when walking through ritem, even via rule->rhs.
* src/reduce.c (dump_grammar, useful_production, reduce_output)
(useful_production, useless_nonterminals): Likewise.
(reduce_grammar_tables): Likewise, plus update nritems.
* src/nullable.c (set_nullable): Likewise.
* src/lalr.c (build_relations): Likewise.
* tests/sets.at (Nullable): Adjust.
Fortunately, now, the $axiom is no longer nullable.
otherwise, since we range over their symbols, we might look at a
nonterminal which no longer ``exists'', i.e., it is not counted in
`nvars', hence we overflow our arrays.
RHS's properly.
* tests/torture.at (GNU AWK Grammar): New.
* tests/sets.at (Nullable): New.
* src/nullable.c (set_nullable): Instead of blindly looping over
`ritems', loop over the rules, and then over their rhs's.
Work around Autotest bugs.
* src/warshall.c (bitmatrix_print): Don't use `+--+' as table
frame, because Autotest understand lines starting with a `+' as
traces from the shell. Then, they are not processed properly.
Admittedly an Autotest bug, but we don't have time to wait for
Autotest to catch up.
* tests/regression.at (Broken Closure): Adjust to the new table
frames.
Move to...
* tests/sets.at: here.
`destroyed'), we now sometimes examine them, and store information
related to them. Hence we need to know their number, and adjust
memory allocations.
* src/reduce.c, src/reduce.h (nuseless_nonterminals): No longer
static.
* src/LR0.c (allocate_itemsets): The memory allocated to
`symbol_count' was used for two different purpose: once to count
the number of occurrences of each symbol, and later reassigned to
`shift_symbol', containing the symbol that can be shifted from a
given state.
Deobfuscate, i.e., allocate, use and free `symbol_count' here
only, and...
(new_itemsets): Allocate `shift_symbol' here.
(allocate_itemsets): symbol_count includes useless nonterminals.
Make room for them.
(free_storage): Use `free', not `XFREE', for pointers that cannot
be null.
Bison's `allocate.c' and `alloc.h'. This patch was surprisingly
difficult and introduced a lot of core dump. It turns out that
Bison used an implementation of `xmalloc' based on `calloc', and
at various places it does depend upon the initialization to 0. I
have not tried to isolate the pertinent places, and all the former
calls to Bison's `xmalloc' are now using `XCALLOC'. Someday,
someone should address this issue.
* src/allocate.c, src/alloc.h, m4/bison-decl.m4: Remove.
* lib/xmalloc.c, lib/xalloc.h, m4/malloc.m4, m4/realloc.m4: New
files.
Adjust dependencies.
* src/warshall.h: New file.
Propagate.
* src/machine.h: Remove, leaving its contents in...
* src/system.h: here.
Include stdio.h.
Adjust all dependencies on stdio.h and machine.h.
* src/getargs.h: New file.
Let all `extern' declarations about getargs.c be replaced with
inclusion of `getargs.h'.
* src/Makefile.am (noinst_HEADERS): Adjust.
* tests/calc.m4 (yyin): Be initialized in main, not on the global
scope.
(yyerror): Returns void, not int.
* doc/bison.texinfo: Formatting changes.
