X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/80a697503460fcf0d52bed9fb3b86e4c276fb37e..e91ff6ce3d2653b41265591dafd4bdb32ae114c2:/src/lalr.c diff --git a/src/lalr.c b/src/lalr.c index 54b0391c..643f8fd9 100644 --- a/src/lalr.c +++ b/src/lalr.c @@ -1,5 +1,6 @@ /* Compute look-ahead criteria for bison, - Copyright 1984, 1986, 1989, 2000, 2001 Free Software Foundation, Inc. + Copyright (C) 1984, 1986, 1989, 2000, 2001, 2002 + Free Software Foundation, Inc. This file is part of Bison, the GNU Compiler Compiler. @@ -24,52 +25,58 @@ tokens they accept. */ #include "system.h" +#include "bitset.h" +#include "bitsetv.h" +#include "quotearg.h" +#include "symtab.h" +#include "gram.h" +#include "reader.h" #include "types.h" #include "LR0.h" -#include "gram.h" #include "complain.h" #include "lalr.h" #include "nullable.h" #include "derives.h" +#include "getargs.h" -/* All the decorated states, indexed by the state number. Warning: - there is a state_TABLE in LR0.c, but it is different and static. - */ -state_t *state_table = NULL; - -int tokensetsize; -short *LAruleno; -unsigned *LA; +/* All the decorated states, indexed by the state number. */ +state_t **states = NULL; -short *goto_map; -short *from_state; -short *to_state; +rule_t **LArule = NULL; +bitsetv LA = NULL; +size_t nLA; -extern void berror PARAMS ((const char *)); - -static int infinity; static int ngotos; +short *goto_map = NULL; +short *from_state = NULL; +short *to_state = NULL; /* And for the famous F variable, which name is so descriptive that a comment is hardly needed. . */ -static unsigned *F = NULL; -#define F(Rule) (F + (Rule) * tokensetsize) +static bitsetv F = NULL; static short **includes; static shorts **lookback; + + +/*---------------------------------------------------------------. +| digraph & traverse. | +| | +| The following variables are used as common storage between the | +| two. | +`---------------------------------------------------------------*/ + static short **R; static short *INDEX; static short *VERTICES; static int top; - +static int infinity; static void traverse (int i) { int j; - size_t k; int height; - size_t size = F (i + 1) - F(i); VERTICES[++top] = i; INDEX[i] = height = top; @@ -83,8 +90,7 @@ traverse (int i) if (INDEX[i] > INDEX[R[i][j]]) INDEX[i] = INDEX[R[i][j]]; - for (k = 0; k < size; ++k) - F (i)[k] |= F (R[i][j])[k]; + bitset_or (F[i], F[i], F[R[i][j]]); } if (INDEX[i] == height) @@ -96,8 +102,7 @@ traverse (int i) if (i == j) break; - for (k = 0; k < size; ++k) - F (i)[k] = F (j)[k]; + bitset_copy (F[j], F[i]); } } @@ -126,192 +131,79 @@ digraph (short **relation) } -/*--------------------. -| Build STATE_TABLE. | -`--------------------*/ - -static void -set_state_table (void) -{ - /* NSTATES + 1 because lookahead for the pseudo state number NSTATES - might be used (see conflicts.c). It is too opaque for me to - provide a probably less hacky implementation. --akim */ - state_table = XCALLOC (state_t, nstates + 1); - - { - core *sp; - for (sp = first_state; sp; sp = sp->next) - { - state_table[sp->number].state = sp; - state_table[sp->number].accessing_symbol = sp->accessing_symbol; - } - } - - { - shifts *sp; - for (sp = first_shift; sp; sp = sp->next) - state_table[sp->number].shift_table = sp; - } - - { - reductions *rp; - for (rp = first_reduction; rp; rp = rp->next) - state_table[rp->number].reduction_table = rp; - } - - /* Initializing the lookaheads members. Please note that it must be - performed after having set some of the other members which are - used below. Change with extreme caution. */ - { - int i; - int count = 0; - for (i = 0; i < nstates; i++) - { - int k; - reductions *rp = state_table[i].reduction_table; - shifts *sp = state_table[i].shift_table; - - state_table[i].lookaheads = count; - - if (rp - && (rp->nreds > 1 - || (sp && !ISVAR (state_table[sp->shifts[0]].accessing_symbol)))) - count += rp->nreds; - else - state_table[i].consistent = 1; - - if (sp) - for (k = 0; k < sp->nshifts; k++) - if (state_table[sp->shifts[k]].accessing_symbol - == error_token_number) - { - state_table[i].consistent = 0; - break; - } - } - state_table[nstates].lookaheads = count; - } -} - - -/*------------------------------------------. -| Return the size of the longest rule RHS. | -`------------------------------------------*/ - -static size_t -maxrhs (void) -{ - short *itemp; - int length; - int max; - - length = 0; - max = 0; - for (itemp = ritem; *itemp; itemp++) - { - if (*itemp > 0) - { - length++; - } - else - { - if (length > max) - max = length; - length = 0; - } - } - - return max; -} - - static void initialize_LA (void) { - int i; + size_t i; int j; - short *np; - reductions *rp; + rule_t **np; - size_t nLA = state_table[nstates].lookaheads; + /* Avoid having to special case 0. */ if (!nLA) nLA = 1; - LA = XCALLOC (unsigned, nLA * tokensetsize); - LAruleno = XCALLOC (short, nLA); + LA = bitsetv_create (nLA, ntokens, BITSET_FIXED); + LArule = XCALLOC (rule_t *, nLA); lookback = XCALLOC (shorts *, nLA); - np = LAruleno; + np = LArule; for (i = 0; i < nstates; i++) - if (!state_table[i].consistent) - if ((rp = state_table[i].reduction_table)) - for (j = 0; j < rp->nreds; j++) - *np++ = rp->rules[j]; + if (!states[i]->consistent) + for (j = 0; j < states[i]->reductions->nreds; j++) + *np++ = &rules[states[i]->reductions->rules[j]]; } static void set_goto_map (void) { - shifts *sp; + size_t state; int i; - int symbol; - int k; short *temp_map; - int state2; - int state1; goto_map = XCALLOC (short, nvars + 1) - ntokens; temp_map = XCALLOC (short, nvars + 1) - ntokens; ngotos = 0; - for (sp = first_shift; sp; sp = sp->next) + for (state = 0; state < nstates; ++state) { - for (i = sp->nshifts - 1; i >= 0; i--) + shifts *sp = states[state]->shifts; + for (i = sp->nshifts - 1; i >= 0 && SHIFT_IS_GOTO (sp, i); --i) { - symbol = state_table[sp->shifts[i]].accessing_symbol; - - if (ISTOKEN (symbol)) - break; - - if (ngotos == MAXSHORT) - fatal (_("too many gotos (max %d)"), MAXSHORT); + if (ngotos == SHRT_MAX) + fatal (_("too many gotos (max %d)"), SHRT_MAX); ngotos++; - goto_map[symbol]++; + goto_map[SHIFT_SYMBOL (sp, i)]++; } } - k = 0; - for (i = ntokens; i < nsyms; i++) - { - temp_map[i] = k; - k += goto_map[i]; - } + { + int k = 0; + for (i = ntokens; i < nsyms; i++) + { + temp_map[i] = k; + k += goto_map[i]; + } - for (i = ntokens; i < nsyms; i++) - goto_map[i] = temp_map[i]; + for (i = ntokens; i < nsyms; i++) + goto_map[i] = temp_map[i]; - goto_map[nsyms] = ngotos; - temp_map[nsyms] = ngotos; + goto_map[nsyms] = ngotos; + temp_map[nsyms] = ngotos; + } from_state = XCALLOC (short, ngotos); to_state = XCALLOC (short, ngotos); - for (sp = first_shift; sp; sp = sp->next) + for (state = 0; state < nstates; ++state) { - state1 = sp->number; - for (i = sp->nshifts - 1; i >= 0; i--) + shifts *sp = states[state]->shifts; + for (i = sp->nshifts - 1; i >= 0 && SHIFT_IS_GOTO (sp, i); --i) { - state2 = sp->shifts[i]; - symbol = state_table[state2].accessing_symbol; - - if (ISTOKEN (symbol)) - break; - - k = temp_map[symbol]++; - from_state[k] = state1; - to_state[k] = state2; + int k = temp_map[SHIFT_SYMBOL (sp, i)]++; + from_state[k] = state; + to_state[k] = sp->shifts[i]; } } @@ -325,7 +217,7 @@ set_goto_map (void) `----------------------------------------------------------*/ static int -map_goto (int state, int symbol) +map_goto (int state, token_number_t symbol) { int high; int low; @@ -356,59 +248,37 @@ map_goto (int state, int symbol) static void initialize_F (void) { + short **reads = XCALLOC (short *, ngotos); + short *edge = XCALLOC (short, ngotos + 1); + int nedges = 0; + int i; - int j; - short *edge; - unsigned *rowp; - short *rp; - short **reads; - int nedges; - int symbol; - int nwords; - - nwords = ngotos * tokensetsize; - F = XCALLOC (unsigned, nwords); - - reads = XCALLOC (short *, ngotos); - edge = XCALLOC (short, ngotos + 1); - nedges = 0; - - rowp = F; + + F = bitsetv_create (ngotos, ntokens, BITSET_FIXED); + for (i = 0; i < ngotos; i++) { int stateno = to_state[i]; - shifts *sp = state_table[stateno].shift_table; - - if (sp) - { - for (j = 0; j < sp->nshifts; j++) - { - symbol = state_table[sp->shifts[j]].accessing_symbol; - if (ISVAR (symbol)) - break; - SETBIT (rowp, symbol); - } + shifts *sp = states[stateno]->shifts; - for (; j < sp->nshifts; j++) - { - symbol = state_table[sp->shifts[j]].accessing_symbol; - if (nullable[symbol]) - edge[nedges++] = map_goto (stateno, symbol); - } + int j; + for (j = 0; j < sp->nshifts && SHIFT_IS_SHIFT (sp, j); j++) + bitset_set (F[i], SHIFT_SYMBOL (sp, j)); - if (nedges) - { - reads[i] = rp = XCALLOC (short, nedges + 1); - - for (j = 0; j < nedges; j++) - rp[j] = edge[j]; - - rp[nedges] = -1; - nedges = 0; - } + for (; j < sp->nshifts; j++) + { + token_number_t symbol = SHIFT_SYMBOL (sp, j); + if (nullable[symbol]) + edge[nedges++] = map_goto (stateno, symbol); } - rowp += tokensetsize; + if (nedges) + { + reads[i] = XCALLOC (short, nedges + 1); + memcpy (reads[i], edge, nedges * sizeof (edge[0])); + reads[i][nedges] = -1; + nedges = 0; + } } digraph (reads); @@ -422,77 +292,106 @@ initialize_F (void) static void -add_lookback_edge (int stateno, int ruleno, int gotono) +add_lookback_edge (state_t *state, int ruleno, int gotono) { int i; - int k; - int found; shorts *sp; - i = state_table[stateno].lookaheads; - k = state_table[stateno + 1].lookaheads; - found = 0; - while (!found && i < k) - { - if (LAruleno[i] == ruleno) - found = 1; - else - i++; - } + for (i = 0; i < state->nlookaheads; ++i) + if (LArule[state->lookaheadsp + i]->number == ruleno) + break; - assert (found); + assert (LArule[state->lookaheadsp + i]->number == ruleno); sp = XCALLOC (shorts, 1); - sp->next = lookback[i]; + sp->next = lookback[state->lookaheadsp + i]; sp->value = gotono; - lookback[i] = sp; + lookback[state->lookaheadsp + i] = sp; } -static short ** -transpose (short **R_arg, int n) +static void +matrix_print (FILE *out, short **matrix, int n) { - short **new_R; - short **temp_R; - short *nedges; - int i; + int i, j; + + for (i = 0; i < n; ++i) + { + fprintf (out, "%3d: ", i); + if (matrix[i]) + for (j = 0; matrix[i][j] != -1; ++j) + fprintf (out, "%3d ", matrix[i][j]); + fputc ('\n', out); + } + fputc ('\n', out); +} - nedges = XCALLOC (short, n); +/*-------------------------------------------------------------------. +| Return the transpose of R_ARG, of size N. Destroy R_ARG, as it is | +| replaced with the result. | +| | +| R_ARG[I] is NULL or a -1 terminated list of numbers. | +| | +| RESULT[NUM] is NULL or the -1 terminated list of the I such as NUM | +| is in R_ARG[I]. | +`-------------------------------------------------------------------*/ - for (i = 0; i < n; i++) +static short ** +transpose (short **R_arg, int n) +{ + /* The result. */ + short **new_R = XCALLOC (short *, n); + /* END_R[I] -- next entry of NEW_R[I]. */ + short **end_R = XCALLOC (short *, n); + /* NEDGES[I] -- total size of NEW_R[I]. */ + short *nedges = XCALLOC (short, n); + int i, j; + + if (trace_flag) { - short *sp = R_arg[i]; - if (sp) - { - while (*sp >= 0) - nedges[*sp++]++; - } + fputs ("transpose: input\n", stderr); + matrix_print (stderr, R_arg, n); } - new_R = XCALLOC (short *, n); - temp_R = XCALLOC (short *, n); + /* Count. */ + for (i = 0; i < n; i++) + if (R_arg[i]) + for (j = 0; R_arg[i][j] >= 0; ++j) + ++nedges[R_arg[i][j]]; + /* Allocate. */ for (i = 0; i < n; i++) if (nedges[i] > 0) { short *sp = XCALLOC (short, nedges[i] + 1); - new_R[i] = sp; - temp_R[i] = sp; sp[nedges[i]] = -1; + new_R[i] = sp; + end_R[i] = sp; } - XFREE (nedges); + /* Store. */ + for (i = 0; i < n; i++) + if (R_arg[i]) + for (j = 0; R_arg[i][j] >= 0; ++j) + { + *end_R[R_arg[i][j]] = i; + ++end_R[R_arg[i][j]]; + } + + free (nedges); + free (end_R); + /* Free the input: it is replaced with the result. */ for (i = 0; i < n; i++) + XFREE (R_arg[i]); + free (R_arg); + + if (trace_flag) { - short *sp = R_arg[i]; - if (sp) - while (*sp >= 0) - *temp_R[*sp++]++ = i; + fputs ("transpose: output\n", stderr); + matrix_print (stderr, new_R, n); } - XFREE (temp_R); - return new_R; } @@ -500,52 +399,43 @@ transpose (short **R_arg, int n) static void build_relations (void) { + short *edge = XCALLOC (short, ngotos + 1); + short *states1 = XCALLOC (short, ritem_longest_rhs () + 1); int i; - int j; - short *rulep; - short *rp; - int nedges; - int done; - int state1; - int stateno; - int symbol1; - short *edge; - short *states; - short **new_includes; includes = XCALLOC (short *, ngotos); - edge = XCALLOC (short, ngotos + 1); - states = XCALLOC (short, maxrhs () + 1); for (i = 0; i < ngotos; i++) { - nedges = 0; - state1 = from_state[i]; - symbol1 = state_table[to_state[i]].accessing_symbol; + int nedges = 0; + token_number_t symbol1 = states[to_state[i]]->accessing_symbol; + short *rulep; for (rulep = derives[symbol1]; *rulep > 0; rulep++) { + int done; int length = 1; - states[0] = state1; - stateno = state1; + item_number_t *rp; + state_t *state = states[from_state[i]]; + states1[0] = state->number; - for (rp = ritem + rule_table[*rulep].rhs; *rp > 0; rp++) + for (rp = rules[*rulep].rhs; *rp >= 0; rp++) { - int symbol2 = *rp; - shifts *sp = state_table[stateno].shift_table; - + shifts *sp = state->shifts; + int j; for (j = 0; j < sp->nshifts; j++) { - stateno = sp->shifts[j]; - if (state_table[stateno].accessing_symbol == symbol2) + state = states[sp->shifts[j]]; + if (state->accessing_symbol + == item_number_as_token_number (*rp)) break; } - states[length++] = stateno; + states1[length++] = state->number; } - if (!state_table[stateno].consistent) - add_lookback_edge (stateno, *rulep, i); + if (!state->consistent) + add_lookback_edge (state, *rulep, i); length--; done = 0; @@ -556,8 +446,9 @@ build_relations (void) /* JF added rp>=ritem && I hope to god its right! */ if (rp >= ritem && ISVAR (*rp)) { - stateno = states[--length]; - edge[nedges++] = map_goto (stateno, *rp); + /* Downcasting from item_number_t to token_number_t. */ + edge[nedges++] = map_goto (states1[--length], + item_number_as_token_number (*rp)); if (nullable[*rp]) done = 0; } @@ -566,6 +457,7 @@ build_relations (void) if (nedges) { + int j; includes[i] = XCALLOC (short, nedges + 1); for (j = 0; j < nedges; j++) includes[i][j] = edge[j]; @@ -573,16 +465,10 @@ build_relations (void) } } - new_includes = transpose (includes, ngotos); - - for (i = 0; i < ngotos; i++) - XFREE (includes[i]); - XFREE (includes); - - includes = new_includes; - XFREE (edge); - XFREE (states); + XFREE (states1); + + includes = transpose (includes, ngotos); } @@ -604,37 +490,98 @@ compute_FOLLOWS (void) static void compute_lookaheads (void) { - int i; + size_t i; shorts *sp; - for (i = 0; i < state_table[nstates].lookaheads; i++) + for (i = 0; i < nLA; i++) for (sp = lookback[i]; sp; sp = sp->next) - { - unsigned *fp1 = LA (i); - unsigned *fp2 = F (sp->value); - while (fp1 < LA (i + 1)) - *fp1++ |= *fp2++; - } + bitset_or (LA[i], LA[i], F[sp->value]); /* Free LOOKBACK. */ - for (i = 0; i < state_table[nstates].lookaheads; i++) + for (i = 0; i < nLA; i++) LIST_FREE (shorts, lookback[i]); XFREE (lookback); - XFREE (F); + bitsetv_free (F); +} + + +/*--------------------------------------. +| Initializing the lookaheads members. | +`--------------------------------------*/ + +static void +initialize_lookaheads (void) +{ + size_t i; + nLA = 0; + for (i = 0; i < nstates; i++) + { + int k; + int nlookaheads = 0; + reductions *rp = states[i]->reductions; + shifts *sp = states[i]->shifts; + + /* We need a lookahead either to distinguish different + reductions (i.e., there are two or more), or to distinguish a + reduction from a shift. Otherwise, it is straightforward, + and the state is `consistent'. */ + if (rp->nreds > 1 + || (rp->nreds == 1 && sp->nshifts && SHIFT_IS_SHIFT (sp, 0))) + nlookaheads += rp->nreds; + else + states[i]->consistent = 1; + + for (k = 0; k < sp->nshifts; k++) + if (SHIFT_IS_ERROR (sp, k)) + { + states[i]->consistent = 0; + break; + } + + states[i]->nlookaheads = nlookaheads; + states[i]->lookaheadsp = nLA; + nLA += nlookaheads; + } } +/*---------------------------------------. +| Output the lookaheads for each state. | +`---------------------------------------*/ + +static void +lookaheads_print (FILE *out) +{ + size_t i; + int j, k; + fprintf (out, "Lookaheads: BEGIN\n"); + for (i = 0; i < nstates; ++i) + { + fprintf (out, "State %d: %d lookaheads\n", + i, states[i]->nlookaheads); + + for (j = 0; j < states[i]->nlookaheads; ++j) + for (k = 0; k < ntokens; ++k) + if (bitset_test (LA[states[i]->lookaheadsp + j], k)) + fprintf (out, " on %d (%s) -> rule %d\n", + k, quotearg_style (escape_quoting_style, symbols[k]->tag), + LArule[states[i]->lookaheadsp + j]->number - 1); + } + fprintf (out, "Lookaheads: END\n"); +} + void lalr (void) { - tokensetsize = WORDSIZE (ntokens); - - set_state_table (); + initialize_lookaheads (); initialize_LA (); set_goto_map (); initialize_F (); build_relations (); compute_FOLLOWS (); compute_lookaheads (); + + if (trace_flag) + lookaheads_print (stderr); }