]>
Commit | Line | Data |
---|---|---|
166366b2 JD |
1 | /* IELR main implementation. |
2 | ||
3 | Copyright (C) 2009 Free Software Foundation, Inc. | |
4 | ||
5 | This file is part of Bison, the GNU Compiler Compiler. | |
6 | ||
7 | This program 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 3 of the License, or | |
10 | (at your option) any later version. | |
11 | ||
12 | This program 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 this program. If not, see <http://www.gnu.org/licenses/>. */ | |
19 | ||
20 | #include <config.h> | |
21 | #include "system.h" | |
22 | ||
23 | #include "ielr.h" | |
24 | ||
25 | #include <bitset.h> | |
26 | #include <timevar.h> | |
27 | ||
28 | #include "AnnotationList.h" | |
29 | #include "derives.h" | |
30 | #include "getargs.h" | |
31 | #include "lalr.h" | |
23ec25b7 | 32 | #include "muscle-tab.h" |
166366b2 JD |
33 | #include "nullable.h" |
34 | #include "relation.h" | |
35 | #include "state.h" | |
36 | #include "symtab.h" | |
37 | ||
f37495f6 | 38 | /** Records the value of the \%define variable lr.type. */ |
166366b2 JD |
39 | typedef enum { LR_TYPE__LALR, LR_TYPE__IELR, LR_TYPE__CANONICAL_LR } LrType; |
40 | ||
41 | /** | |
42 | * \post: | |
43 | * - \c result = a new \c bitset of size \c ::nritems such that any bit \c i | |
44 | * is set iff <tt>ritem[i]</tt> is a nonterminal after which all ritems in | |
45 | * the same RHS are nullable nonterminals. In other words, the follows of | |
46 | * a goto on <tt>ritem[i]</tt> include the lookahead set of the item. | |
47 | */ | |
48 | static bitset | |
49 | ielr_compute_ritem_sees_lookahead_set (void) | |
50 | { | |
51 | bitset result = bitset_create (nritems, BITSET_FIXED); | |
52 | unsigned int i = nritems-1; | |
53 | while (i>0) | |
54 | { | |
55 | --i; | |
56 | while (!item_number_is_rule_number (ritem[i]) | |
57 | && ISVAR (ritem[i]) | |
58 | && nullable [item_number_as_symbol_number (ritem[i]) - ntokens]) | |
59 | bitset_set (result, i--); | |
60 | if (!item_number_is_rule_number (ritem[i]) && ISVAR (ritem[i])) | |
61 | bitset_set (result, i--); | |
62 | while (!item_number_is_rule_number (ritem[i]) && i>0) | |
63 | --i; | |
64 | } | |
65 | if (trace_flag & trace_ielr) | |
66 | { | |
67 | fprintf (stderr, "ritem_sees_lookahead_set:\n"); | |
68 | debug_bitset (result); | |
69 | fprintf (stderr, "\n"); | |
70 | } | |
71 | return result; | |
72 | } | |
73 | ||
74 | /** | |
75 | * \pre: | |
76 | * - \c ritem_sees_lookahead_set was computed by | |
77 | * \c ielr_compute_ritem_sees_lookahead_set. | |
78 | * \post: | |
79 | * - Each of \c *edgesp and \c *edge_countsp is a new array of size | |
80 | * \c ::ngotos. | |
81 | * - <tt>(*edgesp)[i]</tt> points to a \c goto_number array of size | |
82 | * <tt>(*edge_countsp)[i]+1</tt>. | |
83 | * - In such a \c goto_number array, the last element is \c ::END_NODE. | |
84 | * - All remaining elements are the indices of the gotos to which there is an | |
85 | * internal follow edge from goto \c i. | |
86 | * - There is an internal follow edge from goto \c i to goto \c j iff both: | |
87 | * - The from states of gotos \c i and \c j are the same. | |
88 | * - The transition nonterminal for goto \c i appears as the first RHS | |
89 | * symbol of at least one production for which both: | |
90 | * - The LHS is the transition symbol of goto \c j. | |
91 | * - All other RHS symbols are nullable nonterminals. | |
92 | * - In other words, the follows of goto \c i include the follows of | |
93 | * goto \c j and it's an internal edge because the from states are the | |
94 | * same. | |
95 | */ | |
96 | static void | |
97 | ielr_compute_internal_follow_edges (bitset ritem_sees_lookahead_set, | |
98 | goto_number ***edgesp, int **edge_countsp) | |
99 | { | |
100 | *edgesp = xnmalloc (ngotos, sizeof **edgesp); | |
101 | *edge_countsp = xnmalloc (ngotos, sizeof **edge_countsp); | |
102 | { | |
103 | bitset sources = bitset_create (ngotos, BITSET_FIXED); | |
104 | goto_number i; | |
105 | for (i = 0; i < ngotos; ++i) | |
106 | (*edge_countsp)[i] = 0; | |
107 | for (i = 0; i < ngotos; ++i) | |
108 | { | |
109 | int nsources = 0; | |
110 | { | |
111 | rule **rulep; | |
112 | for (rulep = derives[states[to_state[i]]->accessing_symbol | |
113 | - ntokens]; | |
114 | *rulep; | |
115 | ++rulep) | |
116 | { | |
117 | /* If there is at least one RHS symbol, if the first RHS symbol | |
118 | is a nonterminal, and if all remaining RHS symbols (if any) | |
119 | are nullable nonterminals, create an edge from the LHS | |
120 | symbol's goto to the first RHS symbol's goto such that the RHS | |
121 | symbol's goto will be the source of the edge after the | |
122 | eventual relation_transpose below. | |
123 | ||
124 | Unlike in ielr_compute_always_follows, I use a bitset for | |
125 | edges rather than an array because it is possible that | |
126 | multiple RHS's with the same first symbol could fit and thus | |
127 | that we could end up with redundant edges. With the | |
128 | possibility of redundant edges, it's hard to know ahead of | |
129 | time how large to make such an array. Another possible | |
130 | redundancy is that source and destination might be the same | |
131 | goto. Eliminating all these possible redundancies now might | |
132 | possibly help performance a little. I have not proven any of | |
133 | this, but I'm guessing the bitset shouldn't entail much of a | |
134 | performance penalty, if any. */ | |
135 | if (bitset_test (ritem_sees_lookahead_set, | |
136 | (*rulep)->rhs - ritem)) | |
137 | { | |
138 | goto_number source = | |
139 | map_goto (from_state[i], | |
140 | item_number_as_symbol_number (*(*rulep)->rhs)); | |
141 | if (i != source && !bitset_test (sources, source)) | |
142 | { | |
143 | bitset_set (sources, source); | |
144 | ++nsources; | |
145 | ++(*edge_countsp)[source]; | |
146 | } | |
147 | } | |
148 | } | |
149 | } | |
150 | if (nsources == 0) | |
151 | (*edgesp)[i] = NULL; | |
152 | else | |
153 | { | |
154 | (*edgesp)[i] = xnmalloc (nsources + 1, sizeof *(*edgesp)[i]); | |
155 | { | |
156 | bitset_iterator biter_source; | |
157 | bitset_bindex source; | |
158 | int j = 0; | |
159 | BITSET_FOR_EACH (biter_source, sources, source, 0) | |
160 | (*edgesp)[i][j++] = source; | |
161 | } | |
162 | (*edgesp)[i][nsources] = END_NODE; | |
163 | } | |
164 | bitset_zero (sources); | |
165 | } | |
166 | bitset_free (sources); | |
167 | } | |
168 | ||
169 | relation_transpose (edgesp, ngotos); | |
170 | ||
171 | if (trace_flag & trace_ielr) | |
172 | { | |
173 | fprintf (stderr, "internal_follow_edges:\n"); | |
174 | relation_print (*edgesp, ngotos, stderr); | |
175 | } | |
176 | } | |
177 | ||
178 | /** | |
179 | * \pre: | |
180 | * - \c ritem_sees_lookahead_set was computed by | |
181 | * \c ielr_compute_ritem_sees_lookahead_set. | |
182 | * - \c internal_follow_edges was computed by | |
183 | * \c ielr_compute_internal_follow_edges. | |
184 | * \post: | |
185 | * - \c *follow_kernel_itemsp is a new \c bitsetv in which the number of rows | |
186 | * is \c ngotos and the number of columns is maximum number of kernel items | |
187 | * in any state. | |
188 | * - <tt>(*follow_kernel_itemsp)[i][j]</tt> is set iff the follows of goto | |
189 | * \c i include the lookahead set of item \c j in the from state of goto | |
190 | * \c i. | |
191 | * - Thus, <tt>(*follow_kernel_itemsp)[i][j]</tt> is always unset if there is | |
192 | * no item \c j in the from state of goto \c i. | |
193 | */ | |
194 | static void | |
195 | ielr_compute_follow_kernel_items (bitset ritem_sees_lookahead_set, | |
196 | goto_number **internal_follow_edges, | |
197 | bitsetv *follow_kernel_itemsp) | |
198 | { | |
199 | { | |
200 | size_t max_nitems = 0; | |
201 | state_number i; | |
202 | for (i = 0; i < nstates; ++i) | |
203 | if (states[i]->nitems > max_nitems) | |
204 | max_nitems = states[i]->nitems; | |
205 | *follow_kernel_itemsp = bitsetv_create (ngotos, max_nitems, BITSET_FIXED); | |
206 | } | |
207 | { | |
208 | goto_number i; | |
209 | for (i = 0; i < ngotos; ++i) | |
210 | { | |
211 | size_t nitems = states[from_state[i]]->nitems; | |
212 | item_number *items = states[from_state[i]]->items; | |
213 | size_t j; | |
214 | for (j = 0; j < nitems; ++j) | |
215 | /* If this item has this goto and if all subsequent symbols in this | |
216 | RHS (if any) are nullable nonterminals, then record this item as | |
217 | one whose lookahead set is included in this goto's follows. */ | |
218 | if (item_number_is_symbol_number (ritem[items[j]]) | |
219 | && item_number_as_symbol_number (ritem[items[j]]) | |
220 | == states[to_state[i]]->accessing_symbol | |
221 | && bitset_test (ritem_sees_lookahead_set, items[j])) | |
222 | bitset_set ((*follow_kernel_itemsp)[i], j); | |
223 | } | |
224 | } | |
225 | relation_digraph (internal_follow_edges, ngotos, follow_kernel_itemsp); | |
226 | ||
227 | if (trace_flag & trace_ielr) | |
228 | { | |
229 | fprintf (stderr, "follow_kernel_items:\n"); | |
230 | debug_bitsetv (*follow_kernel_itemsp); | |
231 | } | |
232 | } | |
233 | ||
234 | /** | |
235 | * \pre | |
236 | * - \c *edgesp and \c edge_counts were computed by | |
237 | * \c ielr_compute_internal_follow_edges. | |
238 | * \post | |
239 | * - \c *always_followsp is a new \c bitsetv with \c ngotos rows and | |
240 | * \c ntokens columns. | |
241 | * - <tt>(*always_followsp)[i][j]</tt> is set iff token \c j is an always | |
242 | * follow (that is, it's computed by internal and successor edges) of goto | |
243 | * \c i. | |
244 | * - Rows of \c *edgesp have been realloc'ed and extended to include | |
245 | * successor follow edges. \c edge_counts has not been updated. | |
246 | */ | |
247 | static void | |
248 | ielr_compute_always_follows (goto_number ***edgesp, | |
249 | int const edge_counts[], | |
250 | bitsetv *always_followsp) | |
251 | { | |
252 | *always_followsp = bitsetv_create (ngotos, ntokens, BITSET_FIXED); | |
253 | { | |
254 | goto_number *edge_array = xnmalloc (ngotos, sizeof *edge_array); | |
255 | goto_number i; | |
256 | for (i = 0; i < ngotos; ++i) | |
257 | { | |
258 | goto_number nedges = edge_counts[i]; | |
259 | { | |
260 | int j; | |
261 | transitions *trans = states[to_state[i]]->transitions; | |
262 | FOR_EACH_SHIFT (trans, j) | |
263 | bitset_set ((*always_followsp)[i], TRANSITION_SYMBOL (trans, j)); | |
264 | for (; j < trans->num; ++j) | |
265 | { | |
266 | symbol_number sym = TRANSITION_SYMBOL (trans, j); | |
267 | if (nullable[sym - ntokens]) | |
268 | edge_array[nedges++] = map_goto (to_state[i], sym); | |
269 | } | |
270 | } | |
271 | if (nedges - edge_counts[i]) | |
272 | { | |
273 | (*edgesp)[i] = | |
274 | xnrealloc ((*edgesp)[i], nedges + 1, sizeof *(*edgesp)[i]); | |
275 | memcpy ((*edgesp)[i] + edge_counts[i], edge_array + edge_counts[i], | |
276 | (nedges - edge_counts[i]) * sizeof *(*edgesp)[i]); | |
277 | (*edgesp)[i][nedges] = END_NODE; | |
278 | } | |
279 | } | |
280 | free (edge_array); | |
281 | } | |
282 | relation_digraph (*edgesp, ngotos, always_followsp); | |
283 | ||
284 | if (trace_flag & trace_ielr) | |
285 | { | |
286 | fprintf (stderr, "always follow edges:\n"); | |
287 | relation_print (*edgesp, ngotos, stderr); | |
288 | fprintf (stderr, "always_follows:\n"); | |
289 | debug_bitsetv (*always_followsp); | |
290 | } | |
291 | } | |
292 | ||
293 | /** | |
294 | * \post | |
295 | * - \c result is a new array of size \c ::nstates. | |
296 | * - <tt>result[i]</tt> is an array of pointers to the predecessor | |
297 | * <tt>state</tt>'s of state \c i. | |
298 | * - The last element of such an array is \c NULL. | |
299 | */ | |
300 | static state *** | |
301 | ielr_compute_predecessors (void) | |
302 | { | |
303 | state_number i; | |
304 | int *predecessor_counts = xnmalloc (nstates, sizeof *predecessor_counts); | |
305 | state ***result = xnmalloc (nstates, sizeof *result); | |
306 | for (i = 0; i < nstates; ++i) | |
307 | predecessor_counts[i] = 0; | |
308 | for (i = 0; i < nstates; ++i) | |
309 | { | |
310 | int j; | |
311 | for (j = 0; j < states[i]->transitions->num; ++j) | |
312 | ++predecessor_counts[states[i]->transitions->states[j]->number]; | |
313 | } | |
314 | for (i = 0; i < nstates; ++i) | |
315 | { | |
316 | result[i] = xnmalloc (predecessor_counts[i]+1, sizeof *result[i]); | |
317 | result[i][predecessor_counts[i]] = NULL; | |
318 | predecessor_counts[i] = 0; | |
319 | } | |
320 | for (i = 0; i < nstates; ++i) | |
321 | { | |
322 | int j; | |
323 | for (j = 0; j < states[i]->transitions->num; ++j) | |
324 | { | |
325 | state_number k = states[i]->transitions->states[j]->number; | |
326 | result[k][predecessor_counts[k]++] = states[i]; | |
327 | } | |
328 | } | |
329 | free (predecessor_counts); | |
330 | return result; | |
331 | } | |
332 | ||
333 | /** | |
334 | * \post | |
335 | * - \c *follow_kernel_itemsp and \c *always_followsp were computed by | |
336 | * \c ielr_compute_follow_kernel_items and | |
337 | * \c ielr_compute_always_follows. | |
338 | * - Iff <tt>predecessorsp != NULL</tt>, then \c *predecessorsp was computed | |
339 | * by \c ielr_compute_predecessors. | |
340 | */ | |
341 | static void | |
342 | ielr_compute_auxiliary_tables (bitsetv *follow_kernel_itemsp, | |
343 | bitsetv *always_followsp, | |
344 | state ****predecessorsp) | |
345 | { | |
346 | goto_number **edges; | |
347 | int *edge_counts; | |
348 | { | |
349 | bitset ritem_sees_lookahead_set = ielr_compute_ritem_sees_lookahead_set (); | |
350 | ielr_compute_internal_follow_edges (ritem_sees_lookahead_set, | |
351 | &edges, &edge_counts); | |
352 | ielr_compute_follow_kernel_items (ritem_sees_lookahead_set, edges, | |
353 | follow_kernel_itemsp); | |
354 | bitset_free (ritem_sees_lookahead_set); | |
355 | } | |
356 | ielr_compute_always_follows (&edges, edge_counts, always_followsp); | |
357 | { | |
358 | int i; | |
359 | for (i = 0; i < ngotos; ++i) | |
360 | free (edges[i]); | |
361 | } | |
362 | free (edges); | |
363 | free (edge_counts); | |
364 | if (predecessorsp) | |
365 | *predecessorsp = ielr_compute_predecessors (); | |
366 | } | |
367 | ||
368 | /** | |
369 | * \note | |
370 | * - FIXME: It might be an interesting experiment to compare the space and | |
371 | * time efficiency of computing \c item_lookahead_sets either: | |
372 | * - Fully up front. | |
373 | * - Just-in-time, as implemented below. | |
374 | * - Not at all. That is, just let annotations continue even when | |
375 | * unnecessary. | |
376 | */ | |
377 | bool | |
378 | ielr_item_has_lookahead (state *s, symbol_number lhs, size_t item, | |
379 | symbol_number lookahead, state ***predecessors, | |
380 | bitset **item_lookahead_sets) | |
381 | { | |
382 | if (!item_lookahead_sets[s->number]) | |
383 | { | |
384 | size_t i; | |
385 | item_lookahead_sets[s->number] = | |
386 | xnmalloc (s->nitems, sizeof item_lookahead_sets[s->number][0]); | |
387 | for (i = 0; i < s->nitems; ++i) | |
388 | item_lookahead_sets[s->number][i] = NULL; | |
389 | } | |
390 | if (!item_lookahead_sets[s->number][item]) | |
391 | { | |
392 | item_lookahead_sets[s->number][item] = | |
393 | bitset_create (ntokens, BITSET_FIXED); | |
394 | /* If this kernel item is the beginning of a RHS, it must be the kernel | |
395 | item in the start state, and so its LHS has no follows and no goto to | |
396 | check. If, instead, this kernel item is the successor of the start | |
397 | state's kernel item, there are still no follows and no goto. This | |
398 | situation is fortunate because we want to avoid the - 2 below in both | |
399 | cases. | |
400 | ||
401 | Actually, IELR(1) should never invoke this function for either of | |
402 | those cases because (1) follow_kernel_items will never reference a | |
403 | kernel item for this RHS because the end token blocks sight of the | |
404 | lookahead set from the RHS's only nonterminal, and (2) no reduction | |
405 | has a lookback dependency on this lookahead set. Nevertheless, I | |
406 | didn't change this test to an aver just in case the usage of this | |
407 | function evolves to need those two cases. In both cases, the current | |
408 | implementation returns the right result. */ | |
409 | if (s->items[item] > 1) | |
410 | { | |
411 | /* If the LHS symbol of this item isn't known (because this is a | |
412 | top-level invocation), go get it. */ | |
413 | if (!lhs) | |
414 | { | |
415 | unsigned int i; | |
416 | for (i = s->items[item]; | |
417 | !item_number_is_rule_number (ritem[i]); | |
418 | ++i) | |
419 | ; | |
420 | lhs = rules[item_number_as_rule_number (ritem[i])].lhs->number; | |
421 | } | |
422 | /* If this kernel item is next to the beginning of the RHS, then | |
423 | check all predecessors' goto follows for the LHS. */ | |
424 | if (item_number_is_rule_number (ritem[s->items[item] - 2])) | |
425 | { | |
426 | state **predecessor; | |
427 | aver (lhs != accept->number); | |
428 | for (predecessor = predecessors[s->number]; | |
429 | *predecessor; | |
430 | ++predecessor) | |
431 | bitset_or (item_lookahead_sets[s->number][item], | |
432 | item_lookahead_sets[s->number][item], | |
433 | goto_follows[map_goto ((*predecessor)->number, | |
434 | lhs)]); | |
435 | } | |
436 | /* If this kernel item is later in the RHS, then check all | |
437 | predecessor items' lookahead sets. */ | |
438 | else | |
439 | { | |
440 | state **predecessor; | |
441 | for (predecessor = predecessors[s->number]; | |
442 | *predecessor; | |
443 | ++predecessor) | |
444 | { | |
445 | size_t predecessor_item; | |
446 | for (predecessor_item = 0; | |
447 | predecessor_item < (*predecessor)->nitems; | |
448 | ++predecessor_item) | |
449 | if ((*predecessor)->items[predecessor_item] | |
450 | == s->items[item] - 1) | |
451 | break; | |
452 | aver (predecessor_item != (*predecessor)->nitems); | |
453 | ielr_item_has_lookahead (*predecessor, lhs, | |
454 | predecessor_item, 0 /*irrelevant*/, | |
455 | predecessors, item_lookahead_sets); | |
456 | bitset_or (item_lookahead_sets[s->number][item], | |
457 | item_lookahead_sets[s->number][item], | |
458 | item_lookahead_sets[(*predecessor)->number] | |
459 | [predecessor_item]); | |
460 | } | |
461 | } | |
462 | } | |
463 | } | |
464 | return bitset_test (item_lookahead_sets[s->number][item], lookahead); | |
465 | } | |
466 | ||
467 | /** | |
468 | * \pre | |
469 | * - \c follow_kernel_items, \c always_follows, and \c predecessors | |
470 | * were computed by \c ielr_compute_auxiliary_tables. | |
471 | * \post | |
472 | * - Each of <tt>*inadequacy_listsp</tt> and <tt>*annotation_listsp</tt> | |
473 | * points to a new array of size \c ::nstates. | |
474 | * - For <tt>0 <= i < ::nstates</tt>: | |
475 | * - <tt>(*inadequacy_listsp)[i]</tt> contains the \c InadequacyList head | |
476 | * node for <tt>states[i]</tt>. | |
477 | * - <tt>(*annotation_listsp)[i]</tt> contains the \c AnnotationList head | |
478 | * node for <tt>states[i]</tt>. | |
479 | * - <tt>*max_annotationsp</tt> is the maximum number of annotations per | |
480 | * state. | |
481 | */ | |
482 | static void | |
483 | ielr_compute_annotation_lists (bitsetv follow_kernel_items, | |
484 | bitsetv always_follows, state ***predecessors, | |
485 | AnnotationIndex *max_annotationsp, | |
486 | InadequacyList ***inadequacy_listsp, | |
487 | AnnotationList ***annotation_listsp, | |
488 | struct obstack *annotations_obstackp) | |
489 | { | |
490 | bitset **item_lookahead_sets = | |
491 | xnmalloc (nstates, sizeof *item_lookahead_sets); | |
492 | AnnotationIndex *annotation_counts = | |
493 | xnmalloc (nstates, sizeof *annotation_counts); | |
494 | ContributionIndex max_contributions = 0; | |
495 | unsigned int total_annotations = 0; | |
496 | state_number i; | |
497 | ||
498 | *inadequacy_listsp = xnmalloc (nstates, sizeof **inadequacy_listsp); | |
499 | *annotation_listsp = xnmalloc (nstates, sizeof **annotation_listsp); | |
500 | for (i = 0; i < nstates; ++i) | |
501 | { | |
502 | item_lookahead_sets[i] = NULL; | |
503 | (*inadequacy_listsp)[i] = NULL; | |
504 | (*annotation_listsp)[i] = NULL; | |
505 | annotation_counts[i] = 0; | |
506 | } | |
2728ac7e JD |
507 | { |
508 | InadequacyListNodeCount inadequacy_list_node_count = 0; | |
509 | for (i = 0; i < nstates; ++i) | |
510 | AnnotationList__compute_from_inadequacies ( | |
511 | states[i], follow_kernel_items, always_follows, predecessors, | |
512 | item_lookahead_sets, *inadequacy_listsp, *annotation_listsp, | |
513 | annotation_counts, &max_contributions, annotations_obstackp, | |
514 | &inadequacy_list_node_count); | |
515 | } | |
166366b2 JD |
516 | *max_annotationsp = 0; |
517 | for (i = 0; i < nstates; ++i) | |
518 | { | |
519 | if (annotation_counts[i] > *max_annotationsp) | |
520 | *max_annotationsp = annotation_counts[i]; | |
521 | total_annotations += annotation_counts[i]; | |
522 | } | |
523 | if (trace_flag & trace_ielr) | |
524 | { | |
525 | for (i = 0; i < nstates; ++i) | |
526 | { | |
527 | fprintf (stderr, "Inadequacy annotations for state %d:\n", i); | |
528 | AnnotationList__debug ((*annotation_listsp)[i], | |
529 | states[i]->nitems, 2); | |
530 | } | |
531 | fprintf (stderr, "Number of LR(0)/LALR(1) states: %d\n", nstates); | |
532 | fprintf (stderr, "Average number of annotations per state: %f\n", | |
533 | (float)total_annotations/nstates); | |
534 | fprintf (stderr, "Max number of annotations per state: %d\n", | |
535 | *max_annotationsp); | |
536 | fprintf (stderr, "Max number of contributions per annotation: %d\n", | |
537 | max_contributions); | |
538 | } | |
539 | for (i = 0; i < nstates; ++i) | |
540 | if (item_lookahead_sets[i]) | |
541 | { | |
542 | size_t j; | |
543 | for (j = 0; j < states[i]->nitems; ++j) | |
544 | if (item_lookahead_sets[i][j]) | |
545 | bitset_free (item_lookahead_sets[i][j]); | |
546 | free (item_lookahead_sets[i]); | |
547 | } | |
548 | free (item_lookahead_sets); | |
549 | free (annotation_counts); | |
550 | } | |
551 | ||
552 | typedef struct state_list { | |
553 | struct state_list *next; | |
554 | state *state; | |
555 | /** Has this state been recomputed as a successor of another state? */ | |
556 | bool recomputedAsSuccessor; | |
557 | /** | |
558 | * \c NULL iff all lookahead sets are empty. <tt>lookaheads[i] = NULL</tt> | |
559 | * iff the lookahead set on item \c i is empty. | |
560 | */ | |
561 | bitset *lookaheads; | |
562 | /** | |
563 | * nextIsocore is the next state in a circularly linked-list of all states | |
564 | * with the same core. The one originally computed by generate_states in | |
565 | * LR0.c is lr0Isocore. | |
566 | */ | |
567 | struct state_list *lr0Isocore; | |
568 | struct state_list *nextIsocore; | |
569 | } state_list; | |
570 | ||
571 | /** | |
572 | * \pre | |
573 | * - \c follow_kernel_items and \c always_follows were computed by | |
574 | * \c ielr_compute_auxiliary_tables. | |
575 | * - <tt>n->class = nterm_sym</tt>. | |
576 | * \post | |
577 | * - \c follow_set contains the follow set for the goto on nonterminal \c n | |
578 | * in state \c s based on the lookaheads stored in <tt>s->lookaheads</tt>. | |
579 | */ | |
580 | static void | |
581 | ielr_compute_goto_follow_set (bitsetv follow_kernel_items, | |
582 | bitsetv always_follows, state_list *s, | |
583 | symbol *n, bitset follow_set) | |
584 | { | |
585 | goto_number n_goto = map_goto (s->lr0Isocore->state->number, n->number); | |
586 | bitset_copy (follow_set, always_follows[n_goto]); | |
587 | if (s->lookaheads) | |
588 | { | |
589 | bitset_iterator biter_item; | |
590 | bitset_bindex item; | |
591 | BITSET_FOR_EACH (biter_item, follow_kernel_items[n_goto], item, 0) | |
592 | if (s->lookaheads[item]) | |
593 | bitset_or (follow_set, follow_set, s->lookaheads[item]); | |
594 | } | |
595 | } | |
596 | ||
597 | /** | |
598 | * \pre | |
599 | * - \c follow_kernel_items and \c always_follows were computed by | |
600 | * \c ielr_compute_auxiliary_tables. | |
601 | * - \c lookahead_filter was computed by | |
602 | * \c AnnotationList__computeLookaheadFilter for the original LR(0) isocore | |
603 | * of \c t. | |
604 | * - The number of rows in \c lookaheads is at least the number of items in | |
605 | * \c t, and the number of columns is \c ::ntokens. | |
606 | * \post | |
607 | * - <tt>lookaheads[i][j]</tt> is set iff both: | |
608 | * - <tt>lookahead_filter[i][j]</tt> is set. | |
609 | * - The isocore of \c t that will be the transition successor of \c s will | |
610 | * inherit from \c s token \c j into the lookahead set of item \c i. | |
611 | */ | |
612 | static void | |
613 | ielr_compute_lookaheads (bitsetv follow_kernel_items, bitsetv always_follows, | |
614 | state_list *s, state *t, bitsetv lookahead_filter, | |
615 | bitsetv lookaheads) | |
616 | { | |
617 | size_t s_item = 0; | |
618 | size_t t_item; | |
619 | bitsetv_zero (lookaheads); | |
620 | for (t_item = 0; t_item < t->nitems; ++t_item) | |
621 | { | |
622 | /* If this kernel item is the beginning of a RHS, it must be the | |
623 | kernel item in the start state, but t is supposed to be a successor | |
624 | state. If, instead, this kernel item is the successor of the start | |
625 | state's kernel item, the lookahead set is empty. This second case is | |
626 | a special case to avoid the - 2 below, but the next successor can be | |
627 | handled fine without special casing it. */ | |
628 | aver (t->items[t_item] != 0); | |
629 | if (t->items[t_item] > 1 | |
630 | && !bitset_empty_p (lookahead_filter[t_item])) | |
631 | { | |
632 | if (item_number_is_rule_number (ritem[t->items[t_item] - 2])) | |
633 | { | |
634 | unsigned int rule_item; | |
635 | for (rule_item = t->items[t_item]; | |
636 | !item_number_is_rule_number (ritem[rule_item]); | |
637 | ++rule_item) | |
638 | ; | |
639 | ielr_compute_goto_follow_set ( | |
640 | follow_kernel_items, always_follows, s, | |
641 | rules[item_number_as_rule_number (ritem[rule_item])].lhs, | |
642 | lookaheads[t_item]); | |
643 | } | |
644 | else if (s->lookaheads) | |
645 | { | |
646 | /* We don't have to start the s item search at the beginning | |
647 | every time because items from both states are sorted by their | |
648 | indices in ritem. */ | |
649 | for (; s_item < s->state->nitems; ++s_item) | |
650 | if (s->state->items[s_item] == t->items[t_item] - 1) | |
651 | break; | |
652 | aver (s_item != s->state->nitems); | |
653 | if (s->lookaheads[s_item]) | |
654 | bitset_copy (lookaheads[t_item], s->lookaheads[s_item]); | |
655 | } | |
656 | bitset_and (lookaheads[t_item], | |
657 | lookaheads[t_item], lookahead_filter[t_item]); | |
658 | } | |
659 | } | |
660 | } | |
661 | ||
662 | /** | |
663 | * \pre | |
664 | * - \c follow_kernel_items and \c always_follows were computed by | |
665 | * \c ielr_compute_auxiliary_tables. | |
666 | * - Either: | |
667 | * - <tt>annotation_lists = NULL</tt> and all bits in work2 are set. | |
668 | * - \c annotation_lists was computed by \c ielr_compute_annotation_lists. | |
669 | * - The number of rows in each of \c lookaheads and \c work2 is the maximum | |
670 | * number of items in any state. The number of columns in each is | |
671 | * \c ::ntokens. | |
672 | * - \c lookaheads was computed by \c ielr_compute_lookaheads for \c t. | |
673 | * - \c ::nstates is the total number of states, some not yet fully computed, | |
674 | * in the list ending at \c *last_statep. It is at least the number of | |
675 | * original LR(0) states. | |
676 | * - The size of \c work1 is at least the number of annotations for the LR(0) | |
677 | * isocore of \c t. | |
678 | * \post | |
679 | * - Either: | |
680 | * - In the case that <tt>annotation_lists != NULL</tt>, | |
681 | * <tt>lookaheads \@pre</tt> was merged with some isocore of \c t if | |
682 | * permitted by the annotations for the original LR(0) isocore of \c t. | |
683 | * If this changed the lookaheads in that isocore, those changes were | |
684 | * propagated to all already computed transition successors recursively | |
685 | * possibly resulting in the splitting of some of those successors. | |
686 | * - In the case that <tt>annotation_lists = NULL</tt>, | |
687 | * <tt>lookaheads \@pre</tt> was merged with some isocore of \c t if the | |
688 | * isocore's lookahead sets were identical to those specified by | |
689 | * <tt>lookaheads \@pre</tt>. | |
690 | * - If no such merge was permitted, a new isocore of \c t containing | |
691 | * <tt>lookaheads \@pre</tt> was appended to the state list whose | |
692 | * previous tail was <tt>*last_statep \@pre</tt> and \c ::nstates was | |
693 | * incremented. It was also appended to \c t's isocore list. | |
694 | * - <tt>*tp</tt> = the isocore of \c t into which | |
695 | * <tt>lookaheads \@pre</tt> was placed/merged. | |
696 | * - \c lookaheads, \c work1, and \c work2 may have been altered. | |
697 | */ | |
698 | static void | |
699 | ielr_compute_state (bitsetv follow_kernel_items, bitsetv always_follows, | |
700 | AnnotationList **annotation_lists, state *t, | |
701 | bitsetv lookaheads, state_list **last_statep, | |
702 | ContributionIndex work1[], bitsetv work2, state **tp) | |
703 | { | |
704 | state_list *lr0_isocore = t->state_list->lr0Isocore; | |
705 | state_list **this_isocorep; | |
706 | bool has_lookaheads; | |
707 | ||
708 | /* Determine whether there's an isocore of t with which these lookaheads can | |
709 | be merged. */ | |
710 | { | |
711 | AnnotationIndex ai; | |
712 | AnnotationList *a; | |
713 | if (annotation_lists) | |
714 | for (ai = 0, a = annotation_lists[lr0_isocore->state->number]; | |
715 | a; | |
716 | ++ai, a = a->next) | |
717 | work1[ai] = | |
718 | AnnotationList__computeDominantContribution ( | |
719 | a, lr0_isocore->state->nitems, lookaheads, false); | |
720 | for (this_isocorep = &t->state_list; | |
721 | this_isocorep == &t->state_list || *this_isocorep != t->state_list; | |
722 | this_isocorep = &(*this_isocorep)->nextIsocore) | |
723 | { | |
724 | if (!(*this_isocorep)->recomputedAsSuccessor) | |
725 | break; | |
726 | if (annotation_lists) | |
727 | { | |
728 | for (ai = 0, a = annotation_lists[lr0_isocore->state->number]; | |
729 | a; | |
730 | ++ai, a = a->next) | |
731 | { | |
732 | if (work1[ai] != ContributionIndex__none) | |
733 | { | |
734 | /* This isocore compatibility test depends on the fact | |
735 | that, if the dominant contributions are the same for the | |
736 | two isocores, then merging their lookahead sets will not | |
737 | produce a state with a different dominant contribution. | |
738 | */ | |
739 | ContributionIndex ci = | |
740 | AnnotationList__computeDominantContribution ( | |
741 | a, lr0_isocore->state->nitems, | |
742 | (*this_isocorep)->lookaheads, false); | |
743 | if (ci != ContributionIndex__none && work1[ai] != ci) | |
744 | break; | |
745 | } | |
746 | } | |
747 | if (!a) | |
748 | break; | |
749 | } | |
750 | else | |
751 | { | |
752 | size_t i; | |
753 | for (i = 0; i < t->nitems; ++i) | |
754 | { | |
755 | if (!(*this_isocorep)->lookaheads | |
756 | || !(*this_isocorep)->lookaheads[i]) | |
757 | { | |
758 | if (!bitset_empty_p (lookaheads[i])) | |
759 | break; | |
760 | } | |
761 | // bitset_equal_p uses the size of the first argument, so | |
762 | // lookaheads[i] must be the second argument. | |
763 | else if (!bitset_equal_p ((*this_isocorep)->lookaheads[i], | |
764 | lookaheads[i])) | |
765 | break; | |
766 | } | |
767 | if (i == t->nitems) | |
768 | break; | |
769 | } | |
770 | } | |
771 | } | |
772 | ||
773 | has_lookaheads = false; | |
774 | { | |
775 | size_t i; | |
776 | for (i = 0; i < lr0_isocore->state->nitems; ++i) | |
777 | if (!bitset_empty_p (lookaheads[i])) | |
778 | { | |
779 | has_lookaheads = true; | |
780 | break; | |
781 | } | |
782 | } | |
783 | ||
784 | /* Merge with an existing isocore. */ | |
785 | if (this_isocorep == &t->state_list || *this_isocorep != t->state_list) | |
786 | { | |
787 | bool new_lookaheads = false; | |
788 | *tp = (*this_isocorep)->state; | |
789 | ||
790 | /* Merge lookaheads into the state and record whether any of them are | |
791 | actually new. */ | |
792 | if (has_lookaheads) | |
793 | { | |
794 | size_t i; | |
795 | if (!(*this_isocorep)->lookaheads) | |
796 | { | |
797 | (*this_isocorep)->lookaheads = | |
798 | xnmalloc (t->nitems, sizeof (*this_isocorep)->lookaheads); | |
799 | for (i = 0; i < t->nitems; ++i) | |
800 | (*this_isocorep)->lookaheads[i] = NULL; | |
801 | } | |
802 | for (i = 0; i < t->nitems; ++i) | |
803 | if (!bitset_empty_p (lookaheads[i])) | |
804 | { | |
805 | if (!(*this_isocorep)->lookaheads[i]) | |
806 | (*this_isocorep)->lookaheads[i] = | |
807 | bitset_create (ntokens, BITSET_FIXED); | |
808 | bitset_andn (lookaheads[i], | |
809 | lookaheads[i], (*this_isocorep)->lookaheads[i]); | |
810 | bitset_or ((*this_isocorep)->lookaheads[i], | |
811 | lookaheads[i], (*this_isocorep)->lookaheads[i]); | |
812 | if (!bitset_empty_p (lookaheads[i])) | |
813 | new_lookaheads = true; | |
814 | } | |
815 | } | |
816 | ||
817 | /* If new lookaheads were merged, propagate those lookaheads to the | |
818 | successors, possibly splitting them. If *tp is being recomputed for | |
819 | the first time, this isn't necessary because the main | |
820 | ielr_split_states loop will handle the successors later. */ | |
821 | if (!(*this_isocorep)->recomputedAsSuccessor) | |
822 | (*this_isocorep)->recomputedAsSuccessor = true; | |
823 | else if (new_lookaheads) | |
824 | { | |
825 | int i; | |
826 | /* When merging demands identical lookahead sets, it is impossible to | |
827 | merge new lookaheads. */ | |
828 | aver (annotation_lists); | |
829 | for (i = 0; i < (*tp)->transitions->num; ++i) | |
830 | { | |
831 | state *t2 = (*tp)->transitions->states[i]; | |
832 | /* At any time, there's at most one state for which we have so | |
833 | far initially recomputed only some of its successors in the | |
834 | main ielr_split_states loop. Because we recompute successors | |
835 | in order, we can just stop at the first such successor. Of | |
836 | course, *tp might be a state some of whose successors have | |
837 | been recomputed as successors of other states rather than as | |
838 | successors of *tp. It's fine if we go ahead and propagate to | |
839 | some of those. We'll propagate to them again (but stop when | |
840 | we see nothing changes) and to the others when we reach *tp in | |
841 | the main ielr_split_states loop later. */ | |
842 | if (!t2->state_list->recomputedAsSuccessor) | |
843 | break; | |
844 | AnnotationList__computeLookaheadFilter ( | |
845 | annotation_lists[t2->state_list->lr0Isocore->state->number], | |
846 | t2->nitems, work2); | |
847 | ielr_compute_lookaheads (follow_kernel_items, always_follows, | |
848 | (*this_isocorep), t2, work2, | |
849 | lookaheads); | |
850 | /* FIXME: If splitting t2 here, it's possible that lookaheads | |
851 | that had already propagated from *tp to t2 will be left in t2 | |
852 | after *tp has been removed as t2's predecessor: | |
853 | - We're going to recompute all lookaheads in phase 4, so these | |
854 | extra lookaheads won't appear in the final parser table. | |
855 | - If t2 has just one annotation, then these extra lookaheads | |
856 | cannot alter the dominating contribution to the associated | |
857 | inadequacy and thus cannot needlessly prevent a future merge | |
858 | of some new state with t2. We can be sure of this because: | |
859 | - The fact that we're splitting t2 now means that some | |
860 | predecessors (at least one) other than *tp must be | |
861 | propagating contributions to t2. | |
862 | - The fact that t2 was merged in the first place means that, | |
863 | if *tp propagated any contributions, the dominating | |
864 | contribution must be the same as that from those other | |
865 | predecessors. | |
866 | - Thus, if some new state to be merged with t2 in the future | |
867 | proves to be compatible with the contributions propagated | |
868 | by the other predecessors, it will also be compatible with | |
869 | the contributions made by the extra lookaheads left behind | |
870 | by *tp. | |
871 | - However, if t2 has more than one annotation and these extra | |
872 | lookaheads contribute to one of their inadequacies, it's | |
873 | possible these extra lookaheads may needlessly prevent a | |
874 | future merge with t2. For example: | |
875 | - Let's say there's an inadequacy A that makes the split | |
876 | necessary as follows: | |
877 | - There's currently just one other predecessor and it | |
878 | propagates to t2 some contributions to inadequacy A. | |
879 | - The new lookaheads that we were attempting to propagate | |
880 | from *tp to t2 made contributions to inadequacy A with a | |
881 | different dominating contribution than those from that | |
882 | other predecessor. | |
883 | - The extra lookaheads either make no contribution to | |
884 | inadequacy A or have the same dominating contribution as | |
885 | the contributions from the other predecessor. Either | |
886 | way, as explained above, they can't prevent a future | |
887 | merge. | |
888 | - Let's say there's an inadequacy B that causes the trouble | |
889 | with future merges as follows: | |
890 | - The extra lookaheads make contributions to inadequacy B. | |
891 | - Those extra contributions did not prevent the original | |
892 | merge to create t2 because the other predecessor | |
893 | propagates to t2 no contributions to inadequacy B. | |
894 | - Thus, those extra contributions may prevent a future | |
895 | merge with t2 even though the merge would be fine if *tp | |
896 | had not left them behind. | |
897 | - Is the latter case common enough to worry about? | |
898 | - Perhaps we should track all predecessors and iterate them | |
899 | now to recreate t2 without those extra lookaheads. */ | |
900 | ielr_compute_state (follow_kernel_items, always_follows, | |
901 | annotation_lists, t2, lookaheads, | |
902 | last_statep, work1, work2, | |
903 | &(*tp)->transitions->states[i]); | |
904 | } | |
905 | } | |
906 | } | |
907 | ||
908 | /* Create a new isocore. */ | |
909 | else | |
910 | { | |
911 | state_list *old_isocore = *this_isocorep; | |
912 | (*last_statep)->next = *this_isocorep = xmalloc (sizeof **last_statep); | |
913 | *last_statep = *this_isocorep; | |
914 | (*last_statep)->state = *tp = state_new_isocore (t); | |
915 | (*tp)->state_list = *last_statep; | |
916 | (*last_statep)->recomputedAsSuccessor = true; | |
917 | (*last_statep)->next = NULL; | |
918 | (*last_statep)->lookaheads = NULL; | |
919 | if (has_lookaheads) | |
920 | { | |
921 | size_t i; | |
922 | (*last_statep)->lookaheads = | |
923 | xnmalloc (t->nitems, sizeof (*last_statep)->lookaheads); | |
924 | for (i = 0; i < t->nitems; ++i) | |
925 | { | |
926 | if (bitset_empty_p (lookaheads[i])) | |
927 | (*last_statep)->lookaheads[i] = NULL; | |
928 | else | |
929 | { | |
930 | (*last_statep)->lookaheads[i] = | |
931 | bitset_create (ntokens, BITSET_FIXED); | |
932 | bitset_copy ((*last_statep)->lookaheads[i], lookaheads[i]); | |
933 | } | |
934 | } | |
935 | } | |
936 | (*last_statep)->lr0Isocore = lr0_isocore; | |
937 | (*last_statep)->nextIsocore = old_isocore; | |
938 | } | |
939 | } | |
940 | ||
941 | /** | |
942 | * \pre | |
943 | * - \c follow_kernel_items and \c always_follows were computed by | |
944 | * \c ielr_compute_auxiliary_tables. | |
945 | * - Either: | |
946 | * - <tt>annotation_lists = NULL</tt> and <tt>max_annotations=0</tt>. | |
947 | * - \c annotation_lists and \c max_annotations were computed by | |
948 | * \c ielr_compute_annotation_lists. | |
949 | * \post | |
950 | * - \c ::states is of size \c ::nstates (which might be greater than | |
951 | * <tt>::nstates \@pre</tt>) and no longer contains any LR(1)-relative | |
952 | * inadequacy. \c annotation_lists was used to determine state | |
953 | * compatibility or, if <tt>annotation_lists = NULL</tt>, the canonical | |
954 | * LR(1) state compatibility test was used. | |
955 | * - If <tt>annotation_lists = NULL</tt>, reduction lookahead sets were | |
956 | * computed in all states. TV_IELR_PHASE4 was pushed while they were | |
957 | * computed from item lookahead sets. | |
958 | */ | |
959 | static void | |
960 | ielr_split_states (bitsetv follow_kernel_items, bitsetv always_follows, | |
961 | AnnotationList **annotation_lists, | |
962 | AnnotationIndex max_annotations) | |
963 | { | |
964 | state_list *first_state; | |
965 | state_list *last_state; | |
966 | bitsetv lookahead_filter = NULL; | |
967 | bitsetv lookaheads; | |
968 | ||
969 | /* Set up state list and some reusable bitsets. */ | |
970 | { | |
971 | size_t max_nitems = 0; | |
972 | state_number i; | |
973 | state_list **nodep = &first_state; | |
974 | for (i = 0; i < nstates; ++i) | |
975 | { | |
976 | *nodep = states[i]->state_list = last_state = xmalloc (sizeof **nodep); | |
977 | (*nodep)->state = states[i]; | |
978 | (*nodep)->recomputedAsSuccessor = false; | |
979 | (*nodep)->lookaheads = NULL; | |
980 | (*nodep)->lr0Isocore = *nodep; | |
981 | (*nodep)->nextIsocore = *nodep; | |
982 | nodep = &(*nodep)->next; | |
983 | if (states[i]->nitems > max_nitems) | |
984 | max_nitems = states[i]->nitems; | |
985 | } | |
986 | *nodep = NULL; | |
987 | lookahead_filter = bitsetv_create (max_nitems, ntokens, BITSET_FIXED); | |
988 | if (!annotation_lists) | |
989 | bitsetv_ones (lookahead_filter); | |
990 | lookaheads = bitsetv_create (max_nitems, ntokens, BITSET_FIXED); | |
991 | } | |
992 | ||
993 | /* Recompute states. */ | |
994 | { | |
995 | ContributionIndex *work = xnmalloc (max_annotations, sizeof *work); | |
996 | state_list *this_state; | |
997 | for (this_state = first_state; this_state; this_state = this_state->next) | |
998 | { | |
999 | state *s = this_state->state; | |
1000 | int i; | |
1001 | for (i = 0; i < s->transitions->num; ++i) | |
1002 | { | |
1003 | state *t = s->transitions->states[i]; | |
1004 | if (annotation_lists) | |
1005 | AnnotationList__computeLookaheadFilter ( | |
1006 | annotation_lists[t->state_list->lr0Isocore->state->number], | |
1007 | t->nitems, lookahead_filter); | |
1008 | ielr_compute_lookaheads (follow_kernel_items, always_follows, | |
1009 | this_state, t, lookahead_filter, | |
1010 | lookaheads); | |
1011 | ielr_compute_state (follow_kernel_items, always_follows, | |
1012 | annotation_lists, t, lookaheads, &last_state, | |
1013 | work, lookahead_filter, | |
1014 | &s->transitions->states[i]); | |
1015 | } | |
1016 | } | |
1017 | free (work); | |
1018 | } | |
1019 | ||
1020 | bitsetv_free (lookahead_filter); | |
1021 | bitsetv_free (lookaheads); | |
1022 | ||
1023 | /* Store states back in the states array. */ | |
1024 | states = xnrealloc (states, nstates, sizeof *states); | |
1025 | { | |
1026 | state_list *node; | |
1027 | for (node = first_state; node; node = node->next) | |
1028 | states[node->state->number] = node->state; | |
1029 | } | |
1030 | ||
1031 | /* In the case of canonical LR(1), copy item lookahead sets to reduction | |
1032 | lookahead sets. */ | |
1033 | if (!annotation_lists) | |
1034 | { | |
1035 | timevar_push (TV_IELR_PHASE4); | |
1036 | initialize_LA (); | |
1037 | state_list *node; | |
1038 | for (node = first_state; node; node = node->next) | |
1039 | if (!node->state->consistent) | |
1040 | { | |
1041 | size_t i = 0; | |
1042 | item_number *itemset = node->state->items; | |
1043 | size_t r; | |
1044 | for (r = 0; r < node->state->reductions->num; ++r) | |
1045 | { | |
1046 | rule *this_rule = node->state->reductions->rules[r]; | |
1047 | bitset lookahead_set = | |
1048 | node->state->reductions->lookahead_tokens[r]; | |
1049 | if (item_number_is_rule_number (*this_rule->rhs)) | |
1050 | ielr_compute_goto_follow_set (follow_kernel_items, | |
1051 | always_follows, node, | |
1052 | this_rule->lhs, lookahead_set); | |
1053 | else if (node->lookaheads) | |
1054 | { | |
1055 | /* We don't need to start the kernel item search back at | |
1056 | i=0 because both items and reductions are sorted on rule | |
1057 | number. */ | |
1058 | while (!item_number_is_rule_number (ritem[itemset[i]]) | |
1059 | || item_number_as_rule_number (ritem[itemset[i]]) | |
1060 | != this_rule->number) | |
1061 | { | |
1062 | ++i; | |
1063 | aver (i < node->state->nitems); | |
1064 | } | |
1065 | if (node->lookaheads[i]) | |
1066 | bitset_copy (lookahead_set, node->lookaheads[i]); | |
1067 | } | |
1068 | } | |
1069 | } | |
1070 | timevar_pop (TV_IELR_PHASE4); | |
1071 | } | |
1072 | ||
1073 | /* Free state list. */ | |
1074 | while (first_state) | |
1075 | { | |
1076 | state_list *node = first_state; | |
1077 | if (node->lookaheads) | |
1078 | { | |
1079 | size_t i; | |
1080 | for (i = 0; i < node->state->nitems; ++i) | |
1081 | if (node->lookaheads[i]) | |
1082 | bitset_free (node->lookaheads[i]); | |
1083 | free (node->lookaheads); | |
1084 | } | |
1085 | first_state = node->next; | |
1086 | free (node); | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | void | |
1091 | ielr (void) | |
1092 | { | |
1093 | LrType lr_type; | |
1094 | ||
1095 | /* Examine user options. */ | |
1096 | { | |
1097 | char *type = muscle_percent_define_get ("lr.type"); | |
3a414bbf | 1098 | if (0 == strcmp (type, "lalr")) |
166366b2 | 1099 | lr_type = LR_TYPE__LALR; |
3a414bbf | 1100 | else if (0 == strcmp (type, "ielr")) |
166366b2 | 1101 | lr_type = LR_TYPE__IELR; |
3a414bbf | 1102 | else if (0 == strcmp (type, "canonical-lr")) |
166366b2 JD |
1103 | lr_type = LR_TYPE__CANONICAL_LR; |
1104 | else | |
1105 | aver (false); | |
1106 | free (type); | |
1107 | } | |
1108 | ||
1109 | /* Phase 0: LALR(1). */ | |
1110 | timevar_push (TV_LALR); | |
1111 | if (lr_type == LR_TYPE__CANONICAL_LR) | |
1112 | set_goto_map (); | |
1113 | else | |
1114 | lalr (); | |
1115 | if (lr_type == LR_TYPE__LALR) | |
1116 | { | |
1117 | bitsetv_free (goto_follows); | |
1118 | timevar_pop (TV_LALR); | |
1119 | return; | |
1120 | } | |
1121 | timevar_pop (TV_LALR); | |
1122 | ||
1123 | { | |
1124 | bitsetv follow_kernel_items; | |
1125 | bitsetv always_follows; | |
1126 | InadequacyList **inadequacy_lists = NULL; | |
1127 | AnnotationList **annotation_lists = NULL; | |
1128 | struct obstack annotations_obstack; | |
1129 | AnnotationIndex max_annotations = 0; | |
1130 | ||
1131 | { | |
1132 | /* Phase 1: Compute Auxiliary Tables. */ | |
1133 | state ***predecessors; | |
1134 | timevar_push (TV_IELR_PHASE1); | |
1135 | ielr_compute_auxiliary_tables ( | |
1136 | &follow_kernel_items, &always_follows, | |
1137 | lr_type == LR_TYPE__CANONICAL_LR ? NULL : &predecessors); | |
1138 | timevar_pop (TV_IELR_PHASE1); | |
1139 | ||
1140 | /* Phase 2: Compute Annotations. */ | |
1141 | timevar_push (TV_IELR_PHASE2); | |
1142 | if (lr_type != LR_TYPE__CANONICAL_LR) | |
1143 | { | |
1144 | obstack_init (&annotations_obstack); | |
1145 | ielr_compute_annotation_lists (follow_kernel_items, always_follows, | |
1146 | predecessors, &max_annotations, | |
1147 | &inadequacy_lists, &annotation_lists, | |
1148 | &annotations_obstack); | |
1149 | { | |
1150 | state_number i; | |
1151 | for (i = 0; i < nstates; ++i) | |
1152 | free (predecessors[i]); | |
1153 | } | |
1154 | free (predecessors); | |
1155 | bitsetv_free (goto_follows); | |
1156 | lalr_free (); | |
1157 | } | |
1158 | timevar_pop (TV_IELR_PHASE2); | |
1159 | } | |
1160 | ||
1161 | /* Phase 3: Split States. */ | |
1162 | timevar_push (TV_IELR_PHASE3); | |
1163 | { | |
1164 | state_number nstates_lr0 = nstates; | |
1165 | ielr_split_states (follow_kernel_items, always_follows, | |
1166 | annotation_lists, max_annotations); | |
1167 | if (inadequacy_lists) | |
1168 | { | |
1169 | state_number i; | |
1170 | for (i = 0; i < nstates_lr0; ++i) | |
1171 | InadequacyList__delete (inadequacy_lists[i]); | |
1172 | } | |
1173 | } | |
1174 | free (inadequacy_lists); | |
1175 | if (annotation_lists) | |
1176 | obstack_free (&annotations_obstack, NULL); | |
1177 | free (annotation_lists); | |
1178 | bitsetv_free (follow_kernel_items); | |
1179 | bitsetv_free (always_follows); | |
1180 | timevar_pop (TV_IELR_PHASE3); | |
1181 | } | |
1182 | ||
1183 | /* Phase 4: Compute Reduction Lookaheads. */ | |
1184 | timevar_push (TV_IELR_PHASE4); | |
1185 | free (goto_map); | |
1186 | free (from_state); | |
1187 | free (to_state); | |
1188 | if (lr_type == LR_TYPE__CANONICAL_LR) | |
1189 | { | |
1190 | // Reduction lookaheads are computed in ielr_split_states above but are | |
1191 | // timed as part of phase 4. | |
1192 | set_goto_map (); | |
1193 | } | |
1194 | else | |
1195 | { | |
1196 | lalr (); | |
1197 | bitsetv_free (goto_follows); | |
1198 | } | |
1199 | timevar_pop (TV_IELR_PHASE4); | |
1200 | } |