/* Grammar reduction for Bison.
- Copyright 1988, 1989, 2000, 2001 Free Software Foundation, Inc.
+ Copyright (C) 1988, 1989, 2000, 2001, 2002 Free Software Foundation, Inc.
This file is part of Bison, the GNU Compiler Compiler.
user's parser. */
#include "system.h"
+#include "quotearg.h"
#include "getargs.h"
#include "files.h"
#include "symtab.h"
#include "reduce.h"
#include "reader.h"
#include "getargs.h"
-
-typedef unsigned *BSet;
-typedef short *rule;
-
+#include "bitset.h"
/* Set of all nonterminals which are not useless. */
-static BSet N;
+static bitset N;
/* Set of all rules which have no useless nonterminals in their RHS. */
-static BSet P;
+static bitset P;
/* Set of all accessible symbols. */
-static BSet V;
+static bitset V;
/* Set of symbols used to define rule precedence (so they are
`useless', but no warning should be issued). */
-static BSet V1;
+static bitset V1;
static int nuseful_productions;
static int nuseless_productions;
static int nuseful_nonterminals;
int nuseless_nonterminals;
\f
-static bool
-bits_equal (BSet L, BSet R, int n)
-{
- int i;
-
- for (i = n - 1; i >= 0; i--)
- if (L[i] != R[i])
- return FALSE;
- return TRUE;
-}
-
-
-static int
-nbits (unsigned i)
-{
- int count = 0;
-
- while (i != 0)
- {
- i ^= (i & ((unsigned) (-(int) i)));
- ++count;
- }
- return count;
-}
-
-
-static int
-bits_size (BSet S, int n)
-{
- int i, count = 0;
-
- for (i = n - 1; i >= 0; i--)
- count += nbits (S[i]);
- return count;
-}
-\f
/*-------------------------------------------------------------------.
| Another way to do this would be with a set for each production and |
| then do subset tests against N0, but even for the C grammar the |
`-------------------------------------------------------------------*/
static bool
-useful_production (int i, BSet N0)
+useful_production (int i, bitset N0)
{
- rule r;
+ item_number_t *r;
short n;
/* A production is useful if all of the nonterminals in its appear
in the set of useful nonterminals. */
- for (r = &ritem[rules[i].rhs]; *r >= 0; r++)
- if (ISVAR (n = *r))
- if (!BITISSET (N0, n - ntokens))
- return FALSE;
+ for (r = rules[i].rhs; *r >= 0; r++)
+ if (ISVAR (n = *r) && !bitset_test (N0, n - ntokens))
+ return FALSE;
return TRUE;
}
static void
useless_nonterminals (void)
{
- BSet Np, Ns;
+ bitset Np, Ns;
int i;
/* N is set as built. Np is set being built this iteration. P is
set of all productions which have a RHS all in N. */
- Np = XCALLOC (unsigned, WORDSIZE (nvars));
+ Np = bitset_create (nvars, BITSET_FIXED);
+
/* The set being computed is a set of nonterminals which can derive
the empty string or strings consisting of all terminals. At each
while (1)
{
- for (i = WORDSIZE (nvars) - 1; i >= 0; i--)
- Np[i] = N[i];
- for (i = 1; i <= nrules; i++)
- {
- if (!BITISSET (P, i))
- {
- if (useful_production (i, N))
- {
- SETBIT (Np, rules[i].lhs - ntokens);
- SETBIT (P, i);
- }
- }
- }
- if (bits_equal (N, Np, WORDSIZE (nvars)))
+ bitset_copy (Np, N);
+ for (i = 1; i < nrules + 1; i++)
+ if (!bitset_test (P, i)
+ && useful_production (i, N))
+ {
+ bitset_set (Np, rules[i].lhs->number - ntokens);
+ bitset_set (P, i);
+ }
+ if (bitset_equal_p (N, Np))
break;
Ns = Np;
Np = N;
N = Ns;
}
- XFREE (N);
+ bitset_free (N);
N = Np;
}
static void
inaccessable_symbols (void)
{
- BSet Vp, Vs, Pp;
+ bitset Vp, Vs, Pp;
int i;
short t;
- rule r;
+ item_number_t *r;
/* Find out which productions are reachable and which symbols are
used. Starting with an empty set of productions and a set of
terminals are printed (if running in verbose mode) so that the
user can know. */
- Vp = XCALLOC (unsigned, WORDSIZE (nsyms));
- Pp = XCALLOC (unsigned, WORDSIZE (nrules + 1));
+ Vp = bitset_create (nsyms, BITSET_FIXED);
+ Pp = bitset_create (nrules + 1, BITSET_FIXED);
/* If the start symbol isn't useful, then nothing will be useful. */
- if (BITISSET (N, start_symbol - ntokens))
+ if (bitset_test (N, start_symbol - ntokens))
{
- SETBIT (V, start_symbol);
+ bitset_set (V, start_symbol);
while (1)
{
- for (i = WORDSIZE (nsyms) - 1; i >= 0; i--)
- Vp[i] = V[i];
- for (i = 1; i <= nrules; i++)
+ bitset_copy (Vp, V);
+ for (i = 1; i < nrules + 1; i++)
{
- if (!BITISSET (Pp, i)
- && BITISSET (P, i)
- && BITISSET (V, rules[i].lhs))
+ if (!bitset_test (Pp, i)
+ && bitset_test (P, i)
+ && bitset_test (V, rules[i].lhs->number))
{
- for (r = &ritem[rules[i].rhs]; *r >= 0; r++)
- if (ISTOKEN (t = *r) || BITISSET (N, t - ntokens))
- SETBIT (Vp, t);
- SETBIT (Pp, i);
+ for (r = rules[i].rhs; *r >= 0; r++)
+ if (ISTOKEN (t = *r) || bitset_test (N, t - ntokens))
+ bitset_set (Vp, t);
+ bitset_set (Pp, i);
}
}
- if (bits_equal (V, Vp, WORDSIZE (nsyms)))
+ if (bitset_equal_p (V, Vp))
break;
Vs = Vp;
Vp = V;
}
}
- XFREE (V);
+ bitset_free (V);
V = Vp;
/* Tokens 0, 1, and 2 are internal to Bison. Consider them useful. */
- SETBIT (V, 0); /* end-of-input token */
- SETBIT (V, 1); /* error token */
- SETBIT (V, 2); /* some undefined token */
+ bitset_set (V, eoftoken->number); /* end-of-input token */
+ bitset_set (V, errtoken->number); /* error token */
+ bitset_set (V, undeftoken->number); /* some undefined token */
- XFREE (P);
+ bitset_free (P);
P = Pp;
- nuseful_productions = bits_size (P, WORDSIZE (nrules + 1));
+ nuseful_productions = bitset_count (P);
nuseless_productions = nrules - nuseful_productions;
nuseful_nonterminals = 0;
for (i = ntokens; i < nsyms; i++)
- if (BITISSET (V, i))
+ if (bitset_test (V, i))
nuseful_nonterminals++;
nuseless_nonterminals = nvars - nuseful_nonterminals;
/* A token that was used in %prec should not be warned about. */
- for (i = 1; i < nrules; i++)
+ for (i = 1; i < nrules + 1; i++)
if (rules[i].precsym != 0)
- SETBIT (V1, rules[i].precsym);
+ bitset_set (V1, rules[i].precsym->number);
}
+
+/*-------------------------------------------------------------------.
+| Put the useless productions at the end of RULES, and adjust NRULES |
+| accordingly. |
+`-------------------------------------------------------------------*/
+
static void
reduce_grammar_tables (void)
{
- /* This is turned off because we would need to change the numbers in
- the case statements in the actions file.
-
- We don't disable it via CPP so that it is still checked with the
- rest of the code, to avoid its becoming completely obsolete.
+ /* Flag useless productions. */
+ {
+ int pn;
+ for (pn = 1; pn < nrules + 1; pn++)
+ rules[pn].useful = bitset_test (P, pn);
+ }
- FIXME: I think the comment above demonstrates this code must be
- turned off for *semantic* parser, not in the general case. Try
- to understand this better --akim. */
+ /* Map the nonterminals to their new index: useful first, useless
+ afterwards. Kept for later report. */
+ {
+ int useful = 1;
+ int useless = nrules + 1 - nuseless_productions;
+ rule_t *rules_sorted = XMALLOC (rule_t, nrules + 1) - 1;
+ int i;
+ for (i = 1; i < nrules + 1; ++i)
+ rules_sorted[rules[i].useful ? useful++ : useless++] = rules[i];
+ free (rules + 1);
+ rules = rules_sorted;
- if (0)
- /* remove useless productions */
- if (nuseless_productions > 0)
+ /* Renumber the rules markers in RITEMS. */
+ for (i = 1; i < nrules + 1; ++i)
{
- short np, pn, ni, pi;
-
- np = 0;
- ni = 0;
- for (pn = 1; pn <= nrules; pn++)
- if (BITISSET (P, pn))
- {
- np++;
- if (pn != np)
- {
- rules[np].lhs = rules[pn].lhs;
- rules[np].line = rules[pn].line;
- rules[np].prec = rules[pn].prec;
- rules[np].assoc = rules[pn].assoc;
- rules[np].rhs = rules[pn].rhs;
- if (rules[np].rhs != ni)
- {
- pi = rules[np].rhs;
- rules[np].rhs = ni;
- while (ritem[pi] >= 0)
- ritem[ni++] = ritem[pi++];
- ritem[ni++] = -np;
- }
- }
- else
- {
- while (ritem[ni++] >= 0)
- /* Nothing. */;
- }
- }
-
- ritem[ni] = 0;
- nrules -= nuseless_productions;
- nitems = ni;
- nritems = ni;
-
- /* Is it worth it to reduce the amount of memory for the
- grammar? Probably not. */
+ item_number_t *rhsp = rules[i].rhs;
+ for (/* Nothing. */; *rhsp >= 0; ++rhsp)
+ /* Nothing. */;
+ *rhsp = -i;
+ rules[i].number = i;
}
+ nrules -= nuseless_productions;
+ }
- /* Disable useless productions. */
- if (nuseless_productions > 0)
- {
- int pn;
- for (pn = 1; pn <= nrules; pn++)
- rules[pn].useful = BITISSET (P, pn);
- }
+ /* Adjust NRITEMS and NITEMS. */
+ {
+ int r;
+ int length;
+ for (r = nrules + 1; r < nrules + 1 + nuseless_productions; ++r)
+ {
+ length = rule_rhs_length (&rules[r]);
+ nritems -= length + 1;
+ }
+ }
}
nonterminals_reduce (void)
{
int i, n;
- rule r;
/* Map the nonterminals to their new index: useful first, useless
afterwards. Kept for later report. */
short *nontermmap = XCALLOC (short, nvars) - ntokens;
n = ntokens;
for (i = ntokens; i < nsyms; i++)
- if (BITISSET (V, i))
+ if (bitset_test (V, i))
nontermmap[i] = n++;
for (i = ntokens; i < nsyms; i++)
- if (!BITISSET (V, i))
+ if (!bitset_test (V, i))
nontermmap[i] = n++;
/* Shuffle elements of tables indexed by symbol number. */
{
- bucket **symbols_sorted = XMALLOC (bucket *, nvars) - ntokens;
+ symbol_t **symbols_sorted = XMALLOC (symbol_t *, nvars) - ntokens;
+ for (i = ntokens; i < nsyms; i++)
+ symbols[i]->number = nontermmap[i];
for (i = ntokens; i < nsyms; i++)
symbols_sorted[nontermmap[i]] = symbols[i];
for (i = ntokens; i < nsyms; i++)
free (symbols_sorted + ntokens);
}
- /* Replace all symbol numbers in valid data structures. */
-
- for (i = 1; i <= nrules; i++)
- {
- rules[i].lhs = nontermmap[rules[i].lhs];
- if (ISVAR (rules[i].precsym))
- /* Can this happen? */
- rules[i].precsym = nontermmap[rules[i].precsym];
- }
-
for (i = 0; i < nritems; ++i)
if (ISVAR (ritem[i]))
ritem[i] = nontermmap[ritem[i]];
int i;
fprintf (out, "%s\n\n", _("Useless nonterminals:"));
for (i = 0; i < nuseless_nonterminals; ++i)
- fprintf (out, " %s\n", symbols[nsyms + i]->tag);
+ fprintf (out, " %s\n", quotearg_style (escape_quoting_style,
+ symbols[nsyms + i]->tag));
fputs ("\n\n", out);
}
bool b = FALSE;
int i;
for (i = 0; i < ntokens; i++)
- if (!BITISSET (V, i) && !BITISSET (V1, i))
+ if (!bitset_test (V, i) && !bitset_test (V1, i))
{
if (!b)
fprintf (out, "%s\n\n", _("Terminals which are not used:"));
b = TRUE;
- fprintf (out, " %s\n", symbols[i]->tag);
+ fprintf (out, " %s\n", quotearg_style (escape_quoting_style,
+ symbols[i]->tag));
}
if (b)
fputs ("\n\n", out);
{
int i;
fprintf (out, "%s\n\n", _("Useless rules:"));
- for (i = 1; i <= nrules; i++)
- if (!rules[i].useful)
- {
- rule r;
- fprintf (out, "#%-4d ", i - 1);
- fprintf (out, "%s:", symbols[rules[i].lhs]->tag);
- for (r = &ritem[rules[i].rhs]; *r >= 0; r++)
- fprintf (out, " %s", symbols[*r]->tag);
- fputs (";\n", out);
- }
+ for (i = nrules + 1; i < nuseless_productions + nrules + 1; i++)
+ {
+ item_number_t *r;
+ fprintf (out, "#%-4d ", rules[i].user_number - 1);
+ fprintf (out, "%s:", quotearg_style (escape_quoting_style,
+ rules[i].lhs->tag));
+ for (r = rules[i].rhs; *r >= 0; r++)
+ fprintf (out, " %s", quotearg_style (escape_quoting_style,
+ symbols[*r]->tag));
+ fputs (";\n", out);
+ }
fputs ("\n\n", out);
}
}
dump_grammar (FILE *out)
{
int i;
- rule r;
+ item_number_t *r;
fprintf (out, "REDUCED GRAMMAR\n\n");
fprintf (out,
- "ntokens = %d, nvars = %d, nsyms = %d, nrules = %d, nitems = %d\n\n",
- ntokens, nvars, nsyms, nrules, nitems);
+ "ntokens = %d, nvars = %d, nsyms = %d, nrules = %d, nritems = %d\n\n",
+ ntokens, nvars, nsyms, nrules, nritems);
fprintf (out, "Variables\n---------\n\n");
fprintf (out, "Value Sprec Sassoc Tag\n");
for (i = ntokens; i < nsyms; i++)
fprintf (out, "%5d %5d %5d %s\n",
i,
- symbols[i]->prec, symbols[i]->assoc, symbols[i]->tag);
+ symbols[i]->prec, symbols[i]->assoc,
+ quotearg_style (escape_quoting_style, symbols[i]->tag));
fprintf (out, "\n\n");
fprintf (out, "Rules\n-----\n\n");
fprintf (out, "Num (Prec, Assoc, Useful, Ritem Range) Lhs -> Rhs (Ritem range) [Num]\n");
- for (i = 1; i <= nrules; i++)
+ for (i = 1; i < nrules + nuseless_productions + 1; i++)
{
int rhs_count = 0;
/* Find the last RHS index in ritems. */
- for (r = &ritem[rules[i].rhs]; *r >= 0; ++r)
+ for (r = rules[i].rhs; *r >= 0; ++r)
++rhs_count;
fprintf (out, "%3d (%2d, %2d, %2d, %2d-%2d) %2d ->",
i - 1,
- rules[i].prec, rules[i].assoc, rules[i].useful,
- rules[i].rhs, rules[i].rhs + rhs_count - 1,
- rules[i].lhs);
+ rules[i].prec->prec,
+ rules[i].prec->assoc,
+ rules[i].useful,
+ rules[i].rhs - ritem,
+ rules[i].rhs - ritem + rhs_count - 1,
+ rules[i].lhs->number);
/* Dumped the RHS. */
- for (r = &ritem[rules[i].rhs]; *r >= 0; r++)
+ for (r = rules[i].rhs; *r >= 0; r++)
fprintf (out, "%3d", *r);
fprintf (out, " [%d]\n", -(*r) - 1);
}
fprintf (out, "\n\n");
fprintf (out, "Rules interpreted\n-----------------\n\n");
- for (i = 1; i <= nrules; i++)
+ for (i = 1; i < nrules + nuseless_productions + 1; i++)
{
- fprintf (out, "%-5d %s :", i, symbols[rules[i].lhs]->tag);
- for (r = &ritem[rules[i].rhs]; *r >= 0; r++)
- fprintf (out, " %s", symbols[*r]->tag);
+ fprintf (out, "%-5d %s :",
+ i, quotearg_style (escape_quoting_style, rules[i].lhs->tag));
+ for (r = rules[i].rhs; *r >= 0; r++)
+ fprintf (out, " %s",
+ quotearg_style (escape_quoting_style, symbols[*r]->tag));
fputc ('\n', out);
}
fprintf (out, "\n\n");
/* Allocate the global sets used to compute the reduced grammar */
- N = XCALLOC (unsigned, WORDSIZE (nvars));
- P = XCALLOC (unsigned, WORDSIZE (nrules + 1));
- V = XCALLOC (unsigned, WORDSIZE (nsyms));
- V1 = XCALLOC (unsigned, WORDSIZE (nsyms));
+ N = bitset_create (nvars, BITSET_FIXED);
+ P = bitset_create (nrules + 1, BITSET_FIXED);
+ V = bitset_create (nsyms, BITSET_FIXED);
+ V1 = bitset_create (nsyms, BITSET_FIXED);
useless_nonterminals ();
inaccessable_symbols ();
reduced = (bool) (nuseless_nonterminals + nuseless_productions > 0);
-
if (!reduced)
return;
reduce_print ();
- if (!BITISSET (N, start_symbol - ntokens))
+ if (!bitset_test (N, start_symbol - ntokens))
fatal (_("Start symbol %s does not derive any sentence"),
- symbols[start_symbol]->tag);
+ quotearg_style (escape_quoting_style, symbols[start_symbol]->tag));
- reduce_grammar_tables ();
+ /* First reduce the nonterminals, as they renumber themselves in the
+ whole grammar. If you change the order, nonterms would be
+ renumbered only in the reduced grammar. */
if (nuseless_nonterminals > 0)
nonterminals_reduce ();
+ if (nuseless_productions > 0)
+ reduce_grammar_tables ();
if (trace_flag)
{
void
reduce_free (void)
{
- XFREE (N);
- XFREE (V);
- XFREE (V1);
- XFREE (P);
+ bitset_free (N);
+ bitset_free (V);
+ bitset_free (V1);
+ bitset_free (P);
}
projects / bison.git / blobdiff