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