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