* src/reader.c (readgram): Be sure to attach mid-rule actions to
the empty-rule associated to the dummy symbol, not to the host
rule.
* tests/actions.at (Mid-rule actions): New.
* src/LR0.c (new_itemsets): Don't allocate `shift_symbol' here,
since it allocates it for each state, although only one is needed.
(allocate_storage): Do it here.
and rule_table.
* src/reader.c (packsymbols): Fill this table.
Drop sprec.
* src/conflicts.c (resolve_sr_conflict): Adjust.
* src/reduce.c (reduce_grammar): Adjust: just sort symbols, a
single table.
Use symbols[i]->tag instead of tags[i].
output as expected output. The contents *seem* right to me, but I
can't pretend reading perfectly parser tables... Nonetheless, all
the other tests pass correctly, the table look OK, even though the
presence of `$axiom' is to be noted: AFAICS it is useless (but
harmless).
* src/LR0.c (final_state): Initialize to -1 so that we do compute
the reductions of the first state which was mistakenly confused
with the final state because precisely final_state was initialized
to 0.
* tests/sets.at (Nullable): Adjust: state 0 does have lookaheads,
now noticed by Bison.
* tests/regression.at (Rule Line Numbers): Adjust: state 0 does
have a reduction on $default.
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 $axiom is no longer nullable.
cases for non-GNU systems like AIX, HP-UX, SGI, Sun, and
Sparc, as they were causing more porting problems than the
(minor) performance improvement was worth.
Also, catch up with 1.31's YYSTD.
0-sentinel. See below.
Use -1 as separator, not 0.
* src/bison.simple (yyparse): Subtract 1 to the rule numbers.
Rely on -1 as separator in yyrhs, instead of 0.
* tests/calc.at (AT_CHECK_CALC): Now, the parsers no longer issue
twice `Now at end of input', therefore there are two lines less to
expect.
(Solved SR Conflicts, Rule Line Numbers): Adjust to the changes
below.
* src/LR0.c (new_state): Recognize the final state by the fact it
is reached by eoftoken.
(insert_start_shifting_state, insert_eof_shifting_state)
(insert_accepting_state, augment_automaton): Remove, since now
these states are automatically computed from the initial state.
(generate_states): Adjust.
* src/print.c: When reporting a rule number to the user, substract
1, so that the axiom rule is rule 0, and the first user rule is 1.
* src/reduce.c: Likewise.
* src/print_graph.c (print_core): For the time being, just as for
the report, depend upon --trace-flags to dump the full set of
items.
* src/reader.c (readgram): Once the grammar read, insert the rule
0: `$axiom: START-SYMBOL $'.
* tests/set.at: Adjust: rule 0 is now displayed, and since the
number of the states has changed (the final state is no longer
necessarily the last), catch up.
* tests/regression.at: Partly catch up.
is reached by eoftoken.
(insert_start_shifting_state, insert_eof_shifting_state)
(insert_accepting_state, augment_automaton): Remove, since now
these states are automatically computed from the initial state.
(generate_states): Adjust.
* src/print.c: When reporting a rule number to the user, substract
1, so that the axiom rule is rule 0, and the first user rule is 1.
* src/reduce.c: Likewise.
* src/print_graph.c (print_core): For the time being, just as for
the report, depend upon --trace-flags to dump the full set of
items.
* src/reader.c (readgram): Once the grammar read, insert the rule
0: `$axiom: START-SYMBOL $'.
* tests/set.at: Adjust: rule 0 is now displayed, and since the
number of the states has changed (the final state is no longer
necessarily the last), catch up.
is 0 which was also used as a sentinel in ritem, (i) make sure >= 0
is used instead of > 0 where appropriate, (ii), depend upon nritems
instead of the 0-sentinel.
* src/gram.h, src/gram.c (nritems): New.
Expected to be duplication of nitems, but for the time being...
* src/reader.c (packgram): Assert nritems and nitems are equal.
* src/LR0.c (allocate_itemsets, new_itemsets): Adjust.
* src/closure.c (print_closure, print_fderives): Likewise.
* src/gram.c (ritem_print): Likewise.
* src/print.c (print_core, print_grammar): Likewise.
* src/print_graph.c: Likewise.
reductions, then output the report anyway if requested, then die.
* src/symtab.c (bucket_new): Initialize `value' to -1, not 0.
* src/reader.c (eoftoken): New.
(parse_token_decl): If the token being defined has value `0', it
is the eoftoken.
(packsymbols): No longer hack `tags' to insert `$' by hand.
Be sure to preserve the value of the eoftoken.
(reader): Make sure eoftoken is defined.
Initialize nsyms to 0: now eoftoken is created just like the others.
* src/print.c (print_grammar): Don't special case the eof token.
* src/regression.at: Adjust: `$' has value 0, not -1, which was a
lie anyway, albeit pleasant.
* tests/calc.at: Exercise error messages with eoftoken.
Change the grammar so that empty input is invalid.
Adjust expectations.
When yyungeting, be sure to use a valid yylloc: use last_yylloc.
Replace strchr if needed.
* src/system.h: Provide the protos of strchr, strspn and memchr if
missing.
* lib/strchr.c: New.
* src/reader.c (symbols_save): Use strchr.
Use it to quote the TAGS outputs.
* src/print_graph.c (print_state): Now errors are in red, and
reductions in green.
Prefer high to wide: output the state number on a line of its own.
* src/LR0.c (set_state_table): Let all the states have a
`reductions', even if reduced to 0.
(save_reductions): Adjust.
* src/lalr.c (initialize_LA, initialize_lookaheads): Adjust.
* src/print.c (print_reductions, print_actions): Adjust.
* src/output.c (action_row): Adjust.
* src/LR0.c (set_state_table): Let all the states have an errs,
even if reduced to 0.
* src/print.c (print_errs, print_reductions): Adjust.
* src/output.c (output_actions, action_row): Adjust.
* src/conflicts.c (resolve_sr_conflict): Adjust.
* src/output.c (output): Adjust.
* src/reader.c (parse_braces): New, factoring...
(copy_action, copy_guard): these two which are renamed as...
(parse_action, parse_guard): these.
As a voluntary consequence, using braces around guards is now
mandatory.
* src/reader.c (symbol_list): `guard' and `guard_line' are new
members.
(symbol_list_new): Adjust.
(copy_action): action_line is the first line, not the last.
(copy_guard): Just as for actions, store the `action' only, not
the switch/case/break flesh.
Don't parse the user action that might follow the guard, let...
(readgram): do it, i.e., now, there can be an action after a
guard.
In other words the guard is just explicitly optional.
(packgram): Adjust.
* src/output.c (guards_output): New.
(output_parser): Call it when needed.
(output): Also free the guard and attrs obstacks.
* src/files.c, src/files.h (obstack_save): Remove.
(output_files): Remove.
As a result, if one needs the former `.act' file, using an
appropriate skeleton which requires actions and guards is now
required.
* src/main.c (main): Adjust.
* tests/semantic.at: New.
* tests/regression.at: Use `input.y' as input file name.
Avoid 8+3 problems by requiring input.c when the test needs the
parser.
* src/LR0.c (set_state_table): state_table holds exactly nstates.
* src/lalr.c (nLA): New.
(initialize_LA, compute_lookaheads, initialize_lookaheads): Use it
instead of lookaheadsp from the pseudo state (nstate + 1).
(resolve_sr_conflict, set_conflicts, solve_conflicts)
(count_sr_conflicts, count_rr_conflicts, conflicts_output)
(conflicts_print, print_reductions): Use a state_t instead of an
integer when referring to a state.
As much as possible, depend upon nlookaheads, instead of the
`lookaheadsp' member of the following state (since lookaheads of
successive states are successive, the difference between state n + 1
and n served as the number of lookaheads for state n).
* src/lalr.c (add_lookback_edge): Likewise.
* src/print.c (print_core, print_actions, print_state)
(print_results): Likewise.
* src/print_graph.c (print_core, print_actions, print_state)
(print_graph): Likewise.
* src/conflicts.h: Adjust.
duplicates ntokens.
* src/bison.simple: Formatting/comment changes.
Use YYNTOKENS only, which is documented, but not YYNTBASE, which
is an undocumented synonym.
nlookaheads is a new member.
Adjust all users.
* src/lalr.h (nlookaheads): Remove this orphan declaration.
* src/lalr.c (initialize_lookaheads): Set nlookaheads for each
state.
(format_obstack): this, to avoid any confusion with files.c's
output_obstack.
* src/reader.h (muscle_obstack): Move to...
* src/output.h: here, since it's defined in output.c.
* src/main.c (main): Call print and print_graph conditionally.
* src/print.c (print): Work unconditionally.
* src/print_graph.c (print_graph): Work unconditionally.
* src/conflicts.c (log_resolution): Output only if verbose_flag.
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.
simultaneously, use one to define the `stype' muscle, and use the
value of the latter to fill defines_obstack.
(copy_comment): Remove.
(copy_comment2): Work for a single obstack.
Rename as...
(copy_comment): this.
process, (iii) output: now we can directly output the parser file
without using table_obstack at all.
* src/files.c, src/files.h (table_obstack): Bye bye.
(parser_file_name): New.
* src/files.c (compute_output_file_names): Compute it.
* src/output.c (actions_output, output_parser)
(output_master_parser): To a file instead of an obstack.
actions_obstack.
* src/gram.h (rule_t): action and action_line are new members.
* src/reader.c (symbol_list): Likewise.
(copy_action): Save the actions within the rule.
(packgram): Save them in rule_table.
* src/output.c (actions_output): New.
(output_parser): Use it on `%%actions'.
(output_rule_data): Don't free rule_table.
(output): Do it.
(prepare): Don't save the `action' muscle.
* src/bison.simple: s/%%action/%%actions/.
just decode them, and keep them as char (before, eol was output as
the 2 char string `n' etc.).
* src/output.c (output_rule_data): Use quotearg to output the
token strings.
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.
traces do: shifting the token EOF. Before EOF was treated as a
nonterminal.
* tests/regression.at: Adjust some tests.
* src/print_graph.c (print_core): Complete the set of items via
closure. The next-to-final and final states are still unsatisfying,
but that's to be addressed elsewhere.
No longer output the rule numbers, but do output the state number.
A single loop for the shifts + gotos is enough, but picked a
distinct color for each.
(print_graph): Initialize and finalize closure.
shifts is useless: just fill directly the state's shifts member.
* src/state.h (shifts): Remove the `next' member.
* src/LR0.c (first_state, last_state): Remove.
Adjust the callers.
(augment_automaton): Don't look for the shifts that must be added
a shift on EOF: it is those of the state we looked for! But now,
since shifts are attached, it is no longer needed to looking
merely by its id: its number.
Remove an impossible branch: if there is a state corresponding to
the start symbol being shifted, then there is shift for the start
symbol from the initial state.
* src/LR0.c: here.
* src/lalr.c (lalr): Don't call it...
* src/LR0.c (generate_states): do it.
* src/LR0.h (first_state): Remove, only the table is used.
(insert_start_shifting_state): this.
(insert_eof_shifting_state, insert_accepting_state): New.
(augment_automaton): Adjust.
Better locality of the variables.
When looking if the start_symbol is shifted from the initial
state, using `while (... symbol != start_symbol ...)' sounds
better than `while (... symbol < start_symbol ...)': If fail
to see how the order between symbols could be relevant!
reductions is useless: just fill directly the state's reductions
member.
* src/state.h (struct reductions): Remove member `number' and
`next'.
* src/LR0.c (first_reduction, last_reduction): Remove.
(save_reductions): Don't link the new reductions, store them in
this_state.
* src/lalr.c (set_state_table): No need to attach reductions to
states, it's already done.
* src/output.c (output_actions): No longer free the shifts, then
the reductions, then the states: free all the states and their
members.
Akim Demaille