]>
Commit | Line | Data |
---|---|---|
1 | /* Generate the nondeterministic finite state machine for bison, | |
2 | Copyright 1984, 1986, 1989, 2000 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 | /* See comments in state.h for the data structures that represent it. | |
23 | The entry point is generate_states. */ | |
24 | ||
25 | #include "system.h" | |
26 | #include "gram.h" | |
27 | #include "state.h" | |
28 | #include "complain.h" | |
29 | #include "closure.h" | |
30 | #include "LR0.h" | |
31 | ||
32 | ||
33 | int nstates; | |
34 | int final_state; | |
35 | core *first_state = NULL; | |
36 | shifts *first_shift = NULL; | |
37 | reductions *first_reduction = NULL; | |
38 | ||
39 | static core *this_state = NULL; | |
40 | static core *last_state = NULL; | |
41 | static shifts *last_shift = NULL; | |
42 | static reductions *last_reduction = NULL; | |
43 | ||
44 | static int nshifts; | |
45 | static short *shift_symbol = NULL; | |
46 | ||
47 | static short *redset = NULL; | |
48 | static short *shiftset = NULL; | |
49 | ||
50 | static short **kernel_base = NULL; | |
51 | static short **kernel_end = NULL; | |
52 | static short *kernel_items = NULL; | |
53 | ||
54 | /* hash table for states, to recognize equivalent ones. */ | |
55 | ||
56 | #define STATE_TABLE_SIZE 1009 | |
57 | static core **state_table = NULL; | |
58 | ||
59 | \f | |
60 | static void | |
61 | allocate_itemsets (void) | |
62 | { | |
63 | short *itemp = NULL; | |
64 | int symbol; | |
65 | int i; | |
66 | int count; | |
67 | short *symbol_count = NULL; | |
68 | ||
69 | count = 0; | |
70 | symbol_count = XCALLOC (short, nsyms); | |
71 | ||
72 | itemp = ritem; | |
73 | symbol = *itemp++; | |
74 | while (symbol) | |
75 | { | |
76 | if (symbol > 0) | |
77 | { | |
78 | count++; | |
79 | symbol_count[symbol]++; | |
80 | } | |
81 | symbol = *itemp++; | |
82 | } | |
83 | ||
84 | /* See comments before new_itemsets. All the vectors of items | |
85 | live inside KERNEL_ITEMS. The number of active items after | |
86 | some symbol cannot be more than the number of times that symbol | |
87 | appears as an item, which is symbol_count[symbol]. | |
88 | We allocate that much space for each symbol. */ | |
89 | ||
90 | kernel_base = XCALLOC (short *, nsyms); | |
91 | if (count) | |
92 | kernel_items = XCALLOC (short, count); | |
93 | ||
94 | count = 0; | |
95 | for (i = 0; i < nsyms; i++) | |
96 | { | |
97 | kernel_base[i] = kernel_items + count; | |
98 | count += symbol_count[i]; | |
99 | } | |
100 | ||
101 | shift_symbol = symbol_count; | |
102 | kernel_end = XCALLOC (short *, nsyms); | |
103 | } | |
104 | ||
105 | ||
106 | static void | |
107 | allocate_storage (void) | |
108 | { | |
109 | allocate_itemsets (); | |
110 | ||
111 | shiftset = XCALLOC (short, nsyms); | |
112 | redset = XCALLOC (short, nrules + 1); | |
113 | state_table = XCALLOC (core *, STATE_TABLE_SIZE); | |
114 | } | |
115 | ||
116 | ||
117 | static void | |
118 | free_storage (void) | |
119 | { | |
120 | XFREE (shift_symbol); | |
121 | XFREE (redset); | |
122 | XFREE (shiftset); | |
123 | XFREE (kernel_base); | |
124 | XFREE (kernel_end); | |
125 | XFREE (kernel_items); | |
126 | XFREE (state_table); | |
127 | } | |
128 | ||
129 | ||
130 | ||
131 | ||
132 | /*----------------------------------------------------------------. | |
133 | | Find which symbols can be shifted in the current state, and for | | |
134 | | each one record which items would be active after that shift. | | |
135 | | Uses the contents of itemset. | | |
136 | | | | |
137 | | shift_symbol is set to a vector of the symbols that can be | | |
138 | | shifted. For each symbol in the grammar, kernel_base[symbol] | | |
139 | | points to a vector of item numbers activated if that symbol is | | |
140 | | shifted, and kernel_end[symbol] points after the end of that | | |
141 | | vector. | | |
142 | `----------------------------------------------------------------*/ | |
143 | ||
144 | static void | |
145 | new_itemsets (void) | |
146 | { | |
147 | int i; | |
148 | int shiftcount; | |
149 | short *isp; | |
150 | short *ksp; | |
151 | int symbol; | |
152 | ||
153 | #if TRACE | |
154 | fprintf (stderr, "Entering new_itemsets\n"); | |
155 | #endif | |
156 | ||
157 | for (i = 0; i < nsyms; i++) | |
158 | kernel_end[i] = NULL; | |
159 | ||
160 | shiftcount = 0; | |
161 | ||
162 | isp = itemset; | |
163 | ||
164 | while (isp < itemsetend) | |
165 | { | |
166 | i = *isp++; | |
167 | symbol = ritem[i]; | |
168 | if (symbol > 0) | |
169 | { | |
170 | ksp = kernel_end[symbol]; | |
171 | ||
172 | if (!ksp) | |
173 | { | |
174 | shift_symbol[shiftcount++] = symbol; | |
175 | ksp = kernel_base[symbol]; | |
176 | } | |
177 | ||
178 | *ksp++ = i + 1; | |
179 | kernel_end[symbol] = ksp; | |
180 | } | |
181 | } | |
182 | ||
183 | nshifts = shiftcount; | |
184 | } | |
185 | ||
186 | ||
187 | ||
188 | /*-----------------------------------------------------------------. | |
189 | | Subroutine of get_state. Create a new state for those items, if | | |
190 | | necessary. | | |
191 | `-----------------------------------------------------------------*/ | |
192 | ||
193 | static core * | |
194 | new_state (int symbol) | |
195 | { | |
196 | int n; | |
197 | core *p; | |
198 | short *isp1; | |
199 | short *isp2; | |
200 | short *iend; | |
201 | ||
202 | #if TRACE | |
203 | fprintf (stderr, "Entering new_state, symbol = %d\n", symbol); | |
204 | #endif | |
205 | ||
206 | if (nstates >= MAXSHORT) | |
207 | fatal (_("too many states (max %d)"), MAXSHORT); | |
208 | ||
209 | isp1 = kernel_base[symbol]; | |
210 | iend = kernel_end[symbol]; | |
211 | n = iend - isp1; | |
212 | ||
213 | p = | |
214 | (core *) xcalloc ((unsigned) (sizeof (core) + (n - 1) * sizeof (short)), 1); | |
215 | p->accessing_symbol = symbol; | |
216 | p->number = nstates; | |
217 | p->nitems = n; | |
218 | ||
219 | isp2 = p->items; | |
220 | while (isp1 < iend) | |
221 | *isp2++ = *isp1++; | |
222 | ||
223 | last_state->next = p; | |
224 | last_state = p; | |
225 | ||
226 | nstates++; | |
227 | ||
228 | return p; | |
229 | } | |
230 | ||
231 | ||
232 | /*--------------------------------------------------------------. | |
233 | | Find the state number for the state we would get to (from the | | |
234 | | current state) by shifting symbol. Create a new state if no | | |
235 | | equivalent one exists already. Used by append_states. | | |
236 | `--------------------------------------------------------------*/ | |
237 | ||
238 | static int | |
239 | get_state (int symbol) | |
240 | { | |
241 | int key; | |
242 | short *isp1; | |
243 | short *isp2; | |
244 | short *iend; | |
245 | core *sp; | |
246 | int found; | |
247 | ||
248 | int n; | |
249 | ||
250 | #if TRACE | |
251 | fprintf (stderr, "Entering get_state, symbol = %d\n", symbol); | |
252 | #endif | |
253 | ||
254 | isp1 = kernel_base[symbol]; | |
255 | iend = kernel_end[symbol]; | |
256 | n = iend - isp1; | |
257 | ||
258 | /* add up the target state's active item numbers to get a hash key */ | |
259 | key = 0; | |
260 | while (isp1 < iend) | |
261 | key += *isp1++; | |
262 | ||
263 | key = key % STATE_TABLE_SIZE; | |
264 | ||
265 | sp = state_table[key]; | |
266 | ||
267 | if (sp) | |
268 | { | |
269 | found = 0; | |
270 | while (!found) | |
271 | { | |
272 | if (sp->nitems == n) | |
273 | { | |
274 | found = 1; | |
275 | isp1 = kernel_base[symbol]; | |
276 | isp2 = sp->items; | |
277 | ||
278 | while (found && isp1 < iend) | |
279 | { | |
280 | if (*isp1++ != *isp2++) | |
281 | found = 0; | |
282 | } | |
283 | } | |
284 | ||
285 | if (!found) | |
286 | { | |
287 | if (sp->link) | |
288 | { | |
289 | sp = sp->link; | |
290 | } | |
291 | else /* bucket exhausted and no match */ | |
292 | { | |
293 | sp = sp->link = new_state (symbol); | |
294 | found = 1; | |
295 | } | |
296 | } | |
297 | } | |
298 | } | |
299 | else /* bucket is empty */ | |
300 | { | |
301 | state_table[key] = sp = new_state (symbol); | |
302 | } | |
303 | ||
304 | return sp->number; | |
305 | } | |
306 | ||
307 | /*------------------------------------------------------------------. | |
308 | | Use the information computed by new_itemsets to find the state | | |
309 | | numbers reached by each shift transition from the current state. | | |
310 | | | | |
311 | | shiftset is set up as a vector of state numbers of those states. | | |
312 | `------------------------------------------------------------------*/ | |
313 | ||
314 | static void | |
315 | append_states (void) | |
316 | { | |
317 | int i; | |
318 | int j; | |
319 | int symbol; | |
320 | ||
321 | #if TRACE | |
322 | fprintf (stderr, "Entering append_states\n"); | |
323 | #endif | |
324 | ||
325 | /* first sort shift_symbol into increasing order */ | |
326 | ||
327 | for (i = 1; i < nshifts; i++) | |
328 | { | |
329 | symbol = shift_symbol[i]; | |
330 | j = i; | |
331 | while (j > 0 && shift_symbol[j - 1] > symbol) | |
332 | { | |
333 | shift_symbol[j] = shift_symbol[j - 1]; | |
334 | j--; | |
335 | } | |
336 | shift_symbol[j] = symbol; | |
337 | } | |
338 | ||
339 | for (i = 0; i < nshifts; i++) | |
340 | { | |
341 | symbol = shift_symbol[i]; | |
342 | shiftset[i] = get_state (symbol); | |
343 | } | |
344 | } | |
345 | ||
346 | ||
347 | static void | |
348 | new_states (void) | |
349 | { | |
350 | core *p; | |
351 | ||
352 | p = (core *) xcalloc ((unsigned) (sizeof (core) - sizeof (short)), 1); | |
353 | first_state = last_state = this_state = p; | |
354 | nstates = 1; | |
355 | } | |
356 | ||
357 | ||
358 | static void | |
359 | save_shifts (void) | |
360 | { | |
361 | shifts *p; | |
362 | short *sp1; | |
363 | short *sp2; | |
364 | short *send; | |
365 | ||
366 | p = (shifts *) xcalloc ((unsigned) (sizeof (shifts) + | |
367 | (nshifts - 1) * sizeof (short)), 1); | |
368 | ||
369 | p->number = this_state->number; | |
370 | p->nshifts = nshifts; | |
371 | ||
372 | sp1 = shiftset; | |
373 | sp2 = p->shifts; | |
374 | send = shiftset + nshifts; | |
375 | ||
376 | while (sp1 < send) | |
377 | *sp2++ = *sp1++; | |
378 | ||
379 | if (last_shift) | |
380 | { | |
381 | last_shift->next = p; | |
382 | last_shift = p; | |
383 | } | |
384 | else | |
385 | { | |
386 | first_shift = p; | |
387 | last_shift = p; | |
388 | } | |
389 | } | |
390 | ||
391 | ||
392 | /*------------------------------------------------------------------. | |
393 | | Subroutine of augment_automaton. Create the next-to-final state, | | |
394 | | to which a shift has already been made in the initial state. | | |
395 | `------------------------------------------------------------------*/ | |
396 | ||
397 | static void | |
398 | insert_start_shift (void) | |
399 | { | |
400 | core *statep; | |
401 | shifts *sp; | |
402 | ||
403 | statep = (core *) xcalloc ((unsigned) (sizeof (core) - sizeof (short)), 1); | |
404 | statep->number = nstates; | |
405 | statep->accessing_symbol = start_symbol; | |
406 | ||
407 | last_state->next = statep; | |
408 | last_state = statep; | |
409 | ||
410 | /* Make a shift from this state to (what will be) the final state. */ | |
411 | sp = XCALLOC (shifts, 1); | |
412 | sp->number = nstates++; | |
413 | sp->nshifts = 1; | |
414 | sp->shifts[0] = nstates; | |
415 | ||
416 | last_shift->next = sp; | |
417 | last_shift = sp; | |
418 | } | |
419 | ||
420 | ||
421 | /*------------------------------------------------------------------. | |
422 | | Make sure that the initial state has a shift that accepts the | | |
423 | | grammar's start symbol and goes to the next-to-final state, which | | |
424 | | has a shift going to the final state, which has a shift to the | | |
425 | | termination state. Create such states and shifts if they don't | | |
426 | | happen to exist already. | | |
427 | `------------------------------------------------------------------*/ | |
428 | ||
429 | static void | |
430 | augment_automaton (void) | |
431 | { | |
432 | int i; | |
433 | int k; | |
434 | core *statep; | |
435 | shifts *sp; | |
436 | shifts *sp2; | |
437 | shifts *sp1 = NULL; | |
438 | ||
439 | sp = first_shift; | |
440 | ||
441 | if (sp) | |
442 | { | |
443 | if (sp->number == 0) | |
444 | { | |
445 | k = sp->nshifts; | |
446 | statep = first_state->next; | |
447 | ||
448 | /* The states reached by shifts from first_state are numbered 1...K. | |
449 | Look for one reached by start_symbol. */ | |
450 | while (statep->accessing_symbol < start_symbol | |
451 | && statep->number < k) | |
452 | statep = statep->next; | |
453 | ||
454 | if (statep->accessing_symbol == start_symbol) | |
455 | { | |
456 | /* We already have a next-to-final state. | |
457 | Make sure it has a shift to what will be the final state. */ | |
458 | k = statep->number; | |
459 | ||
460 | while (sp && sp->number < k) | |
461 | { | |
462 | sp1 = sp; | |
463 | sp = sp->next; | |
464 | } | |
465 | ||
466 | if (sp && sp->number == k) | |
467 | { | |
468 | sp2 = (shifts *) xcalloc ((unsigned) (sizeof (shifts) | |
469 | + | |
470 | sp->nshifts * | |
471 | sizeof (short)), 1); | |
472 | sp2->number = k; | |
473 | sp2->nshifts = sp->nshifts + 1; | |
474 | sp2->shifts[0] = nstates; | |
475 | for (i = sp->nshifts; i > 0; i--) | |
476 | sp2->shifts[i] = sp->shifts[i - 1]; | |
477 | ||
478 | /* Patch sp2 into the chain of shifts in place of sp, | |
479 | following sp1. */ | |
480 | sp2->next = sp->next; | |
481 | sp1->next = sp2; | |
482 | if (sp == last_shift) | |
483 | last_shift = sp2; | |
484 | XFREE (sp); | |
485 | } | |
486 | else | |
487 | { | |
488 | sp2 = XCALLOC (shifts, 1); | |
489 | sp2->number = k; | |
490 | sp2->nshifts = 1; | |
491 | sp2->shifts[0] = nstates; | |
492 | ||
493 | /* Patch sp2 into the chain of shifts between sp1 and sp. */ | |
494 | sp2->next = sp; | |
495 | sp1->next = sp2; | |
496 | if (sp == 0) | |
497 | last_shift = sp2; | |
498 | } | |
499 | } | |
500 | else | |
501 | { | |
502 | /* There is no next-to-final state as yet. */ | |
503 | /* Add one more shift in first_shift, | |
504 | going to the next-to-final state (yet to be made). */ | |
505 | sp = first_shift; | |
506 | ||
507 | sp2 = (shifts *) xcalloc (sizeof (shifts) | |
508 | + sp->nshifts * sizeof (short), 1); | |
509 | sp2->nshifts = sp->nshifts + 1; | |
510 | ||
511 | /* Stick this shift into the vector at the proper place. */ | |
512 | statep = first_state->next; | |
513 | for (k = 0, i = 0; i < sp->nshifts; k++, i++) | |
514 | { | |
515 | if (statep->accessing_symbol > start_symbol && i == k) | |
516 | sp2->shifts[k++] = nstates; | |
517 | sp2->shifts[k] = sp->shifts[i]; | |
518 | statep = statep->next; | |
519 | } | |
520 | if (i == k) | |
521 | sp2->shifts[k++] = nstates; | |
522 | ||
523 | /* Patch sp2 into the chain of shifts | |
524 | in place of sp, at the beginning. */ | |
525 | sp2->next = sp->next; | |
526 | first_shift = sp2; | |
527 | if (last_shift == sp) | |
528 | last_shift = sp2; | |
529 | ||
530 | XFREE (sp); | |
531 | ||
532 | /* Create the next-to-final state, with shift to | |
533 | what will be the final state. */ | |
534 | insert_start_shift (); | |
535 | } | |
536 | } | |
537 | else | |
538 | { | |
539 | /* The initial state didn't even have any shifts. | |
540 | Give it one shift, to the next-to-final state. */ | |
541 | sp = XCALLOC (shifts, 1); | |
542 | sp->nshifts = 1; | |
543 | sp->shifts[0] = nstates; | |
544 | ||
545 | /* Patch sp into the chain of shifts at the beginning. */ | |
546 | sp->next = first_shift; | |
547 | first_shift = sp; | |
548 | ||
549 | /* Create the next-to-final state, with shift to | |
550 | what will be the final state. */ | |
551 | insert_start_shift (); | |
552 | } | |
553 | } | |
554 | else | |
555 | { | |
556 | /* There are no shifts for any state. | |
557 | Make one shift, from the initial state to the next-to-final state. */ | |
558 | ||
559 | sp = XCALLOC (shifts, 1); | |
560 | sp->nshifts = 1; | |
561 | sp->shifts[0] = nstates; | |
562 | ||
563 | /* Initialize the chain of shifts with sp. */ | |
564 | first_shift = sp; | |
565 | last_shift = sp; | |
566 | ||
567 | /* Create the next-to-final state, with shift to | |
568 | what will be the final state. */ | |
569 | insert_start_shift (); | |
570 | } | |
571 | ||
572 | /* Make the final state--the one that follows a shift from the | |
573 | next-to-final state. | |
574 | The symbol for that shift is 0 (end-of-file). */ | |
575 | statep = (core *) xcalloc ((unsigned) (sizeof (core) - sizeof (short)), 1); | |
576 | statep->number = nstates; | |
577 | last_state->next = statep; | |
578 | last_state = statep; | |
579 | ||
580 | /* Make the shift from the final state to the termination state. */ | |
581 | sp = XCALLOC (shifts, 1); | |
582 | sp->number = nstates++; | |
583 | sp->nshifts = 1; | |
584 | sp->shifts[0] = nstates; | |
585 | last_shift->next = sp; | |
586 | last_shift = sp; | |
587 | ||
588 | /* Note that the variable `final_state' refers to what we sometimes call | |
589 | the termination state. */ | |
590 | final_state = nstates; | |
591 | ||
592 | /* Make the termination state. */ | |
593 | statep = (core *) xcalloc ((unsigned) (sizeof (core) - sizeof (short)), 1); | |
594 | statep->number = nstates++; | |
595 | last_state->next = statep; | |
596 | last_state = statep; | |
597 | } | |
598 | ||
599 | ||
600 | /*----------------------------------------------------------------. | |
601 | | Find which rules can be used for reduction transitions from the | | |
602 | | current state and make a reductions structure for the state to | | |
603 | | record their rule numbers. | | |
604 | `----------------------------------------------------------------*/ | |
605 | ||
606 | static void | |
607 | save_reductions (void) | |
608 | { | |
609 | short *isp; | |
610 | short *rp1; | |
611 | short *rp2; | |
612 | int item; | |
613 | int count; | |
614 | reductions *p; | |
615 | ||
616 | short *rend; | |
617 | ||
618 | /* Find and count the active items that represent ends of rules. */ | |
619 | ||
620 | count = 0; | |
621 | for (isp = itemset; isp < itemsetend; isp++) | |
622 | { | |
623 | item = ritem[*isp]; | |
624 | if (item < 0) | |
625 | redset[count++] = -item; | |
626 | } | |
627 | ||
628 | /* Make a reductions structure and copy the data into it. */ | |
629 | ||
630 | if (count) | |
631 | { | |
632 | p = (reductions *) xcalloc ((unsigned) (sizeof (reductions) + | |
633 | (count - 1) * sizeof (short)), 1); | |
634 | ||
635 | p->number = this_state->number; | |
636 | p->nreds = count; | |
637 | ||
638 | rp1 = redset; | |
639 | rp2 = p->rules; | |
640 | rend = rp1 + count; | |
641 | ||
642 | for (/* nothing */; rp1 < rend; ++rp1, ++rp2) | |
643 | *rp2 = *rp1; | |
644 | ||
645 | if (last_reduction) | |
646 | { | |
647 | last_reduction->next = p; | |
648 | last_reduction = p; | |
649 | } | |
650 | else | |
651 | { | |
652 | first_reduction = p; | |
653 | last_reduction = p; | |
654 | } | |
655 | } | |
656 | } | |
657 | ||
658 | \f | |
659 | /*-------------------------------------------------------------------. | |
660 | | Compute the nondeterministic finite state machine (see state.h for | | |
661 | | details) from the grammar. | | |
662 | `-------------------------------------------------------------------*/ | |
663 | ||
664 | void | |
665 | generate_states (void) | |
666 | { | |
667 | allocate_storage (); | |
668 | new_closure (nitems); | |
669 | new_states (); | |
670 | ||
671 | while (this_state) | |
672 | { | |
673 | /* Set up ruleset and itemset for the transitions out of this | |
674 | state. ruleset gets a 1 bit for each rule that could reduce | |
675 | now. itemset gets a vector of all the items that could be | |
676 | accepted next. */ | |
677 | closure (this_state->items, this_state->nitems); | |
678 | /* record the reductions allowed out of this state */ | |
679 | save_reductions (); | |
680 | /* find the itemsets of the states that shifts can reach */ | |
681 | new_itemsets (); | |
682 | /* find or create the core structures for those states */ | |
683 | append_states (); | |
684 | ||
685 | /* create the shifts structures for the shifts to those states, | |
686 | now that the state numbers transitioning to are known */ | |
687 | if (nshifts > 0) | |
688 | save_shifts (); | |
689 | ||
690 | /* states are queued when they are created; process them all */ | |
691 | this_state = this_state->next; | |
692 | } | |
693 | ||
694 | /* discard various storage */ | |
695 | free_closure (); | |
696 | free_storage (); | |
697 | ||
698 | /* set up initial and final states as parser wants them */ | |
699 | augment_automaton (); | |
700 | } |