/* Generate the nondeterministic finite state machine for bison,
- Copyright 1984, 1986, 1989, 2000, 2001 Free Software Foundation, Inc.
+ Copyright 1984, 1986, 1989, 2000, 2001, 2002 Free Software Foundation, Inc.
This file is part of Bison, the GNU Compiler Compiler.
The entry point is generate_states. */
#include "system.h"
+#include "bitset.h"
+#include "quotearg.h"
+#include "symtab.h"
+#include "gram.h"
#include "getargs.h"
#include "reader.h"
#include "gram.h"
#include "lalr.h"
#include "reduce.h"
-int nstates;
-int final_state;
static state_t *first_state = NULL;
static state_t *this_state = NULL;
static state_t *last_state = NULL;
static int nshifts;
-static short *shift_symbol = NULL;
+static symbol_number_t *shift_symbol = NULL;
static short *redset = NULL;
-static short *shiftset = NULL;
+static state_number_t *shiftset = NULL;
-static short **kernel_base = NULL;
+static item_number_t **kernel_base = NULL;
static int *kernel_size = NULL;
-static short *kernel_items = NULL;
-
-/* hash table for states, to recognize equivalent ones. */
-
-#define STATE_HASH_SIZE 1009
-static state_t **state_hash = NULL;
+static item_number_t *kernel_items = NULL;
\f
static void
allocate_itemsets (void)
{
- int i;
+ int i, r;
+ item_number_t *rhsp;
/* Count the number of occurrences of all the symbols in RITEMS.
Note that useless productions (hence useless nonterminals) are
int count = 0;
short *symbol_count = XCALLOC (short, nsyms + nuseless_nonterminals);
- for (i = 0; ritem[i]; ++i)
- if (ritem[i] > 0)
+ for (r = 1; r < nrules + 1; ++r)
+ for (rhsp = rules[r].rhs; *rhsp >= 0; ++rhsp)
{
count++;
- symbol_count[ritem[i]]++;
+ symbol_count[*rhsp]++;
}
/* See comments before new_itemsets. All the vectors of items
appears as an item, which is symbol_count[symbol].
We allocate that much space for each symbol. */
- kernel_base = XCALLOC (short *, nsyms);
+ kernel_base = XCALLOC (item_number_t *, nsyms);
if (count)
- kernel_items = XCALLOC (short, count);
+ kernel_items = XCALLOC (item_number_t, count);
count = 0;
for (i = 0; i < nsyms; i++)
{
allocate_itemsets ();
- shiftset = XCALLOC (short, nsyms);
+ shiftset = XCALLOC (state_number_t, nsyms);
redset = XCALLOC (short, nrules + 1);
- state_hash = XCALLOC (state_t *, STATE_HASH_SIZE);
+ 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 ();
}
for (i = 0; i < nsyms; i++)
kernel_size[i] = 0;
- shift_symbol = XCALLOC (short, nsyms);
nshifts = 0;
- for (i = 0; i < nitemset; ++i)
- {
- int symbol = ritem[itemset[i]];
- if (symbol > 0)
- {
- if (!kernel_size[symbol])
- {
- shift_symbol[nshifts] = symbol;
- nshifts++;
- }
-
- kernel_base[symbol][kernel_size[symbol]] = itemset[i] + 1;
- kernel_size[symbol]++;
- }
- }
+ for (i = 0; i < nritemset; ++i)
+ if (ritem[itemset[i]] >= 0)
+ {
+ symbol_number_t symbol
+ = item_number_as_symbol_number (ritem[itemset[i]]);
+ if (!kernel_size[symbol])
+ {
+ shift_symbol[nshifts] = symbol;
+ nshifts++;
+ }
+
+ kernel_base[symbol][kernel_size[symbol]] = itemset[i] + 1;
+ kernel_size[symbol]++;
+ }
}
`-----------------------------------------------------------------*/
static state_t *
-new_state (int symbol)
+new_state (symbol_number_t symbol, size_t core_size, item_number_t *core)
{
- state_t *p;
+ state_t *res;
if (trace_flag)
fprintf (stderr, "Entering new_state, state = %d, symbol = %d (%s)\n",
- this_state->number, symbol, tags[symbol]);
-
- if (nstates >= MAXSHORT)
- fatal (_("too many states (max %d)"), MAXSHORT);
+ nstates, symbol, symbol_tag_get (symbols[symbol]));
- p = STATE_ALLOC (kernel_size[symbol]);
- p->accessing_symbol = symbol;
- p->number = nstates;
- p->nitems = kernel_size[symbol];
+ res = state_new (symbol, core_size, core);
+ state_hash_insert (res);
- shortcpy (p->items, kernel_base[symbol], kernel_size[symbol]);
+ /* 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 = res;
- last_state->next = p;
- last_state = p;
- nstates++;
+ if (!first_state)
+ first_state = res;
+ if (last_state)
+ last_state->next = res;
+ last_state = res;
- return p;
+ return res;
}
| equivalent one exists already. Used by append_states. |
`--------------------------------------------------------------*/
-static int
-get_state (int symbol)
+static state_number_t
+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, tags[symbol]);
+ this_state->number, symbol,
+ symbol_tag_get (symbols[symbol]));
- /* Add up the target state's active item numbers to get a hash key.
- */
- key = 0;
- for (i = 0; i < kernel_size[symbol]; ++i)
- key += kernel_base[symbol][i];
- key = key % STATE_HASH_SIZE;
- sp = state_hash[key];
-
- if (sp)
- {
- int found = 0;
- while (!found)
- {
- if (sp->nitems == kernel_size[symbol])
- {
- found = 1;
- for (i = 0; i < kernel_size[symbol]; ++i)
- if (kernel_base[symbol][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);
- found = 1;
- }
- }
- }
- }
- else /* bucket is empty */
- {
- state_hash[key] = sp = new_state (symbol);
- }
+ sp = state_hash_lookup (core_size, core);
+ if (!sp)
+ sp = new_state (symbol, core_size, core);
if (trace_flag)
fprintf (stderr, "Exiting get_state => %d\n", sp->number);
{
int i;
int j;
- int symbol;
+ symbol_number_t symbol;
if (trace_flag)
fprintf (stderr, "Entering append_states, state = %d\n",
}
for (i = 0; i < nshifts; i++)
- shiftset[i] = get_state (shift_symbol[i]);
+ {
+ symbol = shift_symbol[i];
+ shiftset[i] = get_state (symbol,
+ kernel_size[symbol], kernel_base[symbol]);
+ }
}
static void
new_states (void)
{
- first_state = last_state = this_state = STATE_ALLOC (0);
- nstates = 1;
+ /* 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_shifts (void)
{
shifts *p = shifts_new (nshifts);
-
- p->number = this_state->number;
- shortcpy (p->shifts, shiftset, nshifts);
+ memcpy (p->shifts, shiftset, nshifts * sizeof (shiftset[0]));
this_state->shifts = p;
}
-/*------------------------------------------------------------------.
-| Subroutine of augment_automaton. Create the next-to-final state, |
-| to which a shift has already been made in the initial state. |
-| |
-| The task of this state consists in shifting (actually, it's a |
-| goto, but shifts and gotos are both stored in SHIFTS) the start |
-| symbols, hence the name. |
-`------------------------------------------------------------------*/
-
-static void
-insert_start_shifting_state (void)
-{
- state_t *statep;
- shifts *sp;
-
- statep = STATE_ALLOC (0);
- statep->number = nstates;
-
- /* The distinctive feature of this state from the
- eof_shifting_state, is that it is labeled as post-start-symbol
- shifting. I fail to understand why this state, and the
- post-start-start can't be merged into one. But it does fail if
- you try. --akim */
- 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 = shifts_new (1);
- statep->shifts = sp;
- sp->number = nstates++;
- sp->shifts[0] = nstates;
-}
-
-
-/*-----------------------------------------------------------------.
-| Subroutine of augment_automaton. Create the final state, which |
-| shifts `0', the end of file. The initial state shifts the start |
-| symbol, and goes to here. |
-`-----------------------------------------------------------------*/
-
-static void
-insert_eof_shifting_state (void)
-{
- state_t *statep;
- shifts *sp;
-
- /* 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 = STATE_ALLOC (0);
- statep->number = nstates;
-
- last_state->next = statep;
- last_state = statep;
-
- /* Make the shift from the final state to the termination state. */
- sp = shifts_new (1);
- statep->shifts = sp;
- sp->number = nstates++;
- sp->shifts[0] = nstates;
-}
-
-
-/*---------------------------------------------------------------.
-| Subroutine of augment_automaton. Create the accepting state. |
-`---------------------------------------------------------------*/
-
-static void
-insert_accepting_state (void)
-{
- state_t *statep;
-
- /* Note that the variable `final_state' refers to what we sometimes
- call the termination state. */
- final_state = nstates;
-
- /* Make the termination state. */
- statep = STATE_ALLOC (0);
- statep->number = nstates++;
- last_state->next = statep;
- last_state = statep;
-}
-
-
-
-
-
-/*------------------------------------------------------------------.
-| 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)
-{
- if (!first_state->shifts->nshifts)
- {
- /* The first state has no shifts. Make one shift, from the
- initial state to the next-to-final state. */
-
- shifts *sp = shifts_new (1);
- first_state->shifts = sp;
- sp->shifts[0] = nstates;
-
- /* Create the next-to-final state, with shift to
- what will be the final state. */
- insert_start_shifting_state ();
- }
- else
- {
- state_t *statep = first_state->next;
- /* The states reached by shifts from FIRST_STATE are numbered
- 1..(SP->NSHIFTS). Look for one reached by START_SYMBOL.
- This is typical of `start: start ... ;': there is a state
- with the item `start: start . ...'. We want to add a `shift
- on EOF to eof-shifting state here. */
- while (statep->accessing_symbol != start_symbol
- && statep->number < first_state->shifts->nshifts)
- statep = statep->next;
-
- if (statep->accessing_symbol == start_symbol)
- {
- /* We already have STATEP, a next-to-final state for `start:
- start . ...'. Make sure it has a shift to what will be
- the final state. */
- int i;
-
- /* Find the shift of the inital state that leads to STATEP. */
- shifts *sp = statep->shifts;
-
- shifts *sp1 = shifts_new (sp->nshifts + 1);
- sp1->number = statep->number;
- statep->shifts = sp1;
- sp1->shifts[0] = nstates;
- for (i = sp->nshifts; i > 0; i--)
- sp1->shifts[i] = sp->shifts[i - 1];
-
- XFREE (sp);
-
- insert_eof_shifting_state ();
- }
- else
- {
- /* There is no state for `start: start . ...'. */
- int i, k;
- shifts *sp = first_state->shifts;
- shifts *sp1 = NULL;
-
- /* Add one more shift to the initial state, going to the
- next-to-final state (yet to be made). */
- sp1 = shifts_new (sp->nshifts + 1);
- first_state->shifts = sp1;
- /* 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)
- sp1->shifts[k++] = nstates;
- sp1->shifts[k] = sp->shifts[i];
- statep = statep->next;
- }
- if (i == k)
- sp1->shifts[k++] = nstates;
-
- XFREE (sp);
-
- /* Create the next-to-final state, with shift to what will
- be the final state. Corresponds to `start: start . ...'. */
- insert_start_shifting_state ();
- }
- }
-
- insert_accepting_state ();
-}
-
-
/*----------------------------------------------------------------.
| 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)
{
- int count;
+ int count = 0;
int i;
- /* Find and count the active items that represent ends of rules. */
+ /* If this is the final state, we want it to have no reductions at
+ all, although it has one for `START_SYMBOL EOF .'. */
+ if (final_state && this_state->number == final_state->number)
+ return;
- count = 0;
- for (i = 0; i < nitemset; ++i)
+ /* 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)
}
/* Make a reductions structure and copy the data into it. */
-
- if (count)
- {
- reductions *p = REDUCTIONS_ALLOC (count);
- p->nreds = count;
- shortcpy (p->rules, redset, count);
-
- this_state->reductions = p;
- }
+ this_state->reductions = reductions_new (count);
+ memcpy (this_state->reductions->rules, redset, count * sizeof (redset[0]));
}
\f
-/*--------------------.
-| Build STATE_TABLE. |
-`--------------------*/
+/*---------------.
+| Build STATES. |
+`---------------*/
static void
-set_state_table (void)
+set_states (void)
{
- /* NSTATES + 1 because lookahead for the pseudo state number NSTATES
- might be used (see conflicts.c). It is too opaque for me to
- provide a probably less hacky implementation. --akim */
- state_table = XCALLOC (state_t *, nstates + 1);
-
- {
- state_t *sp;
- for (sp = first_state; sp; sp = sp->next)
- state_table[sp->number] = sp;
- }
-
- /* Pessimization, but simplification of the code: make sure all the
- states have a shifts, even if reduced to 0 shifts. */
- {
- int i;
- for (i = 0; i < nstates; i++)
- if (!state_table[i]->shifts)
- state_table[i]->shifts = shifts_new (0);
- }
-
- /* Initializing the lookaheads members. Please note that it must be
- performed after having set some of the other members which are
- used below. Change with extreme caution. */
- {
- int i;
- int count = 0;
- for (i = 0; i < nstates; i++)
- {
- int k;
- reductions *rp = state_table[i]->reductions;
- shifts *sp = state_table[i]->shifts;
-
- state_table[i]->lookaheads = count;
-
- if (rp
- && (rp->nreds > 1 || (sp->nshifts && SHIFT_IS_SHIFT (sp, 0))))
- count += rp->nreds;
- else
- state_table[i]->consistent = 1;
-
- for (k = 0; k < sp->nshifts; k++)
- if (SHIFT_IS_ERROR (sp, k))
- {
- state_table[i]->consistent = 0;
- break;
- }
- }
+ state_t *sp;
+ states = XCALLOC (state_t *, nstates);
- /* Seems to be needed by conflicts.c. */
- state_table[nstates] = STATE_ALLOC (0);
- state_table[nstates]->lookaheads = count;
- }
+ for (sp = first_state; sp; sp = sp->next)
+ {
+ /* 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;
+ }
}
+
/*-------------------------------------------------------------------.
| Compute the nondeterministic finite state machine (see state.h for |
| details) from the grammar. |
generate_states (void)
{
allocate_storage ();
- new_closure (nitems);
+ new_closure (nritems);
new_states ();
while (this_state)
{
if (trace_flag)
fprintf (stderr, "Processing state %d (reached by %s)\n",
- this_state->number, tags[this_state->accessing_symbol]);
+ this_state->number,
+ symbol_tag_get (symbols[this_state->accessing_symbol]));
/* 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
free_closure ();
free_storage ();
- /* set up initial and final states as parser wants them */
- augment_automaton ();
-
- /* Set up STATE_TABLE. */
- set_state_table ();
+ /* Set up STATES. */
+ set_states ();
}
projects / bison.git / blobdiff