X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/32e1e0a4865d64946a5497aff057d7be28c1cc3d..e62f1a899168bdf1658dd6d934dea407491a80ff:/src/lalr.c diff --git a/src/lalr.c b/src/lalr.c index 84f32894..e69c05c7 100644 --- a/src/lalr.c +++ b/src/lalr.c @@ -1,5 +1,6 @@ -/* Compute look-ahead criteria for bison, - Copyright (C) 1984, 1986, 1989, 2000, 2001, 2002 +/* Compute look-ahead criteria for Bison. + + Copyright (C) 1984, 1986, 1989, 2000, 2001, 2002, 2003, 2004 Free Software Foundation, Inc. This file is part of Bison, the GNU Compiler Compiler. @@ -21,192 +22,116 @@ /* Compute how to make the finite state machine deterministic; find - which rules need lookahead in each state, and which lookahead + which rules need look-ahead in each state, and which look-ahead 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 +#include +#include + #include "LR0.h" #include "complain.h" -#include "lalr.h" -#include "nullable.h" #include "derives.h" #include "getargs.h" +#include "gram.h" +#include "lalr.h" +#include "nullable.h" +#include "reader.h" +#include "relation.h" +#include "symtab.h" -rule_t **LArule = NULL; -bitsetv LA = NULL; -size_t nLA; - -static int ngotos; -short *goto_map = NULL; -state_number_t *from_state = NULL; -state_number_t *to_state = NULL; - -/* And for the famous F variable, which name is so descriptive that a - comment is hardly needed. . */ -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; - int height; - - VERTICES[++top] = i; - INDEX[i] = height = top; - - if (R[i]) - for (j = 0; R[i][j] >= 0; ++j) - { - if (INDEX[R[i][j]] == 0) - traverse (R[i][j]); - - if (INDEX[i] > INDEX[R[i][j]]) - INDEX[i] = INDEX[R[i][j]]; - - bitset_or (F[i], F[i], F[R[i][j]]); - } - - if (INDEX[i] == height) - for (;;) - { - j = VERTICES[top--]; - INDEX[j] = infinity; - - if (i == j) - break; - - bitset_copy (F[j], F[i]); - } -} - +goto_number *goto_map; +static goto_number ngotos; +state_number *from_state; +state_number *to_state; -static void -digraph (short **relation) +/* Linked list of goto numbers. */ +typedef struct goto_list { - int i; + struct goto_list *next; + goto_number value; +} goto_list; - infinity = ngotos + 2; - INDEX = XCALLOC (short, ngotos + 1); - VERTICES = XCALLOC (short, ngotos + 1); - top = 0; - R = relation; +/* LA is a LR by NTOKENS matrix of bits. LA[l, i] is 1 if the rule + LArule[l] is applicable in the appropriate state when the next + token is symbol i. If LA[l, i] and LA[l, j] are both 1 for i != j, + it is a conflict. */ - for (i = 0; i < ngotos; i++) - INDEX[i] = 0; - - for (i = 0; i < ngotos; i++) - if (INDEX[i] == 0 && R[i]) - traverse (i); - - XFREE (INDEX); - XFREE (VERTICES); -} +static bitsetv LA = NULL; +size_t nLA; -static void -initialize_LA (void) -{ - state_number_t i; - int j; - rule_t **np; +/* And for the famous F variable, which name is so descriptive that a + comment is hardly needed. . */ +static bitsetv F = NULL; - /* Avoid having to special case 0. */ - if (!nLA) - nLA = 1; +static goto_number **includes; +static goto_list **lookback; - LA = bitsetv_create (nLA, ntokens, BITSET_FIXED); - LArule = XCALLOC (rule_t *, nLA); - lookback = XCALLOC (shorts *, nLA); - np = LArule; - for (i = 0; i < nstates; i++) - 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) { - state_number_t state; - short *temp_map; + state_number s; + goto_number *temp_map; - goto_map = XCALLOC (short, nvars + 1) - ntokens; - temp_map = XCALLOC (short, nvars + 1) - ntokens; + goto_map = xcalloc (nvars + 1, sizeof *goto_map); + temp_map = xnmalloc (nvars + 1, sizeof *temp_map); ngotos = 0; - for (state = 0; state < nstates; ++state) + for (s = 0; s < nstates; ++s) { - shifts_t *sp = states[state]->shifts; + transitions *sp = states[s]->transitions; int i; - for (i = sp->nshifts - 1; i >= 0 && SHIFT_IS_GOTO (sp, i); --i) + for (i = sp->num - 1; i >= 0 && TRANSITION_IS_GOTO (sp, i); --i) { - if (ngotos == SHRT_MAX) - fatal (_("too many gotos (max %d)"), SHRT_MAX); - ngotos++; - goto_map[SHIFT_SYMBOL (sp, i)]++; + + /* Abort if (ngotos + 1) would overflow. */ + if (ngotos == GOTO_NUMBER_MAXIMUM) + abort (); + + goto_map[TRANSITION_SYMBOL (sp, i) - ntokens]++; } } { - int k = 0; + goto_number k = 0; int i; for (i = ntokens; i < nsyms; i++) { - temp_map[i] = k; - k += goto_map[i]; + temp_map[i - ntokens] = k; + k += goto_map[i - ntokens]; } for (i = ntokens; i < nsyms; i++) - goto_map[i] = temp_map[i]; + goto_map[i - ntokens] = temp_map[i - ntokens]; - goto_map[nsyms] = ngotos; - temp_map[nsyms] = ngotos; + goto_map[nsyms - ntokens] = ngotos; + temp_map[nsyms - ntokens] = ngotos; } - from_state = XCALLOC (state_number_t, ngotos); - to_state = XCALLOC (state_number_t, ngotos); + from_state = xcalloc (ngotos, sizeof *from_state); + to_state = xcalloc (ngotos, sizeof *to_state); - for (state = 0; state < nstates; ++state) + for (s = 0; s < nstates; ++s) { - shifts_t *sp = states[state]->shifts; + transitions *sp = states[s]->transitions; int i; - for (i = sp->nshifts - 1; i >= 0 && SHIFT_IS_GOTO (sp, i); --i) + for (i = sp->num - 1; i >= 0 && TRANSITION_IS_GOTO (sp, i); --i) { - int k = temp_map[SHIFT_SYMBOL (sp, i)]++; - from_state[k] = state; - to_state[k] = sp->shifts[i]; + goto_number k = temp_map[TRANSITION_SYMBOL (sp, i) - ntokens]++; + from_state[k] = s; + to_state[k] = sp->states[i]->number; } } - XFREE (temp_map + ntokens); + free (temp_map); } @@ -215,259 +140,160 @@ set_goto_map (void) | Map a state/symbol pair into its numeric representation. | `----------------------------------------------------------*/ -static int -map_goto (state_number_t state, symbol_number_t symbol) +static goto_number +map_goto (state_number s0, symbol_number sym) { - int high; - int low; - int middle; - state_number_t s; + goto_number high; + goto_number low; + goto_number middle; + state_number s; - low = goto_map[symbol]; - high = goto_map[symbol + 1] - 1; + low = goto_map[sym - ntokens]; + high = goto_map[sym - ntokens + 1] - 1; - while (low <= high) + for (;;) { + if (high < low) + abort (); middle = (low + high) / 2; s = from_state[middle]; - if (s == state) + if (s == s0) return middle; - else if (s < state) + else if (s < s0) low = middle + 1; else high = middle - 1; } - - assert (0); - /* NOTREACHED */ - return 0; } static void initialize_F (void) { - short **reads = XCALLOC (short *, ngotos); - short *edge = XCALLOC (short, ngotos + 1); - int nedges = 0; + goto_number **reads = xnmalloc (ngotos, sizeof *reads); + goto_number *edge = xnmalloc (ngotos + 1, sizeof *edge); + goto_number nedges = 0; - int i; + goto_number i; F = bitsetv_create (ngotos, ntokens, BITSET_FIXED); for (i = 0; i < ngotos; i++) { - state_number_t stateno = to_state[i]; - shifts_t *sp = states[stateno]->shifts; + state_number stateno = to_state[i]; + transitions *sp = states[stateno]->transitions; int j; - for (j = 0; j < sp->nshifts && SHIFT_IS_SHIFT (sp, j); j++) - bitset_set (F[i], SHIFT_SYMBOL (sp, j)); + FOR_EACH_SHIFT (sp, j) + bitset_set (F[i], TRANSITION_SYMBOL (sp, j)); - for (; j < sp->nshifts; j++) + for (; j < sp->num; j++) { - symbol_number_t symbol = SHIFT_SYMBOL (sp, j); - if (nullable[symbol]) - edge[nedges++] = map_goto (stateno, symbol); + symbol_number sym = TRANSITION_SYMBOL (sp, j); + if (nullable[sym - ntokens]) + edge[nedges++] = map_goto (stateno, sym); } - if (nedges) + if (nedges == 0) + reads[i] = NULL; + else { - reads[i] = XCALLOC (short, nedges + 1); - memcpy (reads[i], edge, nedges * sizeof (edge[0])); - reads[i][nedges] = -1; + reads[i] = xnmalloc (nedges + 1, sizeof reads[i][0]); + memcpy (reads[i], edge, nedges * sizeof edge[0]); + reads[i][nedges] = END_NODE; nedges = 0; } } - digraph (reads); + relation_digraph (reads, ngotos, &F); for (i = 0; i < ngotos; i++) - XFREE (reads[i]); + free (reads[i]); - XFREE (reads); - XFREE (edge); + free (reads); + free (edge); } static void -add_lookback_edge (state_t *state, int ruleno, int gotono) +add_lookback_edge (state *s, rule *r, goto_number gotono) { - int i; - shorts *sp; - - for (i = 0; i < state->nlookaheads; ++i) - if (state->lookaheads_rule[i]->number == ruleno) - break; - - assert (state->lookaheads_rule[i]->number == ruleno); - - sp = XCALLOC (shorts, 1); - sp->next = lookback[(state->lookaheads - LA) + i]; + int ri = state_reduction_find (s, r); + goto_list *sp = xmalloc (sizeof *sp); + sp->next = lookback[(s->reductions->look_ahead_tokens - LA) + ri]; sp->value = gotono; - lookback[(state->lookaheads - LA) + i] = sp; -} - - -static void -matrix_print (FILE *out, short **matrix, int n) -{ - 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); + lookback[(s->reductions->look_ahead_tokens - LA) + ri] = sp; } -/*-------------------------------------------------------------------. -| 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]. | -`-------------------------------------------------------------------*/ - -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) - { - fputs ("transpose: input\n", stderr); - matrix_print (stderr, R_arg, 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); - sp[nedges[i]] = -1; - new_R[i] = sp; - end_R[i] = sp; - } - - /* 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) - { - fputs ("transpose: output\n", stderr); - matrix_print (stderr, new_R, n); - } - - return new_R; -} static void build_relations (void) { - short *edge = XCALLOC (short, ngotos + 1); - state_number_t *states1 = XCALLOC (state_number_t, ritem_longest_rhs () + 1); - int i; + goto_number *edge = xnmalloc (ngotos + 1, sizeof *edge); + state_number *states1 = xnmalloc (ritem_longest_rhs () + 1, sizeof *states1); + goto_number i; - includes = XCALLOC (short *, ngotos); + includes = xnmalloc (ngotos, sizeof *includes); for (i = 0; i < ngotos; i++) { int nedges = 0; - symbol_number_t symbol1 = states[to_state[i]]->accessing_symbol; - short *rulep; + symbol_number symbol1 = states[to_state[i]]->accessing_symbol; + rule **rulep; - for (rulep = derives[symbol1]; *rulep > 0; rulep++) + for (rulep = derives[symbol1 - ntokens]; *rulep; rulep++) { - int done; + bool done; int length = 1; - item_number_t *rp; - state_t *state = states[from_state[i]]; - states1[0] = state->number; + item_number *rp; + state *s = states[from_state[i]]; + states1[0] = s->number; - for (rp = rules[*rulep].rhs; *rp >= 0; rp++) + for (rp = (*rulep)->rhs; *rp >= 0; rp++) { - shifts_t *sp = state->shifts; - int j; - for (j = 0; j < sp->nshifts; j++) - { - state = states[sp->shifts[j]]; - if (state->accessing_symbol - == item_number_as_symbol_number (*rp)) - break; - } - - states1[length++] = state->number; + s = transitions_to (s->transitions, + item_number_as_symbol_number (*rp)); + states1[length++] = s->number; } - if (!state->consistent) - add_lookback_edge (state, *rulep, i); + if (!s->consistent) + add_lookback_edge (s, *rulep, i); length--; - done = 0; + done = false; while (!done) { - done = 1; + done = true; rp--; /* JF added rp>=ritem && I hope to god its right! */ if (rp >= ritem && ISVAR (*rp)) { - /* Downcasting from item_number_t to symbol_number_t. */ + /* Downcasting from item_number to symbol_number. */ edge[nedges++] = map_goto (states1[--length], item_number_as_symbol_number (*rp)); - if (nullable[*rp]) - done = 0; + if (nullable[*rp - ntokens]) + done = false; } } } - if (nedges) + if (nedges == 0) + includes[i] = NULL; + else { int j; - includes[i] = XCALLOC (short, nedges + 1); + includes[i] = xnmalloc (nedges + 1, sizeof includes[i][0]); for (j = 0; j < nedges; j++) includes[i][j] = edge[j]; - includes[i][nedges] = -1; + includes[i][nedges] = END_NODE; } } - XFREE (edge); - XFREE (states1); + free (edge); + free (states1); - includes = transpose (includes, ngotos); + relation_transpose (&includes, ngotos); } @@ -475,22 +301,22 @@ build_relations (void) static void compute_FOLLOWS (void) { - int i; + goto_number i; - digraph (includes); + relation_digraph (includes, ngotos, &F); for (i = 0; i < ngotos; i++) - XFREE (includes[i]); + free (includes[i]); - XFREE (includes); + free (includes); } static void -compute_lookaheads (void) +compute_look_ahead_tokens (void) { size_t i; - shorts *sp; + goto_list *sp; for (i = 0; i < nLA; i++) for (sp = lookback[i]; sp; sp = sp->next) @@ -498,115 +324,139 @@ compute_lookaheads (void) /* Free LOOKBACK. */ for (i = 0; i < nLA; i++) - LIST_FREE (shorts, lookback[i]); + LIST_FREE (goto_list, lookback[i]); - XFREE (lookback); + free (lookback); bitsetv_free (F); } -/*-------------------------------------------------------------. -| Count the number of lookaheads required for each state | -| (NLOOKAHEADS member). Compute the total number of LA, NLA. | -`-------------------------------------------------------------*/ +/*-----------------------------------------------------. +| Count the number of look-ahead tokens required for S | +| (N_LOOK_AHEAD_TOKENS member). | +`-----------------------------------------------------*/ -static void -states_lookaheads_count (void) +static int +state_look_ahead_tokens_count (state *s) { - state_number_t i; - nLA = 0; - - /* Count */ - for (i = 0; i < nstates; i++) - { - int k; - int nlookaheads = 0; - reductions *rp = states[i]->reductions; - shifts_t *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; - } + int k; + int n_look_ahead_tokens = 0; + reductions *rp = s->reductions; + transitions *sp = s->transitions; + + /* We need a look-ahead 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->num > 1 + || (rp->num == 1 && sp->num && + !TRANSITION_IS_DISABLED (sp, 0) && TRANSITION_IS_SHIFT (sp, 0))) + n_look_ahead_tokens += rp->num; + else + s->consistent = 1; + + for (k = 0; k < sp->num; k++) + if (!TRANSITION_IS_DISABLED (sp, k) && TRANSITION_IS_ERROR (sp, k)) + { + s->consistent = 0; + break; + } - states[i]->nlookaheads = nlookaheads; - nLA += nlookaheads; - } + return n_look_ahead_tokens; } -/*--------------------------------------. -| Initializing the lookaheads members. | -`--------------------------------------*/ +/*-----------------------------------------------------. +| Compute LA, NLA, and the look_ahead_tokens members. | +`-----------------------------------------------------*/ static void -states_lookaheads_initialize (void) +initialize_LA (void) { - state_number_t i; - bitsetv pLA = LA; - rule_t **pLArule = LArule; + state_number i; + bitsetv pLA; - /* Initialize the members LOOKAHEADS and LOOKAHEADS_RULE for each - state. */ + /* Compute the total number of reductions requiring a look-ahead. */ + nLA = 0; + for (i = 0; i < nstates; i++) + nLA += state_look_ahead_tokens_count (states[i]); + /* Avoid having to special case 0. */ + if (!nLA) + nLA = 1; + + pLA = LA = bitsetv_create (nLA, ntokens, BITSET_FIXED); + lookback = xcalloc (nLA, sizeof *lookback); + + /* Initialize the members LOOK_AHEAD_TOKENS for each state whose reductions + require look-ahead tokens. */ for (i = 0; i < nstates; i++) { - states[i]->lookaheads = pLA; - states[i]->lookaheads_rule = pLArule; - pLA += states[i]->nlookaheads; - pLArule += states[i]->nlookaheads; + int count = state_look_ahead_tokens_count (states[i]); + if (count) + { + states[i]->reductions->look_ahead_tokens = pLA; + pLA += count; + } } } -/*---------------------------------------. -| Output the lookaheads for each state. | -`---------------------------------------*/ +/*----------------------------------------------. +| Output the look-ahead tokens for each state. | +`----------------------------------------------*/ static void -lookaheads_print (FILE *out) +look_ahead_tokens_print (FILE *out) { - state_number_t i; + state_number i; int j, k; - fprintf (out, "Lookaheads: BEGIN\n"); + fprintf (out, "Look-ahead tokens: BEGIN\n"); for (i = 0; i < nstates; ++i) { - fprintf (out, "State %d: %d lookaheads\n", - i, states[i]->nlookaheads); + reductions *reds = states[i]->reductions; + bitset_iterator iter; + int n_look_ahead_tokens = 0; + + if (reds->look_ahead_tokens) + for (k = 0; k < reds->num; ++k) + if (reds->look_ahead_tokens[k]) + ++n_look_ahead_tokens; - for (j = 0; j < states[i]->nlookaheads; ++j) - for (k = 0; k < ntokens; ++k) - if (bitset_test (states[i]->lookaheads[j], k)) + fprintf (out, "State %d: %d look-ahead tokens\n", + i, n_look_ahead_tokens); + + if (reds->look_ahead_tokens) + for (j = 0; j < reds->num; ++j) + BITSET_FOR_EACH (iter, reds->look_ahead_tokens[j], k, 0) + { fprintf (out, " on %d (%s) -> rule %d\n", - k, symbol_tag_get (symbols[k]), - states[i]->lookaheads_rule[j]->number - 1); + k, symbols[k]->tag, + reds->rules[j]->number); + }; } - fprintf (out, "Lookaheads: END\n"); + fprintf (out, "Look-ahead tokens: END\n"); } void lalr (void) { - states_lookaheads_count (); initialize_LA (); - states_lookaheads_initialize (); set_goto_map (); initialize_F (); build_relations (); compute_FOLLOWS (); - compute_lookaheads (); + compute_look_ahead_tokens (); + + if (trace_flag & trace_sets) + look_ahead_tokens_print (stderr); +} - if (trace_flag) - lookaheads_print (stderr); + +void +lalr_free (void) +{ + state_number s; + for (s = 0; s < nstates; ++s) + states[s]->reductions->look_ahead_tokens = NULL; + bitsetv_free (LA); }