]>
Commit | Line | Data |
---|---|---|
1 | /* Compute look-ahead criteria for bison, | |
2 | Copyright (C) 1984, 1986, 1989, 2000, 2001, 2002 | |
3 | Free Software Foundation, Inc. | |
4 | ||
5 | This file is part of Bison, the GNU Compiler Compiler. | |
6 | ||
7 | Bison is free software; you can redistribute it and/or modify | |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 2, or (at your option) | |
10 | any later version. | |
11 | ||
12 | Bison is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | GNU General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with Bison; see the file COPYING. If not, write to | |
19 | the Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
20 | Boston, MA 02111-1307, USA. */ | |
21 | ||
22 | ||
23 | /* Compute how to make the finite state machine deterministic; find | |
24 | which rules need lookahead in each state, and which lookahead | |
25 | tokens they accept. */ | |
26 | ||
27 | #include "system.h" | |
28 | #include "bitset.h" | |
29 | #include "bitsetv.h" | |
30 | #include "relation.h" | |
31 | #include "quotearg.h" | |
32 | #include "symtab.h" | |
33 | #include "gram.h" | |
34 | #include "reader.h" | |
35 | #include "types.h" | |
36 | #include "LR0.h" | |
37 | #include "complain.h" | |
38 | #include "lalr.h" | |
39 | #include "nullable.h" | |
40 | #include "derives.h" | |
41 | #include "getargs.h" | |
42 | ||
43 | rule_t **LArule = NULL; | |
44 | bitsetv LA = NULL; | |
45 | size_t nLA; | |
46 | ||
47 | static int ngotos; | |
48 | short *goto_map = NULL; | |
49 | state_number_t *from_state = NULL; | |
50 | state_number_t *to_state = NULL; | |
51 | ||
52 | /* And for the famous F variable, which name is so descriptive that a | |
53 | comment is hardly needed. <grin>. */ | |
54 | static bitsetv F = NULL; | |
55 | ||
56 | static short **includes; | |
57 | static shorts **lookback; | |
58 | ||
59 | ||
60 | ||
61 | ||
62 | static void | |
63 | initialize_LA (void) | |
64 | { | |
65 | state_number_t i; | |
66 | int j; | |
67 | rule_t **np; | |
68 | ||
69 | /* Avoid having to special case 0. */ | |
70 | if (!nLA) | |
71 | nLA = 1; | |
72 | ||
73 | LA = bitsetv_create (nLA, ntokens, BITSET_FIXED); | |
74 | LArule = XCALLOC (rule_t *, nLA); | |
75 | lookback = XCALLOC (shorts *, nLA); | |
76 | ||
77 | np = LArule; | |
78 | for (i = 0; i < nstates; i++) | |
79 | if (!states[i]->consistent) | |
80 | for (j = 0; j < states[i]->reductions->num; j++) | |
81 | *np++ = &rules[states[i]->reductions->rules[j]]; | |
82 | } | |
83 | ||
84 | ||
85 | static void | |
86 | set_goto_map (void) | |
87 | { | |
88 | state_number_t state; | |
89 | short *temp_map; | |
90 | ||
91 | goto_map = XCALLOC (short, nvars + 1) - ntokens; | |
92 | temp_map = XCALLOC (short, nvars + 1) - ntokens; | |
93 | ||
94 | ngotos = 0; | |
95 | for (state = 0; state < nstates; ++state) | |
96 | { | |
97 | transitions_t *sp = states[state]->shifts; | |
98 | int i; | |
99 | for (i = sp->num - 1; i >= 0 && TRANSITION_IS_GOTO (sp, i); --i) | |
100 | { | |
101 | if (ngotos == SHRT_MAX) | |
102 | fatal (_("too many gotos (max %d)"), SHRT_MAX); | |
103 | ||
104 | ngotos++; | |
105 | goto_map[TRANSITION_SYMBOL (sp, i)]++; | |
106 | } | |
107 | } | |
108 | ||
109 | { | |
110 | int k = 0; | |
111 | int i; | |
112 | for (i = ntokens; i < nsyms; i++) | |
113 | { | |
114 | temp_map[i] = k; | |
115 | k += goto_map[i]; | |
116 | } | |
117 | ||
118 | for (i = ntokens; i < nsyms; i++) | |
119 | goto_map[i] = temp_map[i]; | |
120 | ||
121 | goto_map[nsyms] = ngotos; | |
122 | temp_map[nsyms] = ngotos; | |
123 | } | |
124 | ||
125 | from_state = XCALLOC (state_number_t, ngotos); | |
126 | to_state = XCALLOC (state_number_t, ngotos); | |
127 | ||
128 | for (state = 0; state < nstates; ++state) | |
129 | { | |
130 | transitions_t *sp = states[state]->shifts; | |
131 | int i; | |
132 | for (i = sp->num - 1; i >= 0 && TRANSITION_IS_GOTO (sp, i); --i) | |
133 | { | |
134 | int k = temp_map[TRANSITION_SYMBOL (sp, i)]++; | |
135 | from_state[k] = state; | |
136 | to_state[k] = sp->states[i]; | |
137 | } | |
138 | } | |
139 | ||
140 | XFREE (temp_map + ntokens); | |
141 | } | |
142 | ||
143 | ||
144 | ||
145 | /*----------------------------------------------------------. | |
146 | | Map a state/symbol pair into its numeric representation. | | |
147 | `----------------------------------------------------------*/ | |
148 | ||
149 | static int | |
150 | map_goto (state_number_t state, symbol_number_t symbol) | |
151 | { | |
152 | int high; | |
153 | int low; | |
154 | int middle; | |
155 | state_number_t s; | |
156 | ||
157 | low = goto_map[symbol]; | |
158 | high = goto_map[symbol + 1] - 1; | |
159 | ||
160 | while (low <= high) | |
161 | { | |
162 | middle = (low + high) / 2; | |
163 | s = from_state[middle]; | |
164 | if (s == state) | |
165 | return middle; | |
166 | else if (s < state) | |
167 | low = middle + 1; | |
168 | else | |
169 | high = middle - 1; | |
170 | } | |
171 | ||
172 | assert (0); | |
173 | /* NOTREACHED */ | |
174 | return 0; | |
175 | } | |
176 | ||
177 | ||
178 | static void | |
179 | initialize_F (void) | |
180 | { | |
181 | short **reads = XCALLOC (short *, ngotos); | |
182 | short *edge = XCALLOC (short, ngotos + 1); | |
183 | int nedges = 0; | |
184 | ||
185 | int i; | |
186 | ||
187 | F = bitsetv_create (ngotos, ntokens, BITSET_FIXED); | |
188 | ||
189 | for (i = 0; i < ngotos; i++) | |
190 | { | |
191 | state_number_t stateno = to_state[i]; | |
192 | transitions_t *sp = states[stateno]->shifts; | |
193 | ||
194 | int j; | |
195 | for (j = 0; j < sp->num && TRANSITION_IS_SHIFT (sp, j); j++) | |
196 | bitset_set (F[i], TRANSITION_SYMBOL (sp, j)); | |
197 | ||
198 | for (; j < sp->num; j++) | |
199 | { | |
200 | symbol_number_t symbol = TRANSITION_SYMBOL (sp, j); | |
201 | if (nullable[symbol]) | |
202 | edge[nedges++] = map_goto (stateno, symbol); | |
203 | } | |
204 | ||
205 | if (nedges) | |
206 | { | |
207 | reads[i] = XCALLOC (short, nedges + 1); | |
208 | memcpy (reads[i], edge, nedges * sizeof (edge[0])); | |
209 | reads[i][nedges] = -1; | |
210 | nedges = 0; | |
211 | } | |
212 | } | |
213 | ||
214 | relation_digraph (reads, ngotos, &F); | |
215 | ||
216 | for (i = 0; i < ngotos; i++) | |
217 | XFREE (reads[i]); | |
218 | ||
219 | XFREE (reads); | |
220 | XFREE (edge); | |
221 | } | |
222 | ||
223 | ||
224 | static void | |
225 | add_lookback_edge (state_t *state, rule_number_t ruleno, int gotono) | |
226 | { | |
227 | int i; | |
228 | shorts *sp; | |
229 | ||
230 | for (i = 0; i < state->nlookaheads; ++i) | |
231 | if (state->lookaheads_rule[i]->number == ruleno) | |
232 | break; | |
233 | ||
234 | assert (state->lookaheads_rule[i]->number == ruleno); | |
235 | ||
236 | sp = XCALLOC (shorts, 1); | |
237 | sp->next = lookback[(state->lookaheads - LA) + i]; | |
238 | sp->value = gotono; | |
239 | lookback[(state->lookaheads - LA) + i] = sp; | |
240 | } | |
241 | ||
242 | ||
243 | ||
244 | static void | |
245 | build_relations (void) | |
246 | { | |
247 | short *edge = XCALLOC (short, ngotos + 1); | |
248 | state_number_t *states1 = XCALLOC (state_number_t, ritem_longest_rhs () + 1); | |
249 | int i; | |
250 | ||
251 | includes = XCALLOC (short *, ngotos); | |
252 | ||
253 | for (i = 0; i < ngotos; i++) | |
254 | { | |
255 | int nedges = 0; | |
256 | symbol_number_t symbol1 = states[to_state[i]]->accessing_symbol; | |
257 | rule_number_t *rulep; | |
258 | ||
259 | for (rulep = derives[symbol1]; *rulep > 0; rulep++) | |
260 | { | |
261 | int done; | |
262 | int length = 1; | |
263 | item_number_t *rp; | |
264 | state_t *state = states[from_state[i]]; | |
265 | states1[0] = state->number; | |
266 | ||
267 | for (rp = rules[*rulep].rhs; *rp >= 0; rp++) | |
268 | { | |
269 | state = transitions_to (state->shifts, | |
270 | item_number_as_symbol_number (*rp)); | |
271 | states1[length++] = state->number; | |
272 | } | |
273 | ||
274 | if (!state->consistent) | |
275 | add_lookback_edge (state, *rulep, i); | |
276 | ||
277 | length--; | |
278 | done = 0; | |
279 | while (!done) | |
280 | { | |
281 | done = 1; | |
282 | rp--; | |
283 | /* JF added rp>=ritem && I hope to god its right! */ | |
284 | if (rp >= ritem && ISVAR (*rp)) | |
285 | { | |
286 | /* Downcasting from item_number_t to symbol_number_t. */ | |
287 | edge[nedges++] = map_goto (states1[--length], | |
288 | item_number_as_symbol_number (*rp)); | |
289 | if (nullable[*rp]) | |
290 | done = 0; | |
291 | } | |
292 | } | |
293 | } | |
294 | ||
295 | if (nedges) | |
296 | { | |
297 | int j; | |
298 | includes[i] = XCALLOC (short, nedges + 1); | |
299 | for (j = 0; j < nedges; j++) | |
300 | includes[i][j] = edge[j]; | |
301 | includes[i][nedges] = -1; | |
302 | } | |
303 | } | |
304 | ||
305 | XFREE (edge); | |
306 | XFREE (states1); | |
307 | ||
308 | relation_transpose (&includes, ngotos); | |
309 | } | |
310 | ||
311 | ||
312 | ||
313 | static void | |
314 | compute_FOLLOWS (void) | |
315 | { | |
316 | int i; | |
317 | ||
318 | relation_digraph (includes, ngotos, &F); | |
319 | ||
320 | for (i = 0; i < ngotos; i++) | |
321 | XFREE (includes[i]); | |
322 | ||
323 | XFREE (includes); | |
324 | } | |
325 | ||
326 | ||
327 | static void | |
328 | compute_lookaheads (void) | |
329 | { | |
330 | size_t i; | |
331 | shorts *sp; | |
332 | ||
333 | for (i = 0; i < nLA; i++) | |
334 | for (sp = lookback[i]; sp; sp = sp->next) | |
335 | bitset_or (LA[i], LA[i], F[sp->value]); | |
336 | ||
337 | /* Free LOOKBACK. */ | |
338 | for (i = 0; i < nLA; i++) | |
339 | LIST_FREE (shorts, lookback[i]); | |
340 | ||
341 | XFREE (lookback); | |
342 | bitsetv_free (F); | |
343 | } | |
344 | ||
345 | ||
346 | /*-------------------------------------------------------------. | |
347 | | Count the number of lookaheads required for each state | | |
348 | | (NLOOKAHEADS member). Compute the total number of LA, NLA. | | |
349 | `-------------------------------------------------------------*/ | |
350 | ||
351 | static void | |
352 | states_lookaheads_count (void) | |
353 | { | |
354 | state_number_t i; | |
355 | nLA = 0; | |
356 | ||
357 | /* Count */ | |
358 | for (i = 0; i < nstates; i++) | |
359 | { | |
360 | int k; | |
361 | int nlookaheads = 0; | |
362 | reductions_t *rp = states[i]->reductions; | |
363 | transitions_t *sp = states[i]->shifts; | |
364 | ||
365 | /* We need a lookahead either to distinguish different | |
366 | reductions (i.e., there are two or more), or to distinguish a | |
367 | reduction from a shift. Otherwise, it is straightforward, | |
368 | and the state is `consistent'. */ | |
369 | if (rp->num > 1 | |
370 | || (rp->num == 1 && sp->num && TRANSITION_IS_SHIFT (sp, 0))) | |
371 | nlookaheads += rp->num; | |
372 | else | |
373 | states[i]->consistent = 1; | |
374 | ||
375 | for (k = 0; k < sp->num; k++) | |
376 | if (TRANSITION_IS_ERROR (sp, k)) | |
377 | { | |
378 | states[i]->consistent = 0; | |
379 | break; | |
380 | } | |
381 | ||
382 | states[i]->nlookaheads = nlookaheads; | |
383 | nLA += nlookaheads; | |
384 | } | |
385 | } | |
386 | ||
387 | ||
388 | /*--------------------------------------. | |
389 | | Initializing the lookaheads members. | | |
390 | `--------------------------------------*/ | |
391 | ||
392 | static void | |
393 | states_lookaheads_initialize (void) | |
394 | { | |
395 | state_number_t i; | |
396 | bitsetv pLA = LA; | |
397 | rule_t **pLArule = LArule; | |
398 | ||
399 | /* Initialize the members LOOKAHEADS and LOOKAHEADS_RULE for each | |
400 | state. */ | |
401 | for (i = 0; i < nstates; i++) | |
402 | { | |
403 | states[i]->lookaheads = pLA; | |
404 | states[i]->lookaheads_rule = pLArule; | |
405 | pLA += states[i]->nlookaheads; | |
406 | pLArule += states[i]->nlookaheads; | |
407 | } | |
408 | } | |
409 | ||
410 | ||
411 | /*---------------------------------------. | |
412 | | Output the lookaheads for each state. | | |
413 | `---------------------------------------*/ | |
414 | ||
415 | static void | |
416 | lookaheads_print (FILE *out) | |
417 | { | |
418 | state_number_t i; | |
419 | int j, k; | |
420 | fprintf (out, "Lookaheads: BEGIN\n"); | |
421 | for (i = 0; i < nstates; ++i) | |
422 | { | |
423 | fprintf (out, "State %d: %d lookaheads\n", | |
424 | i, states[i]->nlookaheads); | |
425 | ||
426 | for (j = 0; j < states[i]->nlookaheads; ++j) | |
427 | BITSET_EXECUTE (states[i]->lookaheads[j], 0, k, | |
428 | { | |
429 | fprintf (out, " on %d (%s) -> rule %d\n", | |
430 | k, symbols[k]->tag, | |
431 | states[i]->lookaheads_rule[j]->number - 1); | |
432 | }); | |
433 | } | |
434 | fprintf (out, "Lookaheads: END\n"); | |
435 | } | |
436 | ||
437 | void | |
438 | lalr (void) | |
439 | { | |
440 | states_lookaheads_count (); | |
441 | initialize_LA (); | |
442 | states_lookaheads_initialize (); | |
443 | set_goto_map (); | |
444 | initialize_F (); | |
445 | build_relations (); | |
446 | compute_FOLLOWS (); | |
447 | compute_lookaheads (); | |
448 | ||
449 | if (trace_flag) | |
450 | lookaheads_print (stderr); | |
451 | } |