]> git.saurik.com Git - bison.git/blob - src/reduce.c
* src/derives.c (print_derives): Be sure to use `>= 0', not `> 0',
[bison.git] / src / reduce.c
1 /* Grammar reduction for Bison.
2 Copyright 1988, 1989, 2000, 2001 Free Software Foundation, Inc.
3
4 This file is part of Bison, the GNU Compiler Compiler.
5
6 Bison is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2, or (at your option)
9 any later version.
10
11 Bison is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
15
16 You should have received a copy of the GNU General Public License
17 along with Bison; see the file COPYING. If not, write to
18 the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 Boston, MA 02111-1307, USA. */
20
21
22 /* Reduce the grammar: Find and eliminate unreachable terminals,
23 nonterminals, and productions. David S. Bakin. */
24
25 /* Don't eliminate unreachable terminals: They may be used by the
26 user's parser. */
27
28 #include "system.h"
29 #include "getargs.h"
30 #include "files.h"
31 #include "gram.h"
32 #include "complain.h"
33 #include "reduce.h"
34 #include "reader.h"
35 #include "getargs.h"
36
37 typedef unsigned *BSet;
38 typedef short *rule;
39
40
41 /* Set of all nonterminals which are not useless. */
42 static BSet N;
43
44 /* Set of all rules which have no useless nonterminals in their RHS. */
45 static BSet P;
46
47 /* Set of all accessible symbols. */
48 static BSet V;
49
50 /* Set of symbols used to define rule precedence (so they are
51 `useless', but no warning should be issued). */
52 static BSet V1;
53
54 static int nuseful_productions;
55 static int nuseless_productions;
56 static int nuseful_nonterminals;
57 int nuseless_nonterminals;
58 \f
59 static bool
60 bits_equal (BSet L, BSet R, int n)
61 {
62 int i;
63
64 for (i = n - 1; i >= 0; i--)
65 if (L[i] != R[i])
66 return FALSE;
67 return TRUE;
68 }
69
70
71 static int
72 nbits (unsigned i)
73 {
74 int count = 0;
75
76 while (i != 0)
77 {
78 i ^= (i & ((unsigned) (-(int) i)));
79 ++count;
80 }
81 return count;
82 }
83
84
85 static int
86 bits_size (BSet S, int n)
87 {
88 int i, count = 0;
89
90 for (i = n - 1; i >= 0; i--)
91 count += nbits (S[i]);
92 return count;
93 }
94 \f
95 /*-------------------------------------------------------------------.
96 | Another way to do this would be with a set for each production and |
97 | then do subset tests against N0, but even for the C grammar the |
98 | whole reducing process takes only 2 seconds on my 8Mhz AT. |
99 `-------------------------------------------------------------------*/
100
101 static bool
102 useful_production (int i, BSet N0)
103 {
104 rule r;
105 short n;
106
107 /* A production is useful if all of the nonterminals in its appear
108 in the set of useful nonterminals. */
109
110 for (r = &ritem[rule_table[i].rhs]; *r >= 0; r++)
111 if (ISVAR (n = *r))
112 if (!BITISSET (N0, n - ntokens))
113 return FALSE;
114 return TRUE;
115 }
116
117
118 /*---------------------------------------------------------.
119 | Remember that rules are 1-origin, symbols are 0-origin. |
120 `---------------------------------------------------------*/
121
122 static void
123 useless_nonterminals (void)
124 {
125 BSet Np, Ns;
126 int i;
127
128 /* N is set as built. Np is set being built this iteration. P is
129 set of all productions which have a RHS all in N. */
130
131 Np = XCALLOC (unsigned, WORDSIZE (nvars));
132
133 /* The set being computed is a set of nonterminals which can derive
134 the empty string or strings consisting of all terminals. At each
135 iteration a nonterminal is added to the set if there is a
136 production with that nonterminal as its LHS for which all the
137 nonterminals in its RHS are already in the set. Iterate until
138 the set being computed remains unchanged. Any nonterminals not
139 in the set at that point are useless in that they will never be
140 used in deriving a sentence of the language.
141
142 This iteration doesn't use any special traversal over the
143 productions. A set is kept of all productions for which all the
144 nonterminals in the RHS are in useful. Only productions not in
145 this set are scanned on each iteration. At the end, this set is
146 saved to be used when finding useful productions: only
147 productions in this set will appear in the final grammar. */
148
149 while (1)
150 {
151 for (i = WORDSIZE (nvars) - 1; i >= 0; i--)
152 Np[i] = N[i];
153 for (i = 1; i <= nrules; i++)
154 {
155 if (!BITISSET (P, i))
156 {
157 if (useful_production (i, N))
158 {
159 SETBIT (Np, rule_table[i].lhs - ntokens);
160 SETBIT (P, i);
161 }
162 }
163 }
164 if (bits_equal (N, Np, WORDSIZE (nvars)))
165 break;
166 Ns = Np;
167 Np = N;
168 N = Ns;
169 }
170 XFREE (N);
171 N = Np;
172 }
173
174
175 static void
176 inaccessable_symbols (void)
177 {
178 BSet Vp, Vs, Pp;
179 int i;
180 short t;
181 rule r;
182
183 /* Find out which productions are reachable and which symbols are
184 used. Starting with an empty set of productions and a set of
185 symbols which only has the start symbol in it, iterate over all
186 productions until the set of productions remains unchanged for an
187 iteration. For each production which has a LHS in the set of
188 reachable symbols, add the production to the set of reachable
189 productions, and add all of the nonterminals in the RHS of the
190 production to the set of reachable symbols.
191
192 Consider only the (partially) reduced grammar which has only
193 nonterminals in N and productions in P.
194
195 The result is the set P of productions in the reduced grammar,
196 and the set V of symbols in the reduced grammar.
197
198 Although this algorithm also computes the set of terminals which
199 are reachable, no terminal will be deleted from the grammar. Some
200 terminals might not be in the grammar but might be generated by
201 semantic routines, and so the user might want them available with
202 specified numbers. (Is this true?) However, the nonreachable
203 terminals are printed (if running in verbose mode) so that the
204 user can know. */
205
206 Vp = XCALLOC (unsigned, WORDSIZE (nsyms));
207 Pp = XCALLOC (unsigned, WORDSIZE (nrules + 1));
208
209 /* If the start symbol isn't useful, then nothing will be useful. */
210 if (BITISSET (N, start_symbol - ntokens))
211 {
212 SETBIT (V, start_symbol);
213
214 while (1)
215 {
216 for (i = WORDSIZE (nsyms) - 1; i >= 0; i--)
217 Vp[i] = V[i];
218 for (i = 1; i <= nrules; i++)
219 {
220 if (!BITISSET (Pp, i)
221 && BITISSET (P, i)
222 && BITISSET (V, rule_table[i].lhs))
223 {
224 for (r = &ritem[rule_table[i].rhs]; *r >= 0; r++)
225 if (ISTOKEN (t = *r) || BITISSET (N, t - ntokens))
226 SETBIT (Vp, t);
227 SETBIT (Pp, i);
228 }
229 }
230 if (bits_equal (V, Vp, WORDSIZE (nsyms)))
231 break;
232 Vs = Vp;
233 Vp = V;
234 V = Vs;
235 }
236 }
237
238 XFREE (V);
239 V = Vp;
240
241 /* Tokens 0, 1, and 2 are internal to Bison. Consider them useful. */
242 SETBIT (V, 0); /* end-of-input token */
243 SETBIT (V, 1); /* error token */
244 SETBIT (V, 2); /* some undefined token */
245
246 XFREE (P);
247 P = Pp;
248
249 nuseful_productions = bits_size (P, WORDSIZE (nrules + 1));
250 nuseless_productions = nrules - nuseful_productions;
251
252 nuseful_nonterminals = 0;
253 for (i = ntokens; i < nsyms; i++)
254 if (BITISSET (V, i))
255 nuseful_nonterminals++;
256 nuseless_nonterminals = nvars - nuseful_nonterminals;
257
258 /* A token that was used in %prec should not be warned about. */
259 for (i = 1; i < nrules; i++)
260 if (rule_table[i].precsym != 0)
261 SETBIT (V1, rule_table[i].precsym);
262 }
263
264 static void
265 reduce_grammar_tables (void)
266 {
267 /* This is turned off because we would need to change the numbers in
268 the case statements in the actions file.
269
270 We don't disable it via CPP so that it is still checked with the
271 rest of the code, to avoid its becoming completely obsolete.
272
273 FIXME: I think the comment above demonstrates this code must be
274 turned off for *semantic* parser, not in the general case. Try
275 to understand this better --akim. */
276
277 if (0)
278 /* remove useless productions */
279 if (nuseless_productions > 0)
280 {
281 short np, pn, ni, pi;
282
283 np = 0;
284 ni = 0;
285 for (pn = 1; pn <= nrules; pn++)
286 if (BITISSET (P, pn))
287 {
288 np++;
289 if (pn != np)
290 {
291 rule_table[np].lhs = rule_table[pn].lhs;
292 rule_table[np].line = rule_table[pn].line;
293 rule_table[np].prec = rule_table[pn].prec;
294 rule_table[np].assoc = rule_table[pn].assoc;
295 rule_table[np].rhs = rule_table[pn].rhs;
296 if (rule_table[np].rhs != ni)
297 {
298 pi = rule_table[np].rhs;
299 rule_table[np].rhs = ni;
300 while (ritem[pi] >= 0)
301 ritem[ni++] = ritem[pi++];
302 ritem[ni++] = -np;
303 }
304 }
305 else
306 {
307 while (ritem[ni++] >= 0)
308 /* Nothing. */;
309 }
310 }
311
312 ritem[ni] = 0;
313 nrules -= nuseless_productions;
314 nitems = ni;
315 nritems = ni;
316
317 /* Is it worth it to reduce the amount of memory for the
318 grammar? Probably not. */
319 }
320
321 /* Disable useless productions. */
322 if (nuseless_productions > 0)
323 {
324 int pn;
325 for (pn = 1; pn <= nrules; pn++)
326 rule_table[pn].useful = BITISSET (P, pn);
327 }
328 }
329
330
331 /*------------------------------.
332 | Remove useless nonterminals. |
333 `------------------------------*/
334
335 static void
336 nonterminals_reduce (void)
337 {
338 int i, n;
339 rule r;
340
341 /* Map the nonterminals to their new index: useful first, useless
342 afterwards. Kept for later report. */
343
344 short *nontermmap = XCALLOC (short, nvars) - ntokens;
345 n = ntokens;
346 for (i = ntokens; i < nsyms; i++)
347 if (BITISSET (V, i))
348 nontermmap[i] = n++;
349 for (i = ntokens; i < nsyms; i++)
350 if (!BITISSET (V, i))
351 nontermmap[i] = n++;
352
353
354 /* Shuffle elements of tables indexed by symbol number. */
355 {
356 short *sassoc_sorted = XMALLOC (short, nvars) - ntokens;
357 short *sprec_sorted = XMALLOC (short, nvars) - ntokens;
358 char **tags_sorted = XMALLOC (char *, nvars) - ntokens;
359
360 for (i = ntokens; i < nsyms; i++)
361 {
362 n = nontermmap[i];
363 sassoc_sorted[n] = sassoc[i];
364 sprec_sorted[n] = sprec[i];
365 tags_sorted[n] = tags[i];
366 }
367 for (i = ntokens; i < nsyms; i++)
368 {
369 sassoc[i] = sassoc_sorted[i];
370 sprec[i] = sprec_sorted[i];
371 tags[i] = tags_sorted[i];
372 }
373 free (sassoc_sorted + ntokens);
374 free (sprec_sorted + ntokens);
375 free (tags_sorted + ntokens);
376 }
377
378 /* Replace all symbol numbers in valid data structures. */
379
380 for (i = 1; i <= nrules; i++)
381 {
382 rule_table[i].lhs = nontermmap[rule_table[i].lhs];
383 if (ISVAR (rule_table[i].precsym))
384 /* Can this happen? */
385 rule_table[i].precsym = nontermmap[rule_table[i].precsym];
386 }
387
388 for (i = 0; i < nritems; ++i)
389 if (ISVAR (ritem[i]))
390 ritem[i] = nontermmap[ritem[i]];
391
392 start_symbol = nontermmap[start_symbol];
393
394 nsyms -= nuseless_nonterminals;
395 nvars -= nuseless_nonterminals;
396
397 free (nontermmap + ntokens);
398 }
399
400
401 /*------------------------------------------------------------------.
402 | Output the detailed results of the reductions. For FILE.output. |
403 `------------------------------------------------------------------*/
404
405 void
406 reduce_output (FILE *out)
407 {
408 if (nuseless_nonterminals > 0)
409 {
410 int i;
411 fprintf (out, "%s\n\n", _("Useless nonterminals:"));
412 for (i = 0; i < nuseless_nonterminals; ++i)
413 fprintf (out, " %s\n", tags[nsyms + i]);
414 fputs ("\n\n", out);
415 }
416
417 {
418 bool b = FALSE;
419 int i;
420 for (i = 0; i < ntokens; i++)
421 if (!BITISSET (V, i) && !BITISSET (V1, i))
422 {
423 if (!b)
424 fprintf (out, "%s\n\n", _("Terminals which are not used:"));
425 b = TRUE;
426 fprintf (out, " %s\n", tags[i]);
427 }
428 if (b)
429 fputs ("\n\n", out);
430 }
431
432 if (nuseless_productions > 0)
433 {
434 int i;
435 fprintf (out, "%s\n\n", _("Useless rules:"));
436 for (i = 1; i <= nrules; i++)
437 if (!rule_table[i].useful)
438 {
439 rule r;
440 fprintf (out, "#%-4d ", i - 1);
441 fprintf (out, "%s:", tags[rule_table[i].lhs]);
442 for (r = &ritem[rule_table[i].rhs]; *r >= 0; r++)
443 fprintf (out, " %s", tags[*r]);
444 fputs (";\n", out);
445 }
446 fputs ("\n\n", out);
447 }
448 }
449 \f
450 static void
451 dump_grammar (FILE *out)
452 {
453 int i;
454 rule r;
455
456 fprintf (out, "REDUCED GRAMMAR\n\n");
457 fprintf (out,
458 "ntokens = %d, nvars = %d, nsyms = %d, nrules = %d, nitems = %d\n\n",
459 ntokens, nvars, nsyms, nrules, nitems);
460 fprintf (out, "Variables\n---------\n\n");
461 fprintf (out, "Value Sprec Sassoc Tag\n");
462 for (i = ntokens; i < nsyms; i++)
463 fprintf (out, "%5d %5d %5d %s\n", i, sprec[i], sassoc[i], tags[i]);
464 fprintf (out, "\n\n");
465 fprintf (out, "Rules\n-----\n\n");
466 fprintf (out, "Num (Prec, Assoc, Useful, Ritem Range) Lhs -> Rhs (Ritem range) [Num]\n");
467 for (i = 1; i <= nrules; i++)
468 {
469 int rhs_count = 0;
470 /* Find the last RHS index in ritems. */
471 for (r = &ritem[rule_table[i].rhs]; *r >= 0; ++r)
472 ++rhs_count;
473 fprintf (out, "%3d (%2d, %2d, %2d, %2d-%2d) %2d ->",
474 i - 1,
475 rule_table[i].prec, rule_table[i].assoc, rule_table[i].useful,
476 rule_table[i].rhs, rule_table[i].rhs + rhs_count - 1,
477 rule_table[i].lhs);
478 /* Dumped the RHS. */
479 for (r = &ritem[rule_table[i].rhs]; *r >= 0; r++)
480 fprintf (out, "%3d", *r);
481 fprintf (out, " [%d]\n", -(*r) - 1);
482 }
483 fprintf (out, "\n\n");
484 fprintf (out, "Rules interpreted\n-----------------\n\n");
485 for (i = 1; i <= nrules; i++)
486 {
487 fprintf (out, "%-5d %s :", i, tags[rule_table[i].lhs]);
488 for (r = &ritem[rule_table[i].rhs]; *r >= 0; r++)
489 fprintf (out, " %s", tags[*r]);
490 fputc ('\n', out);
491 }
492 fprintf (out, "\n\n");
493 }
494
495
496
497 /*-------------------------------.
498 | Report the results to STDERR. |
499 `-------------------------------*/
500
501 static void
502 reduce_print (void)
503 {
504 if (yacc_flag && nuseless_productions)
505 fprintf (stderr, ngettext ("%d rule never reduced\n",
506 "%d rules never reduced\n",
507 nuseless_productions),
508 nuseless_productions);
509
510 fprintf (stderr, _("%s contains "), infile);
511
512 if (nuseless_nonterminals > 0)
513 fprintf (stderr, ngettext ("%d useless nonterminal",
514 "%d useless nonterminals",
515 nuseless_nonterminals),
516 nuseless_nonterminals);
517
518 if (nuseless_nonterminals > 0 && nuseless_productions > 0)
519 fprintf (stderr, _(" and "));
520
521 if (nuseless_productions > 0)
522 fprintf (stderr, ngettext ("%d useless rule",
523 "%d useless rules",
524 nuseless_productions),
525 nuseless_productions);
526 fprintf (stderr, "\n");
527 fflush (stderr);
528 }
529 \f
530 void
531 reduce_grammar (void)
532 {
533 bool reduced;
534
535 /* Allocate the global sets used to compute the reduced grammar */
536
537 N = XCALLOC (unsigned, WORDSIZE (nvars));
538 P = XCALLOC (unsigned, WORDSIZE (nrules + 1));
539 V = XCALLOC (unsigned, WORDSIZE (nsyms));
540 V1 = XCALLOC (unsigned, WORDSIZE (nsyms));
541
542 useless_nonterminals ();
543 inaccessable_symbols ();
544
545 reduced = (bool) (nuseless_nonterminals + nuseless_productions > 0);
546
547 if (!reduced)
548 return;
549
550 reduce_print ();
551
552 if (!BITISSET (N, start_symbol - ntokens))
553 fatal (_("Start symbol %s does not derive any sentence"),
554 tags[start_symbol]);
555
556 reduce_grammar_tables ();
557 if (nuseless_nonterminals > 0)
558 nonterminals_reduce ();
559
560 if (trace_flag)
561 {
562 dump_grammar (stderr);
563
564 fprintf (stderr, "reduced %s defines %d terminals, %d nonterminals\
565 , and %d productions.\n",
566 infile, ntokens, nvars, nrules);
567 }
568 }
569
570
571 /*-----------------------------------------------------------.
572 | Free the global sets used to compute the reduced grammar. |
573 `-----------------------------------------------------------*/
574
575 void
576 reduce_free (void)
577 {
578 XFREE (N);
579 XFREE (V);
580 XFREE (V1);
581 XFREE (P);
582 }