Reported by Akim Demaille at
<http://lists.gnu.org/archive/html/bug-bison/2007-12/msg00001.html>
and discussed again starting at
<http://lists.gnu.org/archive/html/bison-patches/2008-11/msg00102.html>.
* data/bison.m4 (b4_percent_define_default): Leave syncline blank
because location information is bogus.
Use angle brackets to delimit fake file name because square brackets
look like over-quoted m4. Choose a better fake file name.
Use negative line numbers.
* src/muscle_tab.c (muscle_percent_define_default): Likewise.
* src/location.c (location_print): If line for a boundary is negative,
only print that boundary's file name.
* src/location.h: Document that.
* tests/skeletons.at (%define Boolean variables: invalid skeleton
defaults): Update output.
skeletons, and use this for %code qualifier complaints.
* data/bison.m4 (b4_error_at): New, invoked by...
(b4_warn_at, b4_complain_at, b4_fatal_at): ... these new macros to wrap
the skeleton scanner's new @warn_at(...@), @complain_at(...@), and
@fatal_at(...@) directives.
(b4_check_percent_code_qualifiers): Rewrite to expect locations for
qualifiers in b4_used_percent_code_qualifiers and to use
b4_complain_at.
* src/location.c, src/location.h (boundary_set_from_string): New global
function.
* src/muscle_tab.c, src/muscle_tab.h (muscle_boundary_grow): New global
function.
* src/parse-gram.y (grammar_declaration): Add locations for qualifiers
to b4_used_percent_code_qualifiers.
* src/scan-skel.l (fail_for_at_directive_too_few_args): New static
function.
(AT_DIRECTIVE_ARGC_MAX): Increase for boundary arguments.
(lineno): Rename to...
(out_lineno): ... this so I don't misunderstand it again.
(SC_AT_DIRECTIVE_SKIP_WS): Don't increment out_lineno for newlines
here; these newlines are in the input but not the output file.
(SC_AT_DIRECTIVE_ARG): Likewise. Extract directive execution to...
(at_directive_perform): ... this new static function, and add handling
of new @warn_at(...@), @complain_at(...@), and @fatal_at(...@)
directives.
* tests/input.at (Reject bad %code qualifiers): Update test output with
locations and extend.
* tests/output.at (Output file name: [, Output file name: ]): Remove
bogus comment about these tests failing.
destructors, printers, and actions into a code_props structure and
interface. This patch merely implements code_props in scan-code.h and
scan-code.l. Future patches will rewrite other modules to use it.
Discussed starting at
<http://lists.gnu.org/archive/html/bison-patches/2006-11/msg00020.html>.
* src/location.h (EMPTY_LOCATION_INIT): Define so that it's easier to
consistently initialize const structs that have an empty location
field.
* src/location.c (empty_location): Initialize with EMPTY_LOCATION_INIT
to ensure consistency.
* src/scan-code.h (code_props): New structure.
(code_props_none_init, CODE_PROPS_NONE_INIT, code_props_none): New
function, macro, and const global variable for initializing a
code_props with no code.
(code_props_plain_init, code_props_symbol_action_init,
code_props_rule_action_init, code_props_translate_code): The rest of
the new code_props functional interface. Among other things, the init
functions set the code_props kind field so that
code_props_translate_code will know whether to behave like
translate_symbol_action, translate_rule_action, or translate_code.
These old translate functions must remain until all other modules are
updated to use the new code_props interface.
(code_scanner_last_string_free): New function similar to
gram_scanner_last_string_free.
(code_scanner_free): Add documentation.
* src/scan-code.l: Implement the new interface.
(code_lex): Make it static, add a code_props* argument, and remove the
rule argument.
(last_string): New static global similar to the one in scan-gram.l.
(SC_RULE_ACTION): Update to use the code_props* argument to code_lex
instead of rule.
(SC_SYMBOL_ACTION): For $$, set the is_value_used member of the
code_props since Bison may one day use this information for destructors
and printers.
(<*><<EOF>>): Use STRING_FINISH so that last_string is set.
(handle_action_dollar): Use symbol_list_n_get and set used flag
directly since symbol_list_n_used_set is removed.
(translate_action): Add a code_props* argument and remove the rule,
action, and location arguments. Pass the code_props* on to code_lex.
(translate_rule_action, translate_symbol_action, translate_code):
Rewrite as wrappers around the new code_props interface.
* src/symlist.h, src/symlist.c (symbol_list_n_used_set): Remove since
it would eventually need to break the encapsulation of code_props.
destructors, printers, and actions into a code_props structure and
interface.
* src/location.h (EMPTY_LOCATION_INIT): Define so that it's easier to
consistently initialize const structs that have an empty location field.
* src/location.c (empty_location): Initialize with EMPTY_LOCATION_INIT
to ensure consistency.
* src/output.c (symbol_destructors_output, symbol_printers_output):
Replace with...
(symbol_code_props_output): ... this to eliminate duplicate code.
(output_skeleton): Update to use symbol_code_props_output.
* src/parse-gram.y (prologue_declaration, braceless, epilogue.opt):
Update all uses of translate_* functions to use the new code_props
interface and to use gram_scanner_last_string_free and
code_scanner_last_string_free where possible.
(grammar_declaration): symbol_list_destructor_set and
symbol_list_printer_set now perform the translation, so don't do it
here. Use gram_scanner_last_string_free where possible.
* src/reader.c: Update to use code_props interface for destructors and
rule actions.
* src/scan-code.h (code_props): New structure.
(code_props_none_init, CODE_PROPS_NONE_INIT, code_props_none): New
function, macro, and const global variable for initializing a
code_props with no code.
(code_props_plain_init, code_props_symbol_action_init,
code_props_rule_action_init, code_props_translate_code,
code_props_code_get, code_props_location_get,
code_props_is_value_used): The rest of the new code_props interface.
(translate_rule_action, translate_symbol_action, translate_code):
Remove as these are now just special cases within
code_props_translate_code, which is switched on the code_props kind.
(code_scanner_last_string_free): New function similar to
gram_scanner_last_string_free.
* src/scan-code.l: Implement the new interface.
(last_string): New static global similar to the one in scan-gram.l.
(SC_SYMBOL_ACTION): For $$, set the is_value_used member of the
code_props since Bison will one day use this information for
destructors and printers.
(<*><<EOF>>): Use STRING_FINISH so that last_string is set.
(handle_action_dollar): Update to use the code_props interface of rule
actions. Use symbol_list_n_get and set is_value_used directly since
symbol_list_n_used_set is removed.
* src/symlist.h, src/symlist.c (symbol_list): Replace action,
action_location, and used members with a code_props member, and update
all uses.
(symbol_list_n_used_set): Remove since it would need to break the
encapsulation of code_props.
(symbol_list_destructor_set, symbol_list_printer_set): Perform code
translation here rather than depending on the caller to do so.
* src/symtab.h (symbol, semantic_type): Remove destructor_location and
printer_location members and change the type of the destructor and
printer members to code_props.
(symbol_destructor_location_get, symbol_printer_location_get): Remove
unneeded.
(symbol_destructor_set, symbol_destructor_get, symbol_printer_set,
symbol_printer_get, semantic_type_destructor_set,
semantic_type_printer_set, default_tagged_destructor_set,
default_tagless_destructor_set, default_tagged_printer_set,
default_tagless_printer_set): Use code_props in arguments and return
types in place of const char * and location.
* src/symtab.c: Update implementation for interface and struct changes.
(default_tagged_destructor_location,
default_tagless_destructor_location, default_tagged_printer_location,
default_tagless_printer_location): Remove since we...
(default_tagged_destructor, default_tagless_destructor,
default_tagged_printer, default_tagless_printer): ... change the type
of these to code_props.
(SYMBOL_CODE_PRINT): New similar to SYMBOL_ATTR_PRINT but for
code_props members.
(symbol_print): Use SYMBOL_CODE_PRINT.
* src/scan-gram.h (gram_last_string): Remove declaration.
* src/scan-gram.l (last_string): Declare it static.
As a consequence, the relation between the grammar scanner and
parser is much simpler. We can also split "composite tokens" back
into simple tokens.
* src/gram.h (ITEM_NUMBER_MAX, RULE_NUMBER_MAX): New.
* src/scan-gram.l (add_column_width, adjust_location): Move to and
rename as...
* src/location.h, src/location.c (add_column_width)
(location_compute): these.
Fix the column count: the initial column is 0.
(location_print): Be robust to ending column being 0.
* src/location.h (boundary_set): New.
* src/main.c: Adjust to scanner_free being renamed as
gram_scanner_free.
* src/output.c: Include scan-code.h.
* src/parse-gram.y: Include scan-gram.h and scan-code.h.
Use boundary_set.
(PERCENT_DESTRUCTOR, PERCENT_PRINTER, PERCENT_INITIAL_ACTION)
(PERCENT_LEX_PARAM, PERCENT_PARSE_PARAM): Remove the {...} part,
which is now, again, a separate token.
Adjust all dependencies.
Whereever actions with $ and @ are used, use translate_code.
(action): Remove this nonterminal which is now useless.
* src/reader.c: Include assert.h, scan-gram.h and scan-code.h.
(grammar_current_rule_action_append): Use translate_code.
(packgram): Bound check ruleno, itemno, and rule_length.
* src/reader.h (gram_in, gram__flex_debug, scanner_cursor)
(last_string, last_braced_code_loc, max_left_semantic_context)
(scanner_initialize, scanner_free, scanner_last_string_free)
(gram_out, gram_lineno, YY_DECL_): Move to...
* src/scan-gram.h: this new file.
(YY_DECL): Rename as...
(GRAM_DECL): this.
* src/scan-code.h, src/scan-code.l, src/scan-code-c.c: New.
* src/scan-gram.l (gram_get_lineno, gram_get_in, gram_get_out):
(gram_get_leng, gram_get_text, gram_set_lineno, gram_set_in):
(gram_set_out, gram_get_debug, gram_set_debug, gram_lex_destroy):
Move these declarations, and...
(obstack_for_string, STRING_GROW, STRING_FINISH, STRING_FREE):
these to...
* src/flex-scanner.h: this new file.
* src/scan-gram.l (rule_length, rule_length_overflow)
(increment_rule_length): Remove.
(last_braced_code_loc): Rename as...
(gram_last_braced_code_loc): this.
Adjust to the changes of the parser.
Move all the handling of $ and @ into...
* src/scan-code.l: here.
* src/scan-gram.l (handle_dollar, handle_at): Remove.
(handle_action_dollar, handle_action_at): Move to...
* src/scan-code.l: here.
* src/Makefile.am (bison_SOURCES): Add flex-scanner.h,
scan-code.h, scan-code-c.c, scan-gram.h.
(EXTRA_bison_SOURCES): Add scan-code.l.
(BUILT_SOURCES): Add scan-code.c.
(yacc): Be robust to white spaces.
* tests/conflicts.at, tests/input.at, tests/reduce.at,
* tests/regression.at: Adjust the column numbers.
* tests/regression.at: Adjust the error message.
* src/assoc.h (assoc_to_string): Give a name to the arg, as
this is the usual Bison style.
* src/location.h (location_print): Likewise.
* src/reader.h (token_name): Likewise.
to INT_MAX indicates an overflow.
* src/scan-gram.l: Include verify.h. Don't include get-errno.h.
(rule_length_overflow, increment_rule_length, add_column_width):
New functions.
(<INITIAL>{id}, <SC_ESCAPED_STRING>"\"", <SC_ESCAPED_CHARACTER>"'"):
(<SC_BRACED_CODE>"}"):
Use increment_rule_length rather than incrementing it by hand.
(adjust_location, handle_syncline): Diagnose overflow.
(handle_action_dollar, handle_action_at):
Fix bug with monstrosities like $-2147483648.
Remove now-unnecessary checks.
(scan_integer): Verify assumptions and remove now-unnecessary checks.
(convert_ucn_to_byte): Verify assumptions.
(handle_syncline): New arg LOC. All callers changed.
Don't store through a value derived from char const * pointer.
(boundary): New type.
(location_t): Use it. This allows locations to span file boundaries.
All member uses changed: file -> start.file or end.file (as needed),
first_line -> start.line, first_column -> start.column,
last_line -> end.line, last_column -> end.column.
(equal_boundaries): New function.
(LOCATION_RESET, LOCATION_STEP): Remove.
(LOCATION_PRINT): Remove. All callers changed to use location_print.
(empty_location): Now const.
(location_print): New decl.
cannot be NULL, not XFREE.
(pack_vector): Use assert, not fatal, for bound violations.
* src/state.c (state_new): Likewise.
* src/reader.c (reader): Likewise.
* src/lalr.c (set_goto_map): Likewise.
* src/location.h (LOCATION_PRINT): If first_line is 0, just issue
the file name.
clashes.
* src/scan-gram.l: Use ['] instead of ['] to pacify
font-lock-mode.
Use complain_at.
Use quote, not quote_n since LOCATION_PRINT no longer uses the
slot 0.
(input): Don't try to initialize yylloc here, do it in the
scanner.
* src/scan-gram.l (YY_USER_INIT): Initialize yylloc.
* src/gram.h (rule_t): Change line and action_line into location
and action_location, of location_t type.
Adjust all dependencies.
* src/location.h, src/location.c (empty_location): New.
* src/reader.h, src/reader.c (grammar_start_symbol_set)
(grammar_symbol_append, grammar_rule_begin, grammar_rule_end)
(grammar_current_rule_symbol_append)
(grammar_current_rule_action_append): Expect a location as argument.
* src/reader.c (grammar_midrule_action): Adjust to attach an
action's location as dummy symbol location.
* src/symtab.h, src/symtab.c (startsymbol_location): New.
* tests/regression.at (Web2c Report, Rule Line Numbers): Adjust
the line numbers.
