#include "system.h"
#include "getargs.h"
+#include "reader.h"
#include "gram.h"
#include "state.h"
#include "complain.h"
#include "closure.h"
#include "LR0.h"
-
+#include "lalr.h"
+#include "reduce.h"
int nstates;
int final_state;
-core *first_state = NULL;
-shifts *first_shift = NULL;
-reductions *first_reduction = NULL;
+static state_t *first_state = NULL;
+static shifts *first_shift = NULL;
-static core *this_state = NULL;
-static core *last_state = NULL;
+static state_t *this_state = NULL;
+static state_t *last_state = NULL;
static shifts *last_shift = NULL;
-static reductions *last_reduction = NULL;
static int nshifts;
static short *shift_symbol = NULL;
static short *shiftset = NULL;
static short **kernel_base = NULL;
-static short **kernel_end = NULL;
+static int *kernel_size = NULL;
static short *kernel_items = NULL;
/* hash table for states, to recognize equivalent ones. */
-#define STATE_TABLE_SIZE 1009
-static core **state_table = NULL;
+#define STATE_HASH_SIZE 1009
+static state_t **state_hash = NULL;
\f
static void
allocate_itemsets (void)
{
- short *itemp = NULL;
- int symbol;
int i;
- int count;
- short *symbol_count = NULL;
- count = 0;
- symbol_count = XCALLOC (short, nsyms);
+ /* 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. */
+ int count = 0;
+ short *symbol_count = XCALLOC (short, nsyms + nuseless_nonterminals);
- itemp = ritem;
- symbol = *itemp++;
- while (symbol)
- {
- if (symbol > 0)
- {
- count++;
- symbol_count[symbol]++;
- }
- symbol = *itemp++;
- }
+ for (i = 0; ritem[i]; ++i)
+ if (ritem[i] > 0)
+ {
+ count++;
+ symbol_count[ritem[i]]++;
+ }
/* See comments before new_itemsets. All the vectors of items
live inside KERNEL_ITEMS. The number of active items after
count += symbol_count[i];
}
- shift_symbol = symbol_count;
- kernel_end = XCALLOC (short *, nsyms);
+ free (symbol_count);
+ kernel_size = XCALLOC (int, nsyms);
}
shiftset = XCALLOC (short, nsyms);
redset = XCALLOC (short, nrules + 1);
- state_table = XCALLOC (core *, STATE_TABLE_SIZE);
+ state_hash = XCALLOC (state_t *, STATE_HASH_SIZE);
}
static void
free_storage (void)
{
- XFREE (shift_symbol);
- XFREE (redset);
- XFREE (shiftset);
- XFREE (kernel_base);
- XFREE (kernel_end);
+ free (shift_symbol);
+ free (redset);
+ free (shiftset);
+ free (kernel_base);
+ free (kernel_size);
XFREE (kernel_items);
- XFREE (state_table);
+ free (state_hash);
}
| 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. |
+| shifted, and kernel_size[symbol] is their numbers. |
`----------------------------------------------------------------*/
static void
new_itemsets (void)
{
int i;
- int shiftcount;
if (trace_flag)
- fprintf (stderr, "Entering new_itemsets, state = %d\n", nstates);
+ fprintf (stderr, "Entering new_itemsets, state = %d\n",
+ this_state->number);
for (i = 0; i < nsyms; i++)
- kernel_end[i] = NULL;
+ kernel_size[i] = 0;
- shiftcount = 0;
+ shift_symbol = XCALLOC (short, nsyms);
+ nshifts = 0;
- for (i = 0; i < itemsetend - itemset; ++i)
+ for (i = 0; i < nitemset; ++i)
{
int symbol = ritem[itemset[i]];
if (symbol > 0)
{
- short *ksp = kernel_end[symbol];
-
- if (!ksp)
+ if (!kernel_size[symbol])
{
- shift_symbol[shiftcount] = symbol;
- ksp = kernel_base[symbol];
- shiftcount++;
+ shift_symbol[nshifts] = symbol;
+ nshifts++;
}
- *ksp++ = itemset[i] + 1;
- kernel_end[symbol] = ksp;
+ kernel_base[symbol][kernel_size[symbol]] = itemset[i] + 1;
+ kernel_size[symbol]++;
}
}
-
- nshifts = shiftcount;
}
| necessary. |
`-----------------------------------------------------------------*/
-static core *
+static state_t *
new_state (int symbol)
{
- int n;
- core *p;
+ state_t *p;
if (trace_flag)
- fprintf (stderr, "Entering new_state, state = %d, symbol = %d\n",
- nstates, symbol);
+ 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);
- n = kernel_end[symbol] - kernel_base[symbol];
-
- p = CORE_ALLOC (n);
+ p = STATE_ALLOC (kernel_size[symbol]);
p->accessing_symbol = symbol;
p->number = nstates;
- p->nitems = n;
+ p->nitems = kernel_size[symbol];
- shortcpy (p->items, kernel_base[symbol], n);
+ shortcpy (p->items, kernel_base[symbol], kernel_size[symbol]);
last_state->next = p;
last_state = p;
get_state (int symbol)
{
int key;
- short *isp2;
int i;
- core *sp;
-
- int n = kernel_end[symbol] - kernel_base[symbol];
+ state_t *sp;
if (trace_flag)
- fprintf (stderr, "Entering get_state, state = %d, symbol = %d\n",
- nstates, symbol);
+ fprintf (stderr, "Entering get_state, state = %d, symbol = %d (%s)\n",
+ this_state->number, symbol, tags[symbol]);
/* Add up the target state's active item numbers to get a hash key.
*/
key = 0;
- for (i = 0; i < n; ++i)
+ for (i = 0; i < kernel_size[symbol]; ++i)
key += kernel_base[symbol][i];
- key = key % STATE_TABLE_SIZE;
- sp = state_table[key];
+ key = key % STATE_HASH_SIZE;
+ sp = state_hash[key];
if (sp)
{
int found = 0;
while (!found)
{
- if (sp->nitems == n)
+ if (sp->nitems == kernel_size[symbol])
{
- int i;
found = 1;
- for (i = 0; i < n; ++i)
+ for (i = 0; i < kernel_size[symbol]; ++i)
if (kernel_base[symbol][i] != sp->items[i])
found = 0;
}
}
else /* bucket is empty */
{
- state_table[key] = sp = new_state (symbol);
+ state_hash[key] = sp = new_state (symbol);
}
+ if (trace_flag)
+ fprintf (stderr, "Exiting get_state => %d\n", sp->number);
+
return sp->number;
}
int symbol;
if (trace_flag)
- fprintf (stderr, "Entering append_states\n");
-
+ fprintf (stderr, "Entering append_states, state = %d\n",
+ this_state->number);
/* first sort shift_symbol into increasing order */
static void
new_states (void)
{
- first_state = last_state = this_state = CORE_ALLOC (0);
+ first_state = last_state = this_state = STATE_ALLOC (0);
nstates = 1;
}
+/*------------------------------------------------------------.
+| Save the NSHIFTS of SHIFTSET into the current linked list. |
+`------------------------------------------------------------*/
+
static void
save_shifts (void)
{
- shifts *p = SHIFTS_ALLOC (nshifts);
+ shifts *p = shifts_new (nshifts);
p->number = this_state->number;
- p->nshifts = nshifts;
-
shortcpy (p->shifts, shiftset, nshifts);
+ this_state->shifts = p;
if (last_shift)
last_shift->next = 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_shift (void)
+insert_start_shifting_state (void)
{
- core *statep;
+ state_t *statep;
shifts *sp;
- statep = CORE_ALLOC (0);
+ 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_ALLOC (1);
+ sp = shifts_new (1);
+ statep->shifts = sp;
sp->number = nstates++;
- sp->nshifts = 1;
sp->shifts[0] = nstates;
last_shift->next = 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. |
-`------------------------------------------------------------------*/
+/*-----------------------------------------------------------------.
+| 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
-augment_automaton (void)
+insert_eof_shifting_state (void)
{
- int i;
- int k;
- core *statep;
+ state_t *statep;
shifts *sp;
- shifts *sp2;
- shifts *sp1 = NULL;
- sp = first_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 = STATE_ALLOC (0);
+ statep->number = nstates;
- if (sp)
- {
- if (sp->number == 0)
- {
- k = sp->nshifts;
- statep = first_state->next;
+ last_state->next = statep;
+ last_state = statep;
- /* 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;
+ /* 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;
- 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;
+ last_shift->next = sp;
+ last_shift = sp;
+}
- while (sp && sp->number < k)
- {
- sp1 = sp;
- sp = sp->next;
- }
- if (sp && sp->number == k)
- {
- sp2 = SHIFTS_ALLOC (sp->nshifts + 1);
- 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;
- XFREE (sp);
- }
- else
- {
- sp2 = SHIFTS_ALLOC (1);
- 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;
+/*---------------------------------------------------------------.
+| Subroutine of augment_automaton. Create the accepting state. |
+`---------------------------------------------------------------*/
- sp2 = SHIFTS_ALLOC (sp->nshifts + 1);
- sp2->nshifts = sp->nshifts + 1;
+static void
+insert_accepting_state (void)
+{
+ state_t *statep;
- /* 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;
+ /* Note that the variable `final_state' refers to what we sometimes
+ call the termination state. */
+ final_state = 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;
+ /* Make the termination state. */
+ statep = STATE_ALLOC (0);
+ statep->number = nstates++;
+ last_state->next = statep;
+ last_state = statep;
+}
- XFREE (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 = SHIFTS_ALLOC (1);
- 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
+
+
+
+/*------------------------------------------------------------------.
+| 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)
{
- /* There are no shifts for any state.
- Make one shift, from the initial state to the next-to-final state. */
+ /* The first state has no shifts. Make one shift, from the
+ initial state to the next-to-final state. */
- sp = SHIFTS_ALLOC (1);
- sp->nshifts = 1;
+ shifts *sp = shifts_new (1);
+ first_state->shifts = sp;
sp->shifts[0] = nstates;
/* Initialize the chain of shifts with sp. */
/* Create the next-to-final state, with shift to
what will be the final state. */
- insert_start_shift ();
+ insert_start_shifting_state ();
}
+ else
+ {
+ state_t *statep = first_state->next;
+ shifts *sp = first_state->shifts;
+ shifts *sp1 = NULL;
+ /* 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 < sp->nshifts)
+ statep = statep->next;
+
+ if (statep->accessing_symbol == start_symbol)
+ {
+ /* We already have a next-to-final state, i.e., for `start:
+ start . ...'. Make sure it has a shift to what will be
+ the final state. */
+ while (sp && sp->number < statep->number)
+ {
+ sp1 = sp;
+ sp = sp->next;
+ }
- /* 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_ALLOC (0);
- statep->number = nstates;
- last_state->next = statep;
- last_state = statep;
+ if (sp && sp->number == statep->number)
+ {
+ int i;
+ shifts *sp2 = shifts_new (sp->nshifts + 1);
+ sp2->number = statep->number;
+ statep->shifts = sp2;
+ 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;
+ XFREE (sp);
+ }
+ else
+ {
+ shifts *sp2 = shifts_new (1);
+ sp2->number = statep->number;
+ statep->shifts = sp2;
+ 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;
+ }
+ insert_eof_shifting_state ();
+ }
+ else
+ {
+ /* There is no state for `start: start . ...'. */
+ int i, k;
+ shifts *sp2;
+ sp = first_shift;
+
+ /* 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). */
+ sp2 = shifts_new (sp->nshifts + 1);
+ first_state->shifts = sp2;
+ /* 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;
- /* Make the shift from the final state to the termination state. */
- sp = SHIFTS_ALLOC (1);
- sp->number = nstates++;
- sp->nshifts = 1;
- sp->shifts[0] = nstates;
- last_shift->next = sp;
- last_shift = sp;
+ /* 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;
- /* Note that the variable `final_state' refers to what we sometimes call
- the termination state. */
- final_state = nstates;
+ XFREE (sp);
- /* Make the termination state. */
- statep = CORE_ALLOC (0);
- statep->number = nstates++;
- last_state->next = statep;
- last_state = statep;
+ /* 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 ();
}
static void
save_reductions (void)
{
- short *isp;
- int item;
int count;
- reductions *p;
-
- short *rend;
+ int i;
/* Find and count the active items that represent ends of rules. */
count = 0;
- for (isp = itemset; isp < itemsetend; isp++)
+ for (i = 0; i < nitemset; ++i)
{
- item = ritem[*isp];
+ int item = ritem[itemset[i]];
if (item < 0)
redset[count++] = -item;
}
if (count)
{
- p = REDUCTIONS_ALLOC (count);
-
- p->number = this_state->number;
+ reductions *p = REDUCTIONS_ALLOC (count);
p->nreds = count;
-
shortcpy (p->rules, redset, count);
- if (last_reduction)
- last_reduction->next = p;
- else
- first_reduction = p;
- last_reduction = p;
+ this_state->reductions = p;
}
}
\f
+/*--------------------.
+| Build STATE_TABLE. |
+`--------------------*/
+
+static void
+set_state_table (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;
+ }
+ }
+
+ /* Seems to be needed by conflicts.c. */
+ state_table[nstates] = STATE_ALLOC (0);
+ state_table[nstates]->lookaheads = count;
+ }
+}
+
/*-------------------------------------------------------------------.
| Compute the nondeterministic finite state machine (see state.h for |
| details) from the grammar. |
while (this_state)
{
+ if (trace_flag)
+ fprintf (stderr, "Processing state %d (reached by %s)\n",
+ this_state->number, tags[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
/* create the shifts structures for the shifts to those states,
now that the state numbers transitioning to are known */
- if (nshifts > 0)
- save_shifts ();
+ save_shifts ();
/* states are queued when they are created; process them all */
this_state = this_state->next;
/* set up initial and final states as parser wants them */
augment_automaton ();
+
+ /* Set up STATE_TABLE. */
+ set_state_table ();
}
projects / bison.git / blobdiff