X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/340ef489223a5d1f9c40674ea2973435cb8ecc1b..796d61fbdd7bd5db9947451a204c76b5d046b308:/src/LR0.c diff --git a/src/LR0.c b/src/LR0.c index 57f6459d..5e8caef0 100644 --- a/src/LR0.c +++ b/src/LR0.c @@ -1,5 +1,5 @@ /* Generate the nondeterministic finite state machine for bison, - Copyright (C) 1984, 1986, 1989, 2000 Free Software Foundation, Inc. + Copyright 1984, 1986, 1989, 2000, 2001 Free Software Foundation, Inc. This file is part of Bison, the GNU Compiler Compiler. @@ -23,64 +23,57 @@ The entry point is generate_states. */ #include "system.h" -#include "alloc.h" +#include "getargs.h" +#include "reader.h" #include "gram.h" #include "state.h" #include "complain.h" #include "closure.h" #include "LR0.h" - +#include "lalr.h" +#include "reduce.h" int nstates; int final_state; -core *first_state; -shifts *first_shift; -reductions *first_reduction; +static state_t *first_state = NULL; -static core *this_state; -static core *last_state; -static shifts *last_shift; -static reductions *last_reduction; +static state_t *this_state = NULL; +static state_t *last_state = NULL; static int nshifts; -static short *shift_symbol; +static short *shift_symbol = NULL; -static short *redset; -static short *shiftset; +static short *redset = NULL; +static short *shiftset = NULL; -static short **kernel_base; -static short **kernel_end; -static short *kernel_items; +static short **kernel_base = NULL; +static int *kernel_size = NULL; +static short *kernel_items = NULL; /* hash table for states, to recognize equivalent ones. */ -#define STATE_TABLE_SIZE 1009 -static core **state_table; +#define STATE_HASH_SIZE 1009 +static state_t **state_hash = NULL; static void allocate_itemsets (void) { - short *itemp; - int symbol; int i; - int count; - short *symbol_count; - count = 0; - symbol_count = NEW2 (nsyms, short); + /* Count the number of occurrences of all the symbols in RITEMS. + Note that useless productions (hence useless nonterminals) are + browsed too, hence we need to allocate room for _all_ the + symbols. */ + int count = 0; + short *symbol_count = XCALLOC (short, nsyms + nuseless_nonterminals); - itemp = ritem; - symbol = *itemp++; - while (symbol) - { - if (symbol > 0) - { - count++; - symbol_count[symbol]++; - } - symbol = *itemp++; - } + for (i = 0; ritem[i]; ++i) + if (ritem[i] > 0) + { + count++; + symbol_count[ritem[i]]++; + } /* See comments before new_itemsets. All the vectors of items live inside KERNEL_ITEMS. The number of active items after @@ -88,8 +81,9 @@ allocate_itemsets (void) appears as an item, which is symbol_count[symbol]. We allocate that much space for each symbol. */ - kernel_base = NEW2 (nsyms, short *); - kernel_items = NEW2 (count, short); + kernel_base = XCALLOC (short *, nsyms); + if (count) + kernel_items = XCALLOC (short, count); count = 0; for (i = 0; i < nsyms; i++) @@ -98,8 +92,8 @@ allocate_itemsets (void) count += symbol_count[i]; } - shift_symbol = symbol_count; - kernel_end = NEW2 (nsyms, short *); + free (symbol_count); + kernel_size = XCALLOC (int, nsyms); } @@ -108,22 +102,22 @@ allocate_storage (void) { allocate_itemsets (); - shiftset = NEW2 (nsyms, short); - redset = NEW2 (nrules + 1, short); - state_table = NEW2 (STATE_TABLE_SIZE, core *); + shiftset = XCALLOC (short, nsyms); + redset = XCALLOC (short, nrules + 1); + state_hash = XCALLOC (state_t *, STATE_HASH_SIZE); } static void free_storage (void) { - FREE (shift_symbol); - FREE (redset); - FREE (shiftset); - FREE (kernel_base); - FREE (kernel_end); - FREE (kernel_items); - FREE (state_table); + free (shift_symbol); + free (redset); + free (shiftset); + free (kernel_base); + free (kernel_size); + XFREE (kernel_items); + free (state_hash); } @@ -137,50 +131,39 @@ free_storage (void) | shift_symbol is set to a vector of the symbols that can be | | shifted. For each symbol in the grammar, kernel_base[symbol] | | points to a vector of item numbers activated if that symbol is | -| shifted, and kernel_end[symbol] points after the end of that | -| vector. | +| shifted, and kernel_size[symbol] is their numbers. | `----------------------------------------------------------------*/ static void new_itemsets (void) { int i; - int shiftcount; - short *isp; - short *ksp; - int symbol; -#if TRACE - fprintf (stderr, "Entering new_itemsets\n"); -#endif + if (trace_flag) + fprintf (stderr, "Entering new_itemsets, state = %d\n", + this_state->number); for (i = 0; i < nsyms; i++) - kernel_end[i] = NULL; - - shiftcount = 0; + kernel_size[i] = 0; - isp = itemset; + shift_symbol = XCALLOC (short, nsyms); + nshifts = 0; - while (isp < itemsetend) + for (i = 0; i < nitemset; ++i) { - i = *isp++; - symbol = ritem[i]; + int symbol = ritem[itemset[i]]; if (symbol > 0) { - ksp = kernel_end[symbol]; - - if (!ksp) + if (!kernel_size[symbol]) { - shift_symbol[shiftcount++] = symbol; - ksp = kernel_base[symbol]; + shift_symbol[nshifts] = symbol; + nshifts++; } - *ksp++ = i + 1; - kernel_end[symbol] = ksp; + kernel_base[symbol][kernel_size[symbol]] = itemset[i] + 1; + kernel_size[symbol]++; } } - - nshifts = shiftcount; } @@ -190,39 +173,27 @@ new_itemsets (void) | necessary. | `-----------------------------------------------------------------*/ -static core * +static state_t * new_state (int symbol) { - int n; - core *p; - short *isp1; - short *isp2; - short *iend; + state_t *p; -#if TRACE - fprintf (stderr, "Entering new_state, symbol = %d\n", symbol); -#endif + if (trace_flag) + fprintf (stderr, "Entering new_state, state = %d, symbol = %d (%s)\n", + this_state->number, symbol, tags[symbol]); if (nstates >= MAXSHORT) fatal (_("too many states (max %d)"), MAXSHORT); - isp1 = kernel_base[symbol]; - iend = kernel_end[symbol]; - n = iend - isp1; - - p = - (core *) xmalloc ((unsigned) (sizeof (core) + (n - 1) * sizeof (short))); + p = STATE_ALLOC (kernel_size[symbol]); p->accessing_symbol = symbol; p->number = nstates; - p->nitems = n; + p->nitems = kernel_size[symbol]; - isp2 = p->items; - while (isp1 < iend) - *isp2++ = *isp1++; + shortcpy (p->items, kernel_base[symbol], kernel_size[symbol]); last_state->next = p; last_state = p; - nstates++; return p; @@ -232,54 +203,39 @@ new_state (int symbol) /*--------------------------------------------------------------. | Find the state number for the state we would get to (from the | | current state) by shifting symbol. Create a new state if no | -| equivalent one exists already. Used by append_states. | +| equivalent one exists already. Used by append_states. | `--------------------------------------------------------------*/ static int get_state (int symbol) { int key; - short *isp1; - short *isp2; - short *iend; - core *sp; - int found; - - int n; - -#if TRACE - fprintf (stderr, "Entering get_state, symbol = %d\n", symbol); -#endif + int i; + state_t *sp; - isp1 = kernel_base[symbol]; - iend = kernel_end[symbol]; - n = iend - isp1; + if (trace_flag) + fprintf (stderr, "Entering get_state, state = %d, symbol = %d (%s)\n", + this_state->number, symbol, tags[symbol]); - /* add up the target state's active item numbers to get a hash key */ + /* Add up the target state's active item numbers to get a hash key. + */ key = 0; - while (isp1 < iend) - key += *isp1++; - - key = key % STATE_TABLE_SIZE; - - sp = state_table[key]; + for (i = 0; i < kernel_size[symbol]; ++i) + key += kernel_base[symbol][i]; + key = key % STATE_HASH_SIZE; + sp = state_hash[key]; if (sp) { - found = 0; + int found = 0; while (!found) { - if (sp->nitems == n) + if (sp->nitems == kernel_size[symbol]) { found = 1; - isp1 = kernel_base[symbol]; - isp2 = sp->items; - - while (found && isp1 < iend) - { - if (*isp1++ != *isp2++) - found = 0; - } + for (i = 0; i < kernel_size[symbol]; ++i) + if (kernel_base[symbol][i] != sp->items[i]) + found = 0; } if (!found) @@ -298,9 +254,12 @@ get_state (int symbol) } else /* bucket is empty */ { - state_table[key] = sp = new_state (symbol); + state_hash[key] = sp = new_state (symbol); } + if (trace_flag) + fprintf (stderr, "Exiting get_state => %d\n", sp->number); + return sp->number; } @@ -318,9 +277,9 @@ append_states (void) int j; int symbol; -#if TRACE - fprintf (stderr, "Entering append_states\n"); -#endif + if (trace_flag) + fprintf (stderr, "Entering append_states, state = %d\n", + this_state->number); /* first sort shift_symbol into increasing order */ @@ -337,87 +296,118 @@ append_states (void) } for (i = 0; i < nshifts; i++) - { - symbol = shift_symbol[i]; - shiftset[i] = get_state (symbol); - } + shiftset[i] = get_state (shift_symbol[i]); } static void new_states (void) { - core *p; - - p = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - first_state = last_state = this_state = p; + first_state = last_state = this_state = STATE_ALLOC (0); nstates = 1; } +/*------------------------------------------------------------. +| Save the NSHIFTS of SHIFTSET into the current linked list. | +`------------------------------------------------------------*/ + static void save_shifts (void) { - shifts *p; - short *sp1; - short *sp2; - short *send; - - p = (shifts *) xmalloc ((unsigned) (sizeof (shifts) + - (nshifts - 1) * sizeof (short))); - - p->number = this_state->number; - p->nshifts = nshifts; - - sp1 = shiftset; - sp2 = p->shifts; - send = shiftset + nshifts; - - while (sp1 < send) - *sp2++ = *sp1++; - - if (last_shift) - { - last_shift->next = p; - last_shift = p; - } - else - { - first_shift = p; - last_shift = p; - } + shifts *p = shifts_new (nshifts); + shortcpy (p->shifts, shiftset, nshifts); + this_state->shifts = p; } /*------------------------------------------------------------------. | Subroutine of augment_automaton. Create the next-to-final state, | | to which a shift has already been made in the initial state. | +| | +| The task of this state consists in shifting (actually, it's a | +| goto, but shifts and gotos are both stored in SHIFTS) the start | +| symbols, hence the name. | `------------------------------------------------------------------*/ static void -insert_start_shift (void) +insert_start_shifting_state (void) { - core *statep; + state_t *statep; shifts *sp; - statep = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - statep->number = nstates; + statep = STATE_ALLOC (0); + statep->number = nstates++; + + /* The distinctive feature of this state from the + eof_shifting_state, is that it is labeled as post-start-symbol + shifting. I fail to understand why this state, and the + post-start-start can't be merged into one. But it does fail if + you try. --akim */ statep->accessing_symbol = start_symbol; last_state->next = statep; last_state = statep; /* Make a shift from this state to (what will be) the final state. */ - sp = NEW (shifts); - sp->number = nstates++; - sp->nshifts = 1; + sp = shifts_new (1); + statep->shifts = sp; + sp->shifts[0] = nstates; +} + + +/*-----------------------------------------------------------------. +| Subroutine of augment_automaton. Create the final state, which | +| shifts `0', the end of file. The initial state shifts the start | +| symbol, and goes to here. | +`-----------------------------------------------------------------*/ + +static void +insert_eof_shifting_state (void) +{ + state_t *statep; + shifts *sp; + + /* Make the final state--the one that follows a shift from the + next-to-final state. + The symbol for that shift is 0 (end-of-file). */ + statep = STATE_ALLOC (0); + statep->number = nstates++; + + last_state->next = statep; + last_state = statep; + + /* Make the shift from the final state to the termination state. */ + sp = shifts_new (1); + statep->shifts = sp; sp->shifts[0] = nstates; +} - last_shift->next = sp; - last_shift = sp; + +/*---------------------------------------------------------------. +| Subroutine of augment_automaton. Create the accepting state. | +`---------------------------------------------------------------*/ + +static void +insert_accepting_state (void) +{ + state_t *statep; + + /* Note that the variable `final_state' refers to what we sometimes + call the termination state. */ + final_state = nstates; + + /* Make the termination state. */ + statep = STATE_ALLOC (0); + statep->number = nstates++; + last_state->next = statep; + last_state = statep; } + + + /*------------------------------------------------------------------. | Make sure that the initial state has a shift that accepts the | | grammar's start symbol and goes to the next-to-final state, which | @@ -429,171 +419,83 @@ insert_start_shift (void) static void augment_automaton (void) { - int i; - int k; - core *statep; - shifts *sp; - shifts *sp2; - shifts *sp1 = NULL; + if (!first_state->shifts->nshifts) + { + /* The first state has no shifts. Make one shift, from the + initial state to the next-to-final state. */ - sp = first_shift; + shifts *sp = shifts_new (1); + first_state->shifts = sp; + sp->shifts[0] = nstates; - if (sp) + /* Create the next-to-final state, with shift to + what will be the final state. */ + insert_start_shifting_state (); + } + else { - if (sp->number == 0) + state_t *statep = first_state->next; + /* The states reached by shifts from FIRST_STATE are numbered + 1..(SP->NSHIFTS). Look for one reached by START_SYMBOL. + This is typical of `start: start ... ;': there is a state + with the item `start: start . ...'. We want to add a `shift + on EOF to eof-shifting state here. */ + while (statep->accessing_symbol != start_symbol + && statep->number < first_state->shifts->nshifts) + statep = statep->next; + + if (statep->accessing_symbol == start_symbol) { - k = sp->nshifts; - statep = first_state->next; + /* We already have STATEP, a next-to-final state for `start: + start . ...'. Make sure it has a shift to what will be + the final state. */ + int i; - /* The states reached by shifts from first_state are numbered 1...K. - Look for one reached by start_symbol. */ - while (statep->accessing_symbol < start_symbol - && statep->number < k) - statep = statep->next; + /* Find the shift of the inital state that leads to STATEP. */ + shifts *sp = statep->shifts; - if (statep->accessing_symbol == start_symbol) - { - /* We already have a next-to-final state. - Make sure it has a shift to what will be the final state. */ - k = statep->number; + shifts *sp1 = shifts_new (sp->nshifts + 1); + statep->shifts = sp1; + sp1->shifts[0] = nstates; + for (i = sp->nshifts; i > 0; i--) + sp1->shifts[i] = sp->shifts[i - 1]; - while (sp && sp->number < k) - { - sp1 = sp; - sp = sp->next; - } + XFREE (sp); - if (sp && sp->number == k) - { - sp2 = (shifts *) xmalloc ((unsigned) (sizeof (shifts) - + - sp->nshifts * - sizeof (short))); - sp2->number = k; - sp2->nshifts = sp->nshifts + 1; - sp2->shifts[0] = nstates; - for (i = sp->nshifts; i > 0; i--) - sp2->shifts[i] = sp->shifts[i - 1]; - - /* Patch sp2 into the chain of shifts in place of sp, - following sp1. */ - sp2->next = sp->next; - sp1->next = sp2; - if (sp == last_shift) - last_shift = sp2; - FREE (sp); - } - else - { - sp2 = NEW (shifts); - sp2->number = k; - sp2->nshifts = 1; - sp2->shifts[0] = nstates; - - /* Patch sp2 into the chain of shifts between sp1 and sp. */ - sp2->next = sp; - sp1->next = sp2; - if (sp == 0) - last_shift = sp2; - } - } - else - { - /* There is no next-to-final state as yet. */ - /* Add one more shift in first_shift, - going to the next-to-final state (yet to be made). */ - sp = first_shift; - - sp2 = (shifts *) xmalloc (sizeof (shifts) - + sp->nshifts * sizeof (short)); - sp2->nshifts = sp->nshifts + 1; - - /* Stick this shift into the vector at the proper place. */ - statep = first_state->next; - for (k = 0, i = 0; i < sp->nshifts; k++, i++) - { - if (statep->accessing_symbol > start_symbol && i == k) - sp2->shifts[k++] = nstates; - sp2->shifts[k] = sp->shifts[i]; - statep = statep->next; - } - if (i == k) - sp2->shifts[k++] = nstates; - - /* Patch sp2 into the chain of shifts - in place of sp, at the beginning. */ - sp2->next = sp->next; - first_shift = sp2; - if (last_shift == sp) - last_shift = sp2; - - FREE (sp); - - /* Create the next-to-final state, with shift to - what will be the final state. */ - insert_start_shift (); - } + insert_eof_shifting_state (); } else { - /* The initial state didn't even have any shifts. - Give it one shift, to the next-to-final state. */ - sp = NEW (shifts); - sp->nshifts = 1; - sp->shifts[0] = nstates; - - /* Patch sp into the chain of shifts at the beginning. */ - sp->next = first_shift; - first_shift = sp; - - /* Create the next-to-final state, with shift to - what will be the final state. */ - insert_start_shift (); - } - } - else - { - /* There are no shifts for any state. - Make one shift, from the initial state to the next-to-final state. */ - - sp = NEW (shifts); - sp->nshifts = 1; - sp->shifts[0] = nstates; + /* There is no state for `start: start . ...'. */ + int i, k; + shifts *sp = first_state->shifts; + shifts *sp1 = NULL; + + /* Add one more shift to the initial state, going to the + next-to-final state (yet to be made). */ + sp1 = shifts_new (sp->nshifts + 1); + first_state->shifts = sp1; + /* Stick this shift into the vector at the proper place. */ + statep = first_state->next; + for (k = 0, i = 0; i < sp->nshifts; k++, i++) + { + if (statep->accessing_symbol > start_symbol && i == k) + sp1->shifts[k++] = nstates; + sp1->shifts[k] = sp->shifts[i]; + statep = statep->next; + } + if (i == k) + sp1->shifts[k++] = nstates; - /* Initialize the chain of shifts with sp. */ - first_shift = sp; - last_shift = sp; + XFREE (sp); - /* Create the next-to-final state, with shift to - what will be the final state. */ - insert_start_shift (); + /* Create the next-to-final state, with shift to what will + be the final state. Corresponds to `start: start . ...'. */ + insert_start_shifting_state (); + } } - /* Make the final state--the one that follows a shift from the - next-to-final state. - The symbol for that shift is 0 (end-of-file). */ - statep = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - statep->number = nstates; - last_state->next = statep; - last_state = statep; - - /* Make the shift from the final state to the termination state. */ - sp = NEW (shifts); - sp->number = nstates++; - sp->nshifts = 1; - sp->shifts[0] = nstates; - last_shift->next = sp; - last_shift = sp; - - /* Note that the variable `final_state' refers to what we sometimes call - the termination state. */ - final_state = nstates; - - /* Make the termination state. */ - statep = (core *) xmalloc ((unsigned) (sizeof (core) - sizeof (short))); - statep->number = nstates++; - last_state->next = statep; - last_state = statep; + insert_accepting_state (); } @@ -606,21 +508,15 @@ augment_automaton (void) static void save_reductions (void) { - short *isp; - short *rp1; - short *rp2; - int item; int count; - reductions *p; - - short *rend; + int i; /* Find and count the active items that represent ends of rules. */ count = 0; - for (isp = itemset; isp < itemsetend; isp++) + for (i = 0; i < nitemset; ++i) { - item = ritem[*isp]; + int item = ritem[itemset[i]]; if (item < 0) redset[count++] = -item; } @@ -629,33 +525,43 @@ save_reductions (void) if (count) { - p = (reductions *) xmalloc ((unsigned) (sizeof (reductions) + - (count - 1) * sizeof (short))); - - p->number = this_state->number; + reductions *p = REDUCTIONS_ALLOC (count); p->nreds = count; + shortcpy (p->rules, redset, count); - rp1 = redset; - rp2 = p->rules; - rend = rp1 + count; - - for (/* nothing */; rp1 < rend; ++rp1, ++rp2) - *rp2 = *rp1; - - if (last_reduction) - { - last_reduction->next = p; - last_reduction = p; - } - else - { - first_reduction = p; - last_reduction = p; - } + this_state->reductions = p; } } +/*--------------------. +| 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); + + { + state_t *sp; + for (sp = first_state; sp; sp = sp->next) + state_table[sp->number] = sp; + } + + /* Pessimization, but simplification of the code: make sure all the + states have a shifts, even if reduced to 0 shifts. */ + { + int i; + for (i = 0; i < nstates; i++) + if (!state_table[i]->shifts) + state_table[i]->shifts = shifts_new (0); + } +} + /*-------------------------------------------------------------------. | Compute the nondeterministic finite state machine (see state.h for | | details) from the grammar. | @@ -670,6 +576,9 @@ generate_states (void) while (this_state) { + if (trace_flag) + fprintf (stderr, "Processing state %d (reached by %s)\n", + this_state->number, tags[this_state->accessing_symbol]); /* Set up ruleset and itemset for the transitions out of this state. ruleset gets a 1 bit for each rule that could reduce now. itemset gets a vector of all the items that could be @@ -684,8 +593,7 @@ generate_states (void) /* create the shifts structures for the shifts to those states, now that the state numbers transitioning to are known */ - if (nshifts > 0) - save_shifts (); + save_shifts (); /* states are queued when they are created; process them all */ this_state = this_state->next; @@ -697,4 +605,7 @@ generate_states (void) /* set up initial and final states as parser wants them */ augment_automaton (); + + /* Set up STATE_TABLE. */ + set_state_table (); }