i = sp->nshifts; i > 0; i--) - sp2->shifts[i] = sp->shifts[i - 1]; - - /* Patch sp2 into the chain of shifts in place of sp, - following sp1. */ - sp2->next = sp->next; - sp1->next = sp2; - if (sp == last_shift) - last_shift = sp2; - FREE (sp); - } - else - { - sp2 = NEW (shifts); - sp2->number = k; - sp2->nshifts = 1; - sp2->shifts[0] = nstates; - - /* Patch sp2 into the chain of shifts between sp1 and sp. */ - sp2->next = sp; - sp1->next = sp2; - if (sp == 0) - last_shift = sp2; - } - } - else - { - /* There is no next-to-final state as yet. */ - /* Add one more shift in first_shift, - going to the next-to-final state (yet to be made). */ - sp = first_shift; - - sp2 = (shifts *) xmalloc (sizeof (shifts) - + sp->nshifts * sizeof (short)); - sp2->nshifts = sp->nshifts + 1; - - /* Stick this shift into the vector at the proper place. */ - statep = first_state->next; - for (k = 0, i = 0; i < sp->nshifts; k++, i++) - { - if (statep->accessing_symbol > start_symbol && i == k) - sp2->shifts[k++] = nstates; - sp2->shifts[k] = sp->shifts[i]; - statep = statep->next; - } - if (i == k) - sp2->shifts[k++] = nstates; - - /* Patch sp2 into the chain of shifts - in place of sp, at the beginning. */ - sp2->next = sp->next; - first_shift = sp2; - if (last_shift == sp) - last_shift = sp2; - - FREE (sp); - - /* Create the next-to-final state, with shift to - what will be the final state. */ - insert_start_shift (); - } - } - else - { - /* The initial state didn't even have any shifts. - Give it one shift, to the next-to-final state. */ - sp = NEW (shifts); - sp->nshifts = 1; - sp->shifts[0] = nstates; - - /* Patch sp into the chain of shifts at the beginning. */ - sp->next = first_shift; - first_shift = sp; - - /* Create the next-to-final state, with shift to - what will be the final state. */ - insert_start_shift (); - } - } - else - { - /* There are no shifts for any state. - Make one shift, from the initial state to the next-to-final state. */ - - sp = NEW (shifts); - sp->nshifts = 1; - sp->shifts[0] = nstates; - - /* Initialize the chain of shifts with sp. */ - first_shift = sp; - last_shift = sp; - - /* Create the next-to-final state, with shift to - what will be the final state. */ - insert_start_shift (); - } - - /* Make the final state--the one that follows a shift from the - next-to-final state. - The symbol for that shift is 0 (end-of-file). */ - statep = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - statep->number = nstates; - last_state->next = statep; - last_state = statep; - - /* Make the shift from the final state to the termination state. */ - sp = NEW (shifts); - sp->number = nstates++; - sp->nshifts = 1; - sp->shifts[0] = nstates; - last_shift->next = sp; - last_shift = sp; - - /* Note that the variable `final_state' refers to what we sometimes call - the termination state. */ - final_state = nstates; - - /* Make the termination state. */ - statep = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - statep->number = nstates++; - last_state->next = statep; - last_state = statep; -} - - /*----------------------------------------------------------------. | Find which rules can be used for reduction transitions from the | | current state and make a reductions structure for the state to | @@ -604,58 +270,57 @@ augment_automaton (void) `----------------------------------------------------------------*/ static void -save_reductions (void) +save_reductions (state *s) { - short *isp; - short *rp1; - short *rp2; - int item; - int count; - reductions *p; - - short *rend; + int count = 0; + size_t i; /* Find and count the active items that represent ends of rules. */ - - count = 0; - for (isp = itemset; isp < itemsetend; isp++) + for (i = 0; i < nritemset; ++i) { - item = ritem[*isp]; + int item = ritem[itemset[i]]; if (item < 0) - redset[count++] = -item; + redset[count++] = &rules[item_number_as_rule_number (item)]; } /* Make a reductions structure and copy the data into it. */ + state_reductions_set (s, count, redset); +} + + +/*---------------. +| Build STATES. | +`---------------*/ - if (count) +static void +set_states (void) +{ + states = xcalloc (nstates, sizeof *states); + + while (first_state) { - p = (reductions *) xmalloc ((unsigned) (sizeof (reductions) + - (count - 1) * sizeof (short))); - - p->number = this_state->number; - p->nreds = count; - - rp1 = redset; - rp2 = p->rules; - rend = rp1 + count; - - for (/* nothing */; rp1 < rend; ++rp1, ++rp2) - *rp2 = *rp1; - - if (last_reduction) - { - last_reduction->next = p; - last_reduction = p; - } - else - { - first_reduction = p; - last_reduction = p; - } + state_list *this = first_state; + + /* Pessimization, but simplification of the code: make sure all + the states have valid transitions and reductions members, + even if reduced to 0. It is too soon for errs, which are + computed later, but set_conflicts. */ + state *s = this->state; + if (!s->transitions) + state_transitions_set (s, 0, 0); + if (!s->reductions) + state_reductions_set (s, 0, 0); + + states[s->number] = s; + + first_state = this->next; + free (this); } + first_state = NULL; + last_state = NULL; } - + /*-------------------------------------------------------------------. | Compute the nondeterministic finite state machine (see state.h for | | details) from the grammar. | @@ -664,37 +329,49 @@ save_reductions (void) void generate_states (void) { + item_number initial_core = 0; + state_list *list = NULL; allocate_storage (); - new_closure (nitems); - new_states (); + new_closure (nritems); + + /* Create the initial state. The 0 at the lhs is the index of the + item of this initial rule. */ + state_list_append (0, 1, &initial_core); + + list = first_state; - while (this_state) + while (list) { + state *s = list->state; + if (trace_flag & trace_automaton) + fprintf (stderr, "Processing state %d (reached by %s)\n", + s->number, + symbols[s->accessing_symbol]->tag); /* Set up ruleset and itemset for the transitions out of this state. ruleset gets a 1 bit for each rule that could reduce now. itemset gets a vector of all the items that could be accepted next. */ - closure (this_state->items, this_state->nitems); - /* record the reductions allowed out of this state */ - save_reductions (); - /* find the itemsets of the states that shifts can reach */ - new_itemsets (); - /* find or create the core structures for those states */ - append_states (); - - /* create the shifts structures for the shifts to those states, - now that the state numbers transitioning to are known */ - if (nshifts > 0) - save_shifts (); - - /* states are queued when they are created; process them all */ - this_state = this_state->next; + closure (s->items, s->nitems); + /* Record the reductions allowed out of this state. */ + save_reductions (s); + /* Find the itemsets of the states that shifts can reach. */ + new_itemsets (s); + /* Find or create the core structures for those states. */ + append_states (s); + + /* Create the shifts structures for the shifts to those states, + now that the state numbers transitioning to are known. */ + state_transitions_set (s, nshifts, shiftset); + + /* states are queued when they are created; process them all. + */ + list = list->next; } /* discard various storage */ free_closure (); free_storage (); - /* set up initial and final states as parser wants them */ - augment_automaton (); + /* Set up STATES. */ + set_states (); }