]>
Commit | Line | Data |
---|---|---|
1 | /* Generate the nondeterministic finite state machine for bison, | |
2 | Copyright 1984, 1986, 1989, 2000, 2001, 2002 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 "bitset.h" | |
27 | #include "quotearg.h" | |
28 | #include "symtab.h" | |
29 | #include "gram.h" | |
30 | #include "getargs.h" | |
31 | #include "reader.h" | |
32 | #include "gram.h" | |
33 | #include "state.h" | |
34 | #include "complain.h" | |
35 | #include "closure.h" | |
36 | #include "LR0.h" | |
37 | #include "lalr.h" | |
38 | #include "reduce.h" | |
39 | ||
40 | unsigned int nstates = 0; | |
41 | /* Initialize the final state to -1, otherwise, it might be set to 0 | |
42 | by default, and since we don't compute the reductions of the final | |
43 | state, we end up not computing the reductions of the initial state, | |
44 | which is of course needed. | |
45 | ||
46 | FINAL_STATE is properly set by new_state when it recognizes the | |
47 | accessing symbol: EOF. */ | |
48 | int final_state = -1; | |
49 | static state_t *first_state = NULL; | |
50 | ||
51 | static state_t *this_state = NULL; | |
52 | static state_t *last_state = NULL; | |
53 | ||
54 | static int nshifts; | |
55 | static token_number_t *shift_symbol = NULL; | |
56 | ||
57 | static short *redset = NULL; | |
58 | static short *shiftset = NULL; | |
59 | ||
60 | static item_number_t **kernel_base = NULL; | |
61 | static int *kernel_size = NULL; | |
62 | static item_number_t *kernel_items = NULL; | |
63 | ||
64 | /* hash table for states, to recognize equivalent ones. */ | |
65 | ||
66 | #define STATE_HASH_SIZE 1009 | |
67 | static state_t **state_hash = NULL; | |
68 | ||
69 | \f | |
70 | static void | |
71 | allocate_itemsets (void) | |
72 | { | |
73 | int i, r; | |
74 | item_number_t *rhsp; | |
75 | ||
76 | /* Count the number of occurrences of all the symbols in RITEMS. | |
77 | Note that useless productions (hence useless nonterminals) are | |
78 | browsed too, hence we need to allocate room for _all_ the | |
79 | symbols. */ | |
80 | int count = 0; | |
81 | short *symbol_count = XCALLOC (short, nsyms + nuseless_nonterminals); | |
82 | ||
83 | for (r = 1; r < nrules + 1; ++r) | |
84 | for (rhsp = rules[r].rhs; *rhsp >= 0; ++rhsp) | |
85 | { | |
86 | count++; | |
87 | symbol_count[*rhsp]++; | |
88 | } | |
89 | ||
90 | /* See comments before new_itemsets. All the vectors of items | |
91 | live inside KERNEL_ITEMS. The number of active items after | |
92 | some symbol cannot be more than the number of times that symbol | |
93 | appears as an item, which is symbol_count[symbol]. | |
94 | We allocate that much space for each symbol. */ | |
95 | ||
96 | kernel_base = XCALLOC (item_number_t *, nsyms); | |
97 | if (count) | |
98 | kernel_items = XCALLOC (item_number_t, count); | |
99 | ||
100 | count = 0; | |
101 | for (i = 0; i < nsyms; i++) | |
102 | { | |
103 | kernel_base[i] = kernel_items + count; | |
104 | count += symbol_count[i]; | |
105 | } | |
106 | ||
107 | free (symbol_count); | |
108 | kernel_size = XCALLOC (int, nsyms); | |
109 | } | |
110 | ||
111 | ||
112 | static void | |
113 | allocate_storage (void) | |
114 | { | |
115 | allocate_itemsets (); | |
116 | ||
117 | shiftset = XCALLOC (short, nsyms); | |
118 | redset = XCALLOC (short, nrules + 1); | |
119 | state_hash = XCALLOC (state_t *, STATE_HASH_SIZE); | |
120 | shift_symbol = XCALLOC (token_number_t, nsyms); | |
121 | } | |
122 | ||
123 | ||
124 | static void | |
125 | free_storage (void) | |
126 | { | |
127 | free (shift_symbol); | |
128 | free (redset); | |
129 | free (shiftset); | |
130 | free (kernel_base); | |
131 | free (kernel_size); | |
132 | XFREE (kernel_items); | |
133 | free (state_hash); | |
134 | } | |
135 | ||
136 | ||
137 | ||
138 | ||
139 | /*----------------------------------------------------------------. | |
140 | | Find which symbols can be shifted in the current state, and for | | |
141 | | each one record which items would be active after that shift. | | |
142 | | Uses the contents of itemset. | | |
143 | | | | |
144 | | shift_symbol is set to a vector of the symbols that can be | | |
145 | | shifted. For each symbol in the grammar, kernel_base[symbol] | | |
146 | | points to a vector of item numbers activated if that symbol is | | |
147 | | shifted, and kernel_size[symbol] is their numbers. | | |
148 | `----------------------------------------------------------------*/ | |
149 | ||
150 | static void | |
151 | new_itemsets (void) | |
152 | { | |
153 | int i; | |
154 | ||
155 | if (trace_flag) | |
156 | fprintf (stderr, "Entering new_itemsets, state = %d\n", | |
157 | this_state->number); | |
158 | ||
159 | for (i = 0; i < nsyms; i++) | |
160 | kernel_size[i] = 0; | |
161 | ||
162 | nshifts = 0; | |
163 | ||
164 | for (i = 0; i < nritemset; ++i) | |
165 | if (ritem[itemset[i]] >= 0) | |
166 | { | |
167 | token_number_t symbol | |
168 | = item_number_as_token_number (ritem[itemset[i]]); | |
169 | if (!kernel_size[symbol]) | |
170 | { | |
171 | shift_symbol[nshifts] = symbol; | |
172 | nshifts++; | |
173 | } | |
174 | ||
175 | kernel_base[symbol][kernel_size[symbol]] = itemset[i] + 1; | |
176 | kernel_size[symbol]++; | |
177 | } | |
178 | } | |
179 | ||
180 | ||
181 | ||
182 | /*-----------------------------------------------------------------. | |
183 | | Subroutine of get_state. Create a new state for those items, if | | |
184 | | necessary. | | |
185 | `-----------------------------------------------------------------*/ | |
186 | ||
187 | static state_t * | |
188 | new_state (token_number_t symbol) | |
189 | { | |
190 | state_t *p; | |
191 | ||
192 | if (trace_flag) | |
193 | fprintf (stderr, "Entering new_state, state = %d, symbol = %d (%s)\n", | |
194 | nstates, symbol, quotearg_style (escape_quoting_style, | |
195 | symbols[symbol]->tag)); | |
196 | ||
197 | if (nstates >= SHRT_MAX) | |
198 | fatal (_("too many states (max %d)"), SHRT_MAX); | |
199 | ||
200 | p = STATE_ALLOC (kernel_size[symbol]); | |
201 | p->accessing_symbol = symbol; | |
202 | p->number = nstates; | |
203 | p->nitems = kernel_size[symbol]; | |
204 | ||
205 | memcpy (p->items, kernel_base[symbol], | |
206 | kernel_size[symbol] * sizeof (kernel_base[symbol][0])); | |
207 | ||
208 | /* If this is the eoftoken, and this is not the initial state, then | |
209 | this is the final state. */ | |
210 | if (symbol == 0 && first_state) | |
211 | final_state = p->number; | |
212 | ||
213 | if (!first_state) | |
214 | first_state = p; | |
215 | if (last_state) | |
216 | last_state->next = p; | |
217 | last_state = p; | |
218 | ||
219 | nstates++; | |
220 | ||
221 | return p; | |
222 | } | |
223 | ||
224 | ||
225 | /*--------------------------------------------------------------. | |
226 | | Find the state number for the state we would get to (from the | | |
227 | | current state) by shifting symbol. Create a new state if no | | |
228 | | equivalent one exists already. Used by append_states. | | |
229 | `--------------------------------------------------------------*/ | |
230 | ||
231 | static int | |
232 | get_state (token_number_t symbol) | |
233 | { | |
234 | int key; | |
235 | int i; | |
236 | state_t *sp; | |
237 | ||
238 | if (trace_flag) | |
239 | fprintf (stderr, "Entering get_state, state = %d, symbol = %d (%s)\n", | |
240 | this_state->number, symbol, quotearg_style (escape_quoting_style, | |
241 | symbols[symbol]->tag)); | |
242 | ||
243 | /* Add up the target state's active item numbers to get a hash key. | |
244 | */ | |
245 | key = 0; | |
246 | for (i = 0; i < kernel_size[symbol]; ++i) | |
247 | key += kernel_base[symbol][i]; | |
248 | key = key % STATE_HASH_SIZE; | |
249 | sp = state_hash[key]; | |
250 | ||
251 | if (sp) | |
252 | { | |
253 | int found = 0; | |
254 | while (!found) | |
255 | { | |
256 | if (sp->nitems == kernel_size[symbol]) | |
257 | { | |
258 | found = 1; | |
259 | for (i = 0; i < kernel_size[symbol]; ++i) | |
260 | if (kernel_base[symbol][i] != sp->items[i]) | |
261 | found = 0; | |
262 | } | |
263 | ||
264 | if (!found) | |
265 | { | |
266 | if (sp->link) | |
267 | { | |
268 | sp = sp->link; | |
269 | } | |
270 | else /* bucket exhausted and no match */ | |
271 | { | |
272 | sp = sp->link = new_state (symbol); | |
273 | found = 1; | |
274 | } | |
275 | } | |
276 | } | |
277 | } | |
278 | else /* bucket is empty */ | |
279 | { | |
280 | state_hash[key] = sp = new_state (symbol); | |
281 | } | |
282 | ||
283 | if (trace_flag) | |
284 | fprintf (stderr, "Exiting get_state => %d\n", sp->number); | |
285 | ||
286 | return sp->number; | |
287 | } | |
288 | ||
289 | /*------------------------------------------------------------------. | |
290 | | Use the information computed by new_itemsets to find the state | | |
291 | | numbers reached by each shift transition from the current state. | | |
292 | | | | |
293 | | shiftset is set up as a vector of state numbers of those states. | | |
294 | `------------------------------------------------------------------*/ | |
295 | ||
296 | static void | |
297 | append_states (void) | |
298 | { | |
299 | int i; | |
300 | int j; | |
301 | token_number_t symbol; | |
302 | ||
303 | if (trace_flag) | |
304 | fprintf (stderr, "Entering append_states, state = %d\n", | |
305 | this_state->number); | |
306 | ||
307 | /* first sort shift_symbol into increasing order */ | |
308 | ||
309 | for (i = 1; i < nshifts; i++) | |
310 | { | |
311 | symbol = shift_symbol[i]; | |
312 | j = i; | |
313 | while (j > 0 && shift_symbol[j - 1] > symbol) | |
314 | { | |
315 | shift_symbol[j] = shift_symbol[j - 1]; | |
316 | j--; | |
317 | } | |
318 | shift_symbol[j] = symbol; | |
319 | } | |
320 | ||
321 | for (i = 0; i < nshifts; i++) | |
322 | shiftset[i] = get_state (shift_symbol[i]); | |
323 | } | |
324 | ||
325 | ||
326 | static void | |
327 | new_states (void) | |
328 | { | |
329 | /* The 0 at the lhs is the index of the item of this initial rule. */ | |
330 | kernel_base[0][0] = 0; | |
331 | kernel_size[0] = 1; | |
332 | this_state = new_state (0); | |
333 | } | |
334 | ||
335 | ||
336 | /*------------------------------------------------------------. | |
337 | | Save the NSHIFTS of SHIFTSET into the current linked list. | | |
338 | `------------------------------------------------------------*/ | |
339 | ||
340 | static void | |
341 | save_shifts (void) | |
342 | { | |
343 | shifts *p = shifts_new (nshifts); | |
344 | memcpy (p->shifts, shiftset, nshifts * sizeof (shiftset[0])); | |
345 | this_state->shifts = p; | |
346 | } | |
347 | ||
348 | ||
349 | /*----------------------------------------------------------------. | |
350 | | Find which rules can be used for reduction transitions from the | | |
351 | | current state and make a reductions structure for the state to | | |
352 | | record their rule numbers. | | |
353 | `----------------------------------------------------------------*/ | |
354 | ||
355 | static void | |
356 | save_reductions (void) | |
357 | { | |
358 | int count = 0; | |
359 | int i; | |
360 | ||
361 | /* If this is the final state, we want it to have no reductions at | |
362 | all, although it has one for `START_SYMBOL EOF .'. */ | |
363 | if (this_state->number == final_state) | |
364 | return; | |
365 | ||
366 | /* Find and count the active items that represent ends of rules. */ | |
367 | for (i = 0; i < nritemset; ++i) | |
368 | { | |
369 | int item = ritem[itemset[i]]; | |
370 | if (item < 0) | |
371 | redset[count++] = -item; | |
372 | } | |
373 | ||
374 | /* Make a reductions structure and copy the data into it. */ | |
375 | this_state->reductions = reductions_new (count); | |
376 | memcpy (this_state->reductions->rules, redset, count * sizeof (redset[0])); | |
377 | } | |
378 | ||
379 | \f | |
380 | /*---------------. | |
381 | | Build STATES. | | |
382 | `---------------*/ | |
383 | ||
384 | static void | |
385 | set_states (void) | |
386 | { | |
387 | state_t *sp; | |
388 | states = XCALLOC (state_t *, nstates); | |
389 | ||
390 | for (sp = first_state; sp; sp = sp->next) | |
391 | { | |
392 | /* Pessimization, but simplification of the code: make sure all | |
393 | the states have a shifts, errs, and reductions, even if | |
394 | reduced to 0. */ | |
395 | if (!sp->shifts) | |
396 | sp->shifts = shifts_new (0); | |
397 | if (!sp->errs) | |
398 | sp->errs = errs_new (0); | |
399 | if (!sp->reductions) | |
400 | sp->reductions = reductions_new (0); | |
401 | ||
402 | states[sp->number] = sp; | |
403 | } | |
404 | } | |
405 | ||
406 | /*-------------------------------------------------------------------. | |
407 | | Compute the nondeterministic finite state machine (see state.h for | | |
408 | | details) from the grammar. | | |
409 | `-------------------------------------------------------------------*/ | |
410 | ||
411 | void | |
412 | generate_states (void) | |
413 | { | |
414 | allocate_storage (); | |
415 | new_closure (nritems); | |
416 | new_states (); | |
417 | ||
418 | while (this_state) | |
419 | { | |
420 | if (trace_flag) | |
421 | fprintf (stderr, "Processing state %d (reached by %s)\n", | |
422 | this_state->number, | |
423 | quotearg_style (escape_quoting_style, | |
424 | symbols[this_state->accessing_symbol]->tag)); | |
425 | /* Set up ruleset and itemset for the transitions out of this | |
426 | state. ruleset gets a 1 bit for each rule that could reduce | |
427 | now. itemset gets a vector of all the items that could be | |
428 | accepted next. */ | |
429 | closure (this_state->items, this_state->nitems); | |
430 | /* record the reductions allowed out of this state */ | |
431 | save_reductions (); | |
432 | /* find the itemsets of the states that shifts can reach */ | |
433 | new_itemsets (); | |
434 | /* find or create the core structures for those states */ | |
435 | append_states (); | |
436 | ||
437 | /* create the shifts structures for the shifts to those states, | |
438 | now that the state numbers transitioning to are known */ | |
439 | save_shifts (); | |
440 | ||
441 | /* states are queued when they are created; process them all */ | |
442 | this_state = this_state->next; | |
443 | } | |
444 | ||
445 | /* discard various storage */ | |
446 | free_closure (); | |
447 | free_storage (); | |
448 | ||
449 | /* Set up STATES. */ | |
450 | set_states (); | |
451 | } |