/* Generate the nondeterministic finite state machine for bison,
- Copyright 1984, 1986, 1989, 2000, 2001, 2002 Free Software Foundation, Inc.
+
+ Copyright (C) 1984, 1986, 1989, 2000, 2001, 2002 Free Software
+ Foundation, Inc.
This file is part of Bison, the GNU Compiler Compiler.
#include "lalr.h"
#include "reduce.h"
-unsigned int nstates = 0;
-/* Initialize the final state to -1, otherwise, it might be set to 0
- by default, and since we don't compute the reductions of the final
- state, we end up not computing the reductions of the initial state,
- which is of course needed.
+typedef struct state_list_s
+{
+ struct state_list_s *next;
+ state_t *state;
+} state_list_t;
+
+static state_list_t *first_state = NULL;
+static state_list_t *last_state = NULL;
+
+
+/*------------------------------------------------------------------.
+| A state was just discovered from another state. Queue it for |
+| later examination, in order to find its transitions. Return it. |
+`------------------------------------------------------------------*/
+
+static state_t *
+state_list_append (symbol_number_t symbol,
+ size_t core_size, item_number_t *core)
+{
+ state_list_t *node = XMALLOC (state_list_t, 1);
+ state_t *state = state_new (symbol, core_size, core);
+
+ if (trace_flag & trace_automaton)
+ fprintf (stderr, "state_list_append (state = %d, symbol = %d (%s))\n",
+ nstates, symbol, symbols[symbol]->tag);
+
+ /* If this is the endtoken, and this is not the initial state, then
+ this is the final state. */
+ if (symbol == 0 && first_state)
+ final_state = state;
- FINAL_STATE is properly set by new_state when it recognizes the
- accessing symbol: EOF. */
-int final_state = -1;
-static state_t *first_state = NULL;
+ node->next = NULL;
+ node->state = state;
-static state_t *this_state = NULL;
-static state_t *last_state = NULL;
+ if (!first_state)
+ first_state = node;
+ if (last_state)
+ last_state->next = node;
+ last_state = node;
+
+ return state;
+}
static int nshifts;
-static token_number_t *shift_symbol = NULL;
+static symbol_number_t *shift_symbol = NULL;
-static short *redset = NULL;
-static short *shiftset = NULL;
+static rule_t **redset = NULL;
+static state_t **shiftset = NULL;
static item_number_t **kernel_base = NULL;
static int *kernel_size = NULL;
static item_number_t *kernel_items = NULL;
-/* hash table for states, to recognize equivalent ones. */
-
-#define STATE_HASH_SIZE 1009
-static state_t **state_hash = NULL;
-
\f
static void
allocate_itemsets (void)
{
- int i, r;
+ symbol_number_t i;
+ rule_number_t r;
item_number_t *rhsp;
/* Count the number of occurrences of all the symbols in RITEMS.
int count = 0;
short *symbol_count = XCALLOC (short, nsyms + nuseless_nonterminals);
- for (r = 1; r < nrules + 1; ++r)
+ for (r = 0; r < nrules; ++r)
for (rhsp = rules[r].rhs; *rhsp >= 0; ++rhsp)
{
count++;
/* 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].
+ appears as an item, which is SYMBOL_COUNT[SYMBOL].
We allocate that much space for each symbol. */
kernel_base = XCALLOC (item_number_t *, nsyms);
{
allocate_itemsets ();
- shiftset = XCALLOC (short, nsyms);
- redset = XCALLOC (short, nrules + 1);
- state_hash = XCALLOC (state_t *, STATE_HASH_SIZE);
- shift_symbol = XCALLOC (token_number_t, nsyms);
+ shiftset = XCALLOC (state_t *, nsyms);
+ redset = XCALLOC (rule_t *, nrules);
+ state_hash_new ();
+ shift_symbol = XCALLOC (symbol_number_t, nsyms);
}
free (kernel_base);
free (kernel_size);
XFREE (kernel_items);
- free (state_hash);
+ 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_size[symbol] is their numbers. |
-`----------------------------------------------------------------*/
+/*---------------------------------------------------------------.
+| Find which symbols can be shifted in 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_size[symbol] is their numbers. |
+`---------------------------------------------------------------*/
static void
-new_itemsets (void)
+new_itemsets (state_t *state)
{
int i;
- if (trace_flag)
+ if (trace_flag & trace_automaton)
fprintf (stderr, "Entering new_itemsets, state = %d\n",
- this_state->number);
+ state->number);
for (i = 0; i < nsyms; i++)
kernel_size[i] = 0;
for (i = 0; i < nritemset; ++i)
if (ritem[itemset[i]] >= 0)
{
- token_number_t symbol
- = item_number_as_token_number (ritem[itemset[i]]);
+ symbol_number_t symbol
+ = item_number_as_symbol_number (ritem[itemset[i]]);
if (!kernel_size[symbol])
{
shift_symbol[nshifts] = symbol;
/*-----------------------------------------------------------------.
-| Subroutine of get_state. Create a new state for those items, if |
-| necessary. |
+| Find 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. |
`-----------------------------------------------------------------*/
static state_t *
-new_state (token_number_t symbol, size_t core_size, item_number_t *core)
-{
- state_t *p;
-
- if (trace_flag)
- fprintf (stderr, "Entering new_state, state = %d, symbol = %d (%s)\n",
- nstates, symbol, quotearg_style (escape_quoting_style,
- symbols[symbol]->tag));
-
- if (nstates >= SHRT_MAX)
- fatal (_("too many states (max %d)"), SHRT_MAX);
-
- p = STATE_ALLOC (core_size);
- p->accessing_symbol = symbol;
- p->number = nstates;
- p->nitems = core_size;
-
- memcpy (p->items, core, core_size * sizeof (core[0]));
-
- /* If this is the eoftoken, and this is not the initial state, then
- this is the final state. */
- if (symbol == 0 && first_state)
- final_state = p->number;
-
- if (!first_state)
- first_state = p;
- if (last_state)
- 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. |
-`--------------------------------------------------------------*/
-
-static int
-get_state (token_number_t symbol, size_t core_size, item_number_t *core)
+get_state (symbol_number_t symbol, size_t core_size, item_number_t *core)
{
- int key;
- int i;
state_t *sp;
- if (trace_flag)
- fprintf (stderr, "Entering get_state, state = %d, symbol = %d (%s)\n",
- this_state->number, symbol, quotearg_style (escape_quoting_style,
- symbols[symbol]->tag));
+ if (trace_flag & trace_automaton)
+ fprintf (stderr, "Entering get_state, symbol = %d (%s)\n",
+ symbol, symbols[symbol]->tag);
- /* Add up the target state's active item numbers to get a hash key.
- */
- key = 0;
- for (i = 0; i < core_size; ++i)
- key += core[i];
- key = key % STATE_HASH_SIZE;
- sp = state_hash[key];
-
- if (sp)
- {
- int found = 0;
- while (!found)
- {
- if (sp->nitems == core_size)
- {
- found = 1;
- for (i = 0; i < core_size; ++i)
- if (core[i] != sp->items[i])
- found = 0;
- }
-
- if (!found)
- {
- if (sp->link)
- {
- sp = sp->link;
- }
- else /* bucket exhausted and no match */
- {
- sp = sp->link = new_state (symbol, core_size, core);
- found = 1;
- }
- }
- }
- }
- else /* bucket is empty */
- {
- state_hash[key] = sp = new_state (symbol, core_size, core);
- }
+ sp = state_hash_lookup (core_size, core);
+ if (!sp)
+ sp = state_list_append (symbol, core_size, core);
- if (trace_flag)
+ if (trace_flag & trace_automaton)
fprintf (stderr, "Exiting get_state => %d\n", sp->number);
- 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 STATE. |
+| |
+| SHIFTSET is set up as a vector of those states. |
+`---------------------------------------------------------------*/
static void
-append_states (void)
+append_states (state_t *state)
{
int i;
int j;
- token_number_t symbol;
+ symbol_number_t symbol;
- if (trace_flag)
+ if (trace_flag & trace_automaton)
fprintf (stderr, "Entering append_states, state = %d\n",
- this_state->number);
+ state->number);
/* first sort shift_symbol into increasing order */
}
-static void
-new_states (void)
-{
- /* The 0 at the lhs is the index of the item of this initial rule. */
- kernel_base[0][0] = 0;
- kernel_size[0] = 1;
- this_state = new_state (0, kernel_size[0], kernel_base[0]);
-}
-
-
-/*------------------------------------------------------------.
-| Save the NSHIFTS of SHIFTSET into the current linked list. |
-`------------------------------------------------------------*/
-
-static void
-save_shifts (void)
-{
- shifts *p = shifts_new (nshifts);
- memcpy (p->shifts, shiftset, nshifts * sizeof (shiftset[0]));
- this_state->shifts = p;
-}
-
-
/*----------------------------------------------------------------.
| Find which rules can be used for reduction transitions from the |
| current state and make a reductions structure for the state to |
`----------------------------------------------------------------*/
static void
-save_reductions (void)
+save_reductions (state_t *state)
{
int count = 0;
int i;
- /* If this is the final state, we want it to have no reductions at
- all, although it has one for `START_SYMBOL EOF .'. */
- if (this_state->number == final_state)
- return;
-
/* Find and count the active items that represent ends of rules. */
for (i = 0; i < nritemset; ++i)
{
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. */
- this_state->reductions = reductions_new (count);
- memcpy (this_state->reductions->rules, redset, count * sizeof (redset[0]));
+ state_reductions_set (state, count, redset);
}
\f
static void
set_states (void)
{
- state_t *sp;
states = XCALLOC (state_t *, nstates);
- for (sp = first_state; sp; sp = sp->next)
+ while (first_state)
{
+ state_list_t *this = first_state;
+
/* Pessimization, but simplification of the code: make sure all
- the states have a shifts, errs, and reductions, even if
- reduced to 0. */
- if (!sp->shifts)
- sp->shifts = shifts_new (0);
- if (!sp->errs)
- sp->errs = errs_new (0);
- if (!sp->reductions)
- sp->reductions = reductions_new (0);
-
- states[sp->number] = sp;
+ 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_t *state = this->state;
+ if (!state->transitions)
+ state_transitions_set (state, 0, 0);
+ if (!state->reductions)
+ state_reductions_set (state, 0, 0);
+
+ states[state->number] = state;
+
+ 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. |
void
generate_states (void)
{
+ state_list_t *list = NULL;
allocate_storage ();
new_closure (nritems);
- new_states ();
- while (this_state)
+ /* Create the initial state. The 0 at the lhs is the index of the
+ item of this initial rule. */
+ kernel_base[0][0] = 0;
+ kernel_size[0] = 1;
+ state_list_append (0, kernel_size[0], kernel_base[0]);
+
+ list = first_state;
+
+ while (list)
{
- if (trace_flag)
+ state_t *state = list->state;
+ if (trace_flag & trace_automaton)
fprintf (stderr, "Processing state %d (reached by %s)\n",
- this_state->number,
- quotearg_style (escape_quoting_style,
- symbols[this_state->accessing_symbol]->tag));
+ state->number,
+ symbols[state->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 */
- save_shifts ();
-
- /* states are queued when they are created; process them all */
- this_state = this_state->next;
+ closure (state->items, state->nitems);
+ /* Record the reductions allowed out of this state. */
+ save_reductions (state);
+ /* Find the itemsets of the states that shifts can reach. */
+ new_itemsets (state);
+ /* Find or create the core structures for those states. */
+ append_states (state);
+
+ /* Create the shifts structures for the shifts to those states,
+ now that the state numbers transitioning to are known. */
+ state_transitions_set (state, nshifts, shiftset);
+
+ /* States are queued when they are created; process them all.
+ */
+ list = list->next;
}
/* discard various storage */
projects / bison.git / blobdiff