Future/bison
1
0
Fork 0
mirror of https://git.savannah.gnu.org/git/bison.git synced 2026-03-11 13:23:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

Let nondeterministic skeletons be usable with deterministic

Browse Source 
tables.
With the patch, GAWK compiled by GCC without -O2 passes its test
suite using a GLR parser driven by LALR tables.  It fails with -O2
because `struct stat' gives two different answers on my machine:
88 (definition of an auto var) and later 96 (memset on this var).
Hence the stack is badly corrumpted.  The headers inclusion is to
blame: if I move the awk.h inclusion before GLR's system header
inclusion, the two struct stat have the same size.
* src/tables.c (pack_table): Always create conflict_table.
(token_actions): Always create conflict_list.
* data/glr.c (YYFLAG): Remove, unused.
This commit is contained in:
Akim Demaille
2002-10-13 18:50:40 +00:00
parent f377f69fec
commit ea99527d23
4 changed files with 44 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
src/output.c
View File

@@ -503,19 +503,19 @@ prepare_actions (void)
muscle_insert_base_table ("check", check,
check[0], 1, high + 1);
if (glr_parser)
{
/* GLR parsing slightly modifies yytable and yycheck
(and thus yypact) so that in states with unresolved conflicts,
the default reduction is not used in the conflicted entries, so
that there is a place to put a conflict pointer. This means that
yyconflp and yyconfl are nonsense for a non-GLR parser, so we
avoid accidents by not writing them out in that case. */
muscle_insert_unsigned_int_table ("conflict_list_heads", conflict_table,
conflict_table[0], 1, high+1);
muscle_insert_unsigned_int_table ("conflicting_rules", conflict_list,
conflict_list[0], 1, conflict_list_cnt);
}
/* GLR parsing slightly modifies YYTABLE and YYCHECK (and thus
YYPACT) so that in states with unresolved conflicts, the default
reduction is not used in the conflicted entries, so that there is
a place to put a conflict pointer.
This means that YYCONFLP and YYCONFL are nonsense for a non-GLR
parser, so we could avoid accidents by not writing them out in
that case. Nevertheless, it seems even better to be able to use
the GLR skeletons even without the non-deterministic tables. */
muscle_insert_unsigned_int_table ("conflict_list_heads", conflict_table,
conflict_table[0], 1, high+1);
muscle_insert_unsigned_int_table ("conflicting_rules", conflict_list,
conflict_list[0], 1, conflict_list_cnt);
}