blob: 23a879cacd5ae64fc3d378636ac1a94906133bf8 [file] [log] [blame]
/* Output the generated parsing program for Bison.
Copyright (C) 1984, 1986, 1989, 1992, 2000-2006, 2009-2015, 2018-2021
Free Software Foundation, Inc.
This file is part of Bison, the GNU Compiler Compiler.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>. */
#include <config.h>
#include "system.h"
#include <bitset.h>
#include <bitsetv.h>
#include "complain.h"
#include "conflicts.h"
#include "files.h"
#include "getargs.h"
#include "gram.h"
#include "lalr.h"
#include "muscle-tab.h"
#include "reader.h"
#include "symtab.h"
#include "tables.h"
/* Several tables are indexed both by state and nonterminal numbers.
We call such an index a 'vector'; i.e., a vector is either a state
or a nonterminal number.
Of course vector_number_t ought to be wide enough to contain
state_number and symbol_number. */
typedef int vector_number;
#if 0 /* Not currently used. */
static inline vector_number
state_number_to_vector_number (state_number s)
{
return s;
}
#endif
static inline vector_number
symbol_number_to_vector_number (symbol_number sym)
{
return state_number_as_int (nstates) + sym - ntokens;
}
int nvectors;
/* FROMS and TOS are indexed by vector_number.
If VECTOR is a nonterminal, (FROMS[VECTOR], TOS[VECTOR]) form an
array of state numbers of the non defaulted GOTO on VECTOR.
If VECTOR is a state, TOS[VECTOR] is the array of actions to do on
the (array of) symbols FROMS[VECTOR].
In both cases, TALLY[VECTOR] is the size of the arrays
FROMS[VECTOR], TOS[VECTOR]; and WIDTH[VECTOR] =
(FROMS[VECTOR][SIZE] - FROMS[VECTOR][0] + 1) where SIZE =
TALLY[VECTOR].
FROMS therefore contains symbol_number and action_number,
TOS state_number and action_number,
TALLY sizes,
WIDTH differences of FROMS.
Let base_number be the type of FROMS, TOS, and WIDTH. */
#define BASE_MAXIMUM INT_MAX
#define BASE_MINIMUM INT_MIN
static base_number **froms;
static base_number **tos;
static int **conflict_tos;
static size_t *tally;
static base_number *width;
/* For a given state, N = ACTROW[SYMBOL]:
If N = 0, stands for 'run the default action'.
If N = MIN, stands for 'raise a syntax error'.
If N > 0, stands for 'shift SYMBOL and go to n'.
If N < 0, stands for 'reduce -N'. */
typedef int action_number;
#define ACTION_NUMBER_MINIMUM INT_MIN
static action_number *actrow;
/* FROMS and TOS are reordered to be compressed. ORDER[VECTOR] is the
new vector number of VECTOR. We skip 'empty' vectors (i.e.,
TALLY[VECTOR] = 0), and call these 'entries'. */
static vector_number *order;
static int nentries;
base_number *base = NULL;
/* A distinguished value of BASE, negative infinite. During the
computation equals to BASE_MINIMUM, later mapped to BASE_NINF to
keep parser tables small. */
base_number base_ninf = 0;
/* Bitset representing an integer set in the range
POS_SET_OFFSET..(POS_SET_OFFSET + SIZE). POS_SET_OFFSET is
nonpositive. */
static bitset pos_set = NULL;
/* The integer denoted by bitno 0 in pos_set. */
static int pos_set_base = 0;
static int *conflrow;
int *conflict_table;
int *conflict_list;
int conflict_list_cnt;
static int conflict_list_free;
/* TABLE_SIZE is the allocated size of both TABLE and CHECK. We start
with more or less the original hard-coded value (which was
SHRT_MAX). */
static int table_size = 32768;
base_number *table;
base_number *check;
/* The value used in TABLE to denote explicit syntax errors
(%nonassoc), a negative infinite. First defaults to ACTION_NUMBER_MINIMUM,
but in order to keep small tables, renumbered as TABLE_ERROR, which
is the smallest (non error) value minus 1. */
base_number table_ninf = 0;
static int lowzero;
int high;
state_number *yydefgoto;
rule_number *yydefact;
/*----------.
| pos_set. |
`----------*/
#if 0
static void
pos_set_dump (void)
{
fprintf (stderr, "pos_set (%ld, %d) =", bitset_size (pos_set), pos_set_base);
bitset_iterator biter;
int i;
BITSET_FOR_EACH (biter, pos_set, i, 0)
fprintf (stderr, " %d", i + pos_set_base);
putc ('\n', stderr);
}
#endif
/* The size and base of POS_SET are not known, we need to be able to
move the base farther "on the left", and grow "on the right".
It would be nice to be able to predict the base accurately, but it
seems difficult (-nstates seems to work most of the time, except
when there are useless tokens).
FIXME: The current approach is correct, but with poor performances.
Bitsets need to support 'assign' and 'shift'. And instead of
extending POS_SET just for the out-of-range new values, we need
something like doubling the size.
*/
static void
pos_set_set (int pos)
{
int bitno = pos - pos_set_base;
if (bitno < 0)
{
// Need more room on the left.
// DELTA is positive. Run 'pos_set >> delta'.
const int delta = pos_set_base - pos;
const int old_size = bitset_size (pos_set);
const int new_size = old_size + delta;
bitset_resize (pos_set, new_size);
// Right-shift all the bits by DELTA. Be sure to reset the new
// bits on the left.
//
// FIXME: add bitset_assign, and bitset_shift?
for (int i = new_size - 1; 0 <= i ; --i)
if (delta <= i && bitset_test (pos_set, i - delta))
bitset_set (pos_set, i);
else
bitset_reset (pos_set, i);
pos_set_base = pos;
bitno = 0;
}
else if (bitset_size (pos_set) <= bitno)
// Need more room on the right.
bitset_resize (pos_set, bitno + 1);
bitset_set (pos_set, bitno);
}
static bool
pos_set_test (int pos)
{
const int bitno = pos - pos_set_base;
return bitset_test (pos_set, bitno);
}
/*-------------------------------------------------------------------.
| If TABLE, CONFLICT_TABLE, and CHECK are too small to be addressed |
| at DESIRED, grow them. TABLE[DESIRED] can be used, so the desired |
| size is at least DESIRED + 1. |
`-------------------------------------------------------------------*/
static void
table_grow (int desired)
{
int old_size = table_size;
while (table_size <= desired)
table_size *= 2;
if (trace_flag & trace_resource)
fprintf (stderr, "growing tables from %d to %d\n",
old_size, table_size);
table = xnrealloc (table, table_size, sizeof *table);
memset (table + old_size, 0,
sizeof *table * (table_size - old_size));
conflict_table = xnrealloc (conflict_table, table_size,
sizeof *conflict_table);
memset (conflict_table + old_size, 0,
sizeof *conflict_table * (table_size - old_size));
check = xnrealloc (check, table_size, sizeof *check);
for (int i = old_size; i < table_size; ++i)
check[i] = -1;
}
/*-------------------------------------------------------------------.
| For GLR parsers, for each conflicted token in S, as indicated |
| by non-zero entries in CONFLROW, create a list of possible |
| reductions that are alternatives to the shift or reduction |
| currently recorded for that token in S. Store the alternative |
| reductions followed by a 0 in CONFLICT_LIST, updating |
| CONFLICT_LIST_CNT, and storing an index to the start of the list |
| back into CONFLROW. |
`-------------------------------------------------------------------*/
static void
conflict_row (state *s)
{
if (!nondeterministic_parser)
return;
const reductions *reds = s->reductions;
for (state_number j = 0; j < ntokens; j += 1)
if (conflrow[j])
{
conflrow[j] = conflict_list_cnt;
/* Find all reductions for token J, and record all that do not
match ACTROW[J]. */
for (int i = 0; i < reds->num; i += 1)
if (bitset_test (reds->lookaheads[i], j)
&& (actrow[j]
!= rule_number_as_item_number (reds->rules[i]->number)))
{
aver (0 < conflict_list_free);
conflict_list[conflict_list_cnt] = reds->rules[i]->number + 1;
conflict_list_cnt += 1;
conflict_list_free -= 1;
}
/* Leave a 0 at the end. */
aver (0 < conflict_list_free);
conflict_list[conflict_list_cnt] = 0;
conflict_list_cnt += 1;
conflict_list_free -= 1;
}
}
/*------------------------------------------------------------------.
| Decide what to do for each type of token if seen as the |
| lookahead in specified state. The value returned is used as the |
| default action (yydefact) for the state. In addition, ACTROW is |
| filled with what to do for each kind of token, index by symbol |
| number, with zero meaning do the default action. The value |
| ACTION_NUMBER_MINIMUM, a very negative number, means this |
| situation is an error. The parser recognizes this value |
| specially. |
| |
| This is where conflicts are resolved. The loop over lookahead |
| rules considered lower-numbered rules last, and the last rule |
| considered that likes a token gets to handle it. |
| |
| For GLR parsers, also sets CONFLROW[SYM] to an index into |
| CONFLICT_LIST iff there is an unresolved conflict (s/r or r/r) |
| with symbol SYM. The default reduction is not used for a symbol |
| that has any such conflicts. |
`------------------------------------------------------------------*/
static rule *
action_row (state *s)
{
for (state_number i = 0; i < ntokens; i++)
actrow[i] = conflrow[i] = 0;
reductions *reds = s->reductions;
bool conflicted = false;
if (reds->lookaheads)
/* loop over all the rules available here which require
lookahead (in reverse order to give precedence to the first
rule) */
for (int i = reds->num - 1; 0 <= i; --i)
/* and find each token which the rule finds acceptable
to come next */
{
bitset_iterator biter;
int j;
BITSET_FOR_EACH (biter, reds->lookaheads[i], j, 0)
{
/* and record this rule as the rule to use if that
token follows. */
if (actrow[j] != 0)
{
conflicted = true;
conflrow[j] = 1;
}
actrow[j] = rule_number_as_item_number (reds->rules[i]->number);
}
}
/* Now see which tokens are allowed for shifts in this state. For
them, record the shift as the thing to do. So shift is preferred
to reduce. */
transitions *trans = s->transitions;
/* Set to nonzero to inhibit having any default reduction. */
bool nodefault = false;
{
int i;
FOR_EACH_SHIFT (trans, i)
{
symbol_number sym = TRANSITION_SYMBOL (trans, i);
state *shift_state = trans->states[i];
if (actrow[sym] != 0)
{
conflicted = true;
conflrow[sym] = 1;
}
actrow[sym] = state_number_as_int (shift_state->number);
/* Do not use any default reduction if there is a shift for
error */
if (sym == errtoken->content->number)
nodefault = true;
}
}
/* See which tokens are an explicit error in this state (due to
%nonassoc). For them, record ACTION_NUMBER_MINIMUM as the
action. */
errs *errp = s->errs;
for (int i = 0; i < errp->num; i++)
{
symbol *sym = errp->symbols[i];
actrow[sym->content->number] = ACTION_NUMBER_MINIMUM;
}
/* Turn off default reductions where requested by the user. See
state_lookaheads_count in lalr.c to understand when states are
labeled as consistent. */
{
char *default_reductions =
muscle_percent_define_get ("lr.default-reduction");
if (STRNEQ (default_reductions, "most") && !s->consistent)
nodefault = true;
free (default_reductions);
}
/* Now find the most common reduction and make it the default action
for this state. */
rule *default_reduction = NULL;
if (reds->num >= 1 && !nodefault)
{
if (s->consistent)
default_reduction = reds->rules[0];
else
{
int max = 0;
for (int i = 0; i < reds->num; i++)
{
int count = 0;
rule *r = reds->rules[i];
for (symbol_number j = 0; j < ntokens; j++)
if (actrow[j] == rule_number_as_item_number (r->number))
count++;
if (count > max)
{
max = count;
default_reduction = r;
}
}
/* GLR parsers need space for conflict lists, so we can't
default conflicted entries. For non-conflicted entries
or as long as we are not building a GLR parser,
actions that match the default are replaced with zero,
which means "use the default". */
if (0 < max)
for (symbol_number j = 0; j < ntokens; j++)
if (actrow[j]
== rule_number_as_item_number (default_reduction->number)
&& ! (nondeterministic_parser && conflrow[j]))
actrow[j] = 0;
}
}
/* If have no default reduction, the default is an error.
So replace any action which says "error" with "use default". */
if (!default_reduction)
for (symbol_number i = 0; i < ntokens; i++)
if (actrow[i] == ACTION_NUMBER_MINIMUM)
actrow[i] = 0;
if (conflicted)
conflict_row (s);
return default_reduction;
}
/*----------------------------------------.
| Set FROMS, TOS, TALLY and WIDTH for S. |
`----------------------------------------*/
static void
save_row (state_number s)
{
/* Number of non default actions in S. */
size_t count = 0;
for (symbol_number i = 0; i < ntokens; i++)
if (actrow[i] != 0)
count++;
if (count)
{
/* Allocate non defaulted actions. */
base_number *sp1 = froms[s] = xnmalloc (count, sizeof *sp1);
base_number *sp2 = tos[s] = xnmalloc (count, sizeof *sp2);
int *sp3 = conflict_tos[s] =
nondeterministic_parser ? xnmalloc (count, sizeof *sp3) : NULL;
/* Store non defaulted actions. */
for (symbol_number i = 0; i < ntokens; i++)
if (actrow[i] != 0)
{
*sp1++ = i;
*sp2++ = actrow[i];
if (nondeterministic_parser)
*sp3++ = conflrow[i];
}
tally[s] = count;
width[s] = sp1[-1] - froms[s][0] + 1;
}
}
/*------------------------------------------------------------------.
| Figure out the actions for the specified state, indexed by |
| lookahead token kind. |
| |
| The YYDEFACT table is output now. The detailed info is saved for |
| putting into YYTABLE later. |
`------------------------------------------------------------------*/
static void
token_actions (void)
{
int nconflict = nondeterministic_parser ? conflicts_total_count () : 0;
yydefact = xnmalloc (nstates, sizeof *yydefact);
actrow = xnmalloc (ntokens, sizeof *actrow);
conflrow = xnmalloc (ntokens, sizeof *conflrow);
conflict_list = xnmalloc (1 + 2 * nconflict, sizeof *conflict_list);
conflict_list_free = 2 * nconflict;
conflict_list_cnt = 1;
/* Find the rules which are reduced. */
if (!nondeterministic_parser)
for (rule_number r = 0; r < nrules; ++r)
rules[r].useful = false;
for (state_number i = 0; i < nstates; ++i)
{
rule *default_reduction = action_row (states[i]);
yydefact[i] = default_reduction ? default_reduction->number + 1 : 0;
save_row (i);
/* Now that the parser was computed, we can find which rules are
really reduced, and which are not because of SR or RR
conflicts. */
if (!nondeterministic_parser)
{
for (symbol_number j = 0; j < ntokens; ++j)
if (actrow[j] < 0 && actrow[j] != ACTION_NUMBER_MINIMUM)
rules[item_number_as_rule_number (actrow[j])].useful = true;
if (yydefact[i])
rules[yydefact[i] - 1].useful = true;
}
}
free (actrow);
free (conflrow);
}
/*------------------------------------------------------------------.
| Compute FROMS[VECTOR], TOS[VECTOR], TALLY[VECTOR], WIDTH[VECTOR], |
| i.e., the information related to non defaulted GOTO on the nterm |
| SYM. |
| |
| DEFAULT_STATE is the principal destination on SYM, i.e., the |
| default GOTO destination on SYM. |
`------------------------------------------------------------------*/
static void
save_column (symbol_number sym, state_number default_state)
{
const goto_number begin = goto_map[sym - ntokens];
const goto_number end = goto_map[sym - ntokens + 1];
/* Number of non default GOTO. */
size_t count = 0;
for (goto_number i = begin; i < end; i++)
if (to_state[i] != default_state)
count++;
if (count)
{
/* Allocate room for non defaulted gotos. */
vector_number symno = symbol_number_to_vector_number (sym);
base_number *sp1 = froms[symno] = xnmalloc (count, sizeof *sp1);
base_number *sp2 = tos[symno] = xnmalloc (count, sizeof *sp2);
/* Store the state numbers of the non defaulted gotos. */
for (goto_number i = begin; i < end; i++)
if (to_state[i] != default_state)
{
*sp1++ = from_state[i];
*sp2++ = to_state[i];
}
tally[symno] = count;
width[symno] = sp1[-1] - froms[symno][0] + 1;
}
}
/*----------------------------------------------------------------.
| The default state for SYM: the state which is 'the' most common |
| GOTO destination on SYM (an nterm). |
`----------------------------------------------------------------*/
static state_number
default_goto (symbol_number sym, size_t state_count[])
{
const goto_number begin = goto_map[sym - ntokens];
const goto_number end = goto_map[sym - ntokens + 1];
/* In the case this symbol is never reduced to ($accept), use state
0. We used to use -1, but as a result the yydefgoto table must
be signed, which (1) might trigger compiler warnings when storing
a value from yydefgoto into a state number (nonnegative), and (2)
wastes bits which might result in using a int16 where a uint8
suffices. */
state_number res = 0;
if (begin != end)
{
for (state_number s = 0; s < nstates; s++)
state_count[s] = 0;
for (goto_number i = begin; i < end; i++)
state_count[to_state[i]]++;
size_t max = 0;
for (state_number s = 0; s < nstates; s++)
if (max < state_count[s])
{
max = state_count[s];
res = s;
}
}
return res;
}
/*-------------------------------------------------------------------.
| Figure out what to do after reducing with each rule, depending on |
| the saved state from before the beginning of parsing the data that |
| matched this rule. |
| |
| The YYDEFGOTO table is output now. The detailed info is saved for |
| putting into YYTABLE later. |
`-------------------------------------------------------------------*/
static void
goto_actions (void)
{
size_t *state_count = xnmalloc (nstates, sizeof *state_count);
yydefgoto = xnmalloc (nnterms, sizeof *yydefgoto);
/* For a given nterm I, STATE_COUNT[S] is the number of times there
is a GOTO to S on I. */
for (symbol_number i = ntokens; i < nsyms; ++i)
{
state_number default_state = default_goto (i, state_count);
save_column (i, default_state);
yydefgoto[i - ntokens] = default_state;
}
free (state_count);
}
/*------------------------------------------------------------------.
| Compute ORDER, a reordering of vectors, in order to decide how to |
| pack the actions and gotos information into yytable. |
`------------------------------------------------------------------*/
static void
sort_actions (void)
{
nentries = 0;
for (int i = 0; i < nvectors; i++)
if (0 < tally[i])
{
const size_t t = tally[i];
const int w = width[i];
int j = nentries - 1;
while (0 <= j && width[order[j]] < w)
j--;
while (0 <= j && width[order[j]] == w && tally[order[j]] < t)
j--;
for (int k = nentries - 1; k > j; k--)
order[k + 1] = order[k];
order[j + 1] = i;
nentries++;
}
}
/* If VECTOR is a state whose actions (reflected by FROMS, TOS, TALLY
and WIDTH of VECTOR) are common to a previous state, return this
state number.
In any other case, return -1. */
static state_number
matching_state (vector_number vector)
{
vector_number i = order[vector];
/* If VECTOR is a nterm, return -1. */
if (i < nstates)
{
size_t t = tally[i];
int w = width[i];
/* If VECTOR has GLR conflicts, return -1 */
if (conflict_tos[i] != NULL)
for (int j = 0; j < t; j += 1)
if (conflict_tos[i][j] != 0)
return -1;
for (int prev = vector - 1; 0 <= prev; prev--)
{
vector_number j = order[prev];
/* Given how ORDER was computed, if the WIDTH or TALLY is
different, there cannot be a matching state. */
if (width[j] != w || tally[j] != t)
return -1;
else
{
bool match = true;
for (int k = 0; match && k < t; k++)
if (tos[j][k] != tos[i][k]
|| froms[j][k] != froms[i][k]
|| (conflict_tos[j] != NULL && conflict_tos[j][k] != 0))
match = false;
if (match)
return j;
}
}
}
return -1;
}
static base_number
pack_vector (vector_number vector)
{
vector_number i = order[vector];
size_t t = tally[i];
base_number *from = froms[i];
base_number *to = tos[i];
int *conflict_to = conflict_tos[i];
aver (t != 0);
for (base_number res = lowzero - from[0]; ; res++)
{
bool ok = true;
aver (res < table_size);
{
for (int k = 0; ok && k < t; k++)
{
int loc = res + state_number_as_int (from[k]);
if (table_size <= loc)
table_grow (loc);
if (table[loc] != 0)
ok = false;
}
if (ok && pos_set_test (res))
ok = false;
}
if (ok)
{
int loc PACIFY_CC (= -1);
for (int k = 0; k < t; k++)
{
loc = res + state_number_as_int (from[k]);
table[loc] = to[k];
if (nondeterministic_parser && conflict_to != NULL)
conflict_table[loc] = conflict_to[k];
check[loc] = from[k];
}
while (table[lowzero] != 0)
lowzero++;
if (high < loc)
high = loc;
aver (BASE_MINIMUM <= res && res <= BASE_MAXIMUM);
return res;
}
}
}
/*-------------------------------------------------------------.
| Remap the negative infinite in TAB from NINF to the greatest |
| possible smallest value. Return it. |
| |
| In most case this allows us to use shorts instead of ints in |
| parsers. |
`-------------------------------------------------------------*/
static base_number
table_ninf_remap (base_number tab[], int size, base_number ninf)
{
base_number res = 0;
for (int i = 0; i < size; i++)
if (tab[i] < res && tab[i] != ninf)
res = tab[i];
--res;
for (int i = 0; i < size; i++)
if (tab[i] == ninf)
tab[i] = res;
return res;
}
static void
pack_table (void)
{
base = xnmalloc (nvectors, sizeof *base);
pos_set = bitset_create (table_size + nstates, BITSET_FRUGAL);
pos_set_base = -nstates;
table = xcalloc (table_size, sizeof *table);
conflict_table = xcalloc (table_size, sizeof *conflict_table);
check = xnmalloc (table_size, sizeof *check);
lowzero = 0;
high = 0;
for (int i = 0; i < nvectors; i++)
base[i] = BASE_MINIMUM;
for (int i = 0; i < table_size; i++)
check[i] = -1;
for (int i = 0; i < nentries; i++)
{
state_number s = matching_state (i);
base_number place;
if (s < 0)
/* A new set of state actions, or a nonterminal. */
place = pack_vector (i);
else
/* Action of I were already coded for S. */
place = base[s];
pos_set_set (place);
base[order[i]] = place;
}
/* Use the greatest possible negative infinites. */
base_ninf = table_ninf_remap (base, nvectors, BASE_MINIMUM);
table_ninf = table_ninf_remap (table, high + 1, ACTION_NUMBER_MINIMUM);
bitset_free (pos_set);
}
/*-----------------------------------------------------------------.
| Compute and output yydefact, yydefgoto, yypact, yypgoto, yytable |
| and yycheck. |
`-----------------------------------------------------------------*/
void
tables_generate (void)
{
/* This is a poor way to make sure the sizes are properly
correlated. In particular the signedness is not taken into
account. But it's not useless. */
verify (sizeof nstates <= sizeof nvectors);
verify (sizeof nnterms <= sizeof nvectors);
nvectors = state_number_as_int (nstates) + nnterms;
froms = xcalloc (nvectors, sizeof *froms);
tos = xcalloc (nvectors, sizeof *tos);
conflict_tos = xcalloc (nvectors, sizeof *conflict_tos);
tally = xcalloc (nvectors, sizeof *tally);
width = xnmalloc (nvectors, sizeof *width);
token_actions ();
goto_actions ();
free (goto_map);
free (from_state);
free (to_state);
order = xcalloc (nvectors, sizeof *order);
sort_actions ();
pack_table ();
free (order);
free (tally);
free (width);
for (int i = 0; i < nvectors; i++)
{
free (froms[i]);
free (tos[i]);
free (conflict_tos[i]);
}
free (froms);
free (tos);
free (conflict_tos);
}
/*-------------------------.
| Free the parser tables. |
`-------------------------*/
void
tables_free (void)
{
free (base);
free (conflict_table);
free (conflict_list);
free (table);
free (check);
free (yydefgoto);
free (yydefact);
}