<http://lists.gnu.org/archive/html/bug-bison/2005-05/msg00042.html>.
* data/yacc.c (yyabortlab): Don't call destructor, and
don't set yychar to EMPTY.
(yyoverflowlab): Don't call destructor.
(yyreturn): Call destructor, if yychar is neither YYEOF nor YYEMPTY.
* tests/calc.at (AT_CHECK_CALC): Expect one fewer output lines,
since we no longer output the message "discarding lookahead token
end of input ()".
fix a small glitch in debugging output.
(yyprocessOneStack, yyrecoverSyntaxError, yyparse): Print newline
after YY_SYMBOL_PRINT where needed.
(struct yyGLRState): Add some comments.
(struct yySemanticOption): Add some comments.
(union yyGLRStackItem): Add comment.
(yymergeOptionSets): Correct this to properly perform the union,
avoiding infinite reported by Michael Rosien.
Update comment.
* tests/glr-regression.at: Add test for GLR merging error reported
by M. Rosien.
From Marcus Holland-Moritz.
* data/yacc.c (yyerrlab): Move the code that destroys the stack
from here....
(yyreturn): to here. That way, destructors are called properly
even if the stack overflows, or the user calls YYACCEPT or
YYABORT. Stack-overflow problem reported by Marcus Holland-Moritz.
(yyoverflowlab): Destroy the lookahead.
associated with tokens.
* src/output.c (prepare_symbols): Don't escape strings,
since users don't want to see C escapes.
* tests/calc.at (AT_CHECK_CALC): Adjust to lack of quotes
in diagnostics.
* tests/regression.at (Token definitions, Web2c Actions): Likewise.
(string_as_id): Quote $1 before using it as a key, since the
lexer no longer quotes it for us.
(string_content): Don't strip quotes, since lexer no longer
quotes it for us.
* src/scan-gram.l: Include quotearg.h.
("\""): Omit quote.
("'"<SC_ESCAPED_CHARACTER>): Quote symbol before using it as
a key, since the rest of the lexer doesn't quote it.
* src/symtab.c (symbol_get): Don't quote symbol; caller does it now.
* tests/regression.at (Token definitions): Check for backslashes
in token strings.
YYSTACK_USE_ALLOCA to 1 if __GNUC__ or alloca are defined;
otherwise, we're not testing alloca. Unfortunately there's no
simple way to consult HAVE_ALLOCA here.
(parser::state_type, parser::semantic_type, parser::location_type):
Private, not public.
(parser::parse): Return ints, not bool.
Returning a bool introduces a problem: 0 corresponds to false, and
it seems weird to return false on success. Returning true changes
the conventions for yyparse.
Alternatively we could return void and send an exception.
There is no clear consensus (yet?).
(state_stack, semantic_stack, location_stack): Rename as...
(state_stack_type, semantic_stack_type, location_stack_type): these.
Private, not public.
* tests/c++.at: New.
* tests/testsuite.at, tests/Makefile.am: Adjust.
* data/lalr1.cc (b4_filename_type): New.
(position::filename): Use it.
(parser.hh): Move the inclusion of stack.hh and location.hh below
the user code, so that needed headers for the filename type can be
included first.
Forward declare them before the user code.
* tests/Makefile.am (check-local, installcheck-local): Pass
TESTSUITEFLAGS to the TESTSUITE.
(error_): Rename as...
(error): this, since it is visible to the user.
Adjust callers.
(Parser::message): Now an automatic variable from...
(Parser::yyreport_syntax_error_): here.
* tests/actions.at, tests/calc.at, tests/regression.at: Adjust to
Parser::error.
* tests/input.at: Escape $.
* data/lalr1.cc (debug_): Rename as...
(yydebug_): so that the parser's internals are always in the yy*
pseudo namespace.
Adjust uses.
(b4_parse_param_decl): Remove the leading comma as it is now only
called as unique argument list.
(Parser::Parser): Remove the constructor accepting a location and
an initial debugging level.
Remove from the other ctor the argument for the debugging level.
(debug_level_type, debug_level, set_debug_level): New.
* tests/actions.at, tests/calc.at, tests/regression.at: Adjust
constructor calls.
specifying whether the test should be skipped. Use it tp
specify that the [%defines %skeleton "lalr1.cc"] tests currently
fail on some hosts, and should be skipped.
* data/lalr1.cc (Parser::yycdebug_): New, a pointer, to allow
changes.
(YYCDEBUG): Adjust.
Use it instead of cdebug_.
(Parser::debug_stream, Parser::set_debug_stream): New.
(Parser::symprint_): Define cdebug_ for temporary backward
compatibility.
* tests/actions.at (_AT_CHECK_PRINTER_AND_DESTRUCTOR): Use
debug_stream ().
Introduce another definition to address simple location arrays.
(yyGLRStack): New member: yyerror_range.
(yyrecoverSyntaxError, yyparse): Update it.
(yyrecoverSyntaxError): Use it when shifting the error token to
have an accurate range, equivalent to the one computed by both
yacc.c and lalr1.cc.
* tests/cxx-type.at (_AT_TEST_GLR_CXXTYPES): Change its yylex so
that column numbers start at column 0, as per GNU Coding
Standards, the others tests, and the doc.
(_AT_RESOLVED_GLR_OUTPUT_WITH_LOC, _AT_AMBIG_GLR_OUTPUT_WITH_LOC):
Adjust to the above change (first column is 0).
And adjust the location of the "<error>", now covering the whole
line.
type names, destructors, and printers.
Reported by Alexandre Duret-Lutz.
Recode the handling of associativity and precedence in terms
of symbol_precedence_set.
Accept no redeclaration at all, not even equal to the previous
value.
(redeclaration): New.
Use it to factor redeclaration complaints.
(symbol_make_alias): Don't set the type of the alias, let
symbol_check_alias_consistency do it as for other features.
* src/symtab.h (symbol): Add new member prec_location, and
type_location.
* src/symtab.c (symbol_precedence_set, symbol_type_set): Set them.
* tests/input.at (Incompatible Aliases): New.
whether the reducion was empty or not. This leaves room to
improve the use of YYLLOC_DEFAULT in such a case.
lalr1.cc is still experimental, so changing this is acceptable.
And finally, there are probably not many users who changed the
handling of locations in GLR, so changing is admissible too.
* data/glr.c, data/lalr1.cc, data/yacc.c (YYLLOC_DEFAULT): On an
empty reduction, set @$ to an empty location ending the previously
stacked symbol.
Adjust uses to make sure the code is triggered on empty
reductions.
* tests/actions.at (_AT_CHECK_PRINTER_AND_DESTRUCTOR): Adjust the
expected output: empty reductions have empty locations.
"initial push" (corresponding to an hypothetical beginning-of-file).
And let lalr1.cc honor %initial-action.
* doc/bison.texinfo (Initial Action Decl): Clarify, and add an
example.
* data/lalr1.cc (Parser::initlocation_): Remove, bad experiment.
(Parser::Parser): Remove the ctor that used to initialize it.
(Parser::parse): Like in the other skeletons, issue the "starting
parse" message before any action.
Honor %initial-action.
Initialize the stacks with the lookahead.
* data/yacc.c: Let $$ and @$ in %initial-action designate the
look-ahead.
Push them in the stacks.
* tests/actions.at, tests/calc.at: Adjust the C++ ctor invocations.
Define when we know YYLTYPE's structure, i.e., when the default
YYLLOC_DEFAULT is used.
* data/c.m4 (b4_yysymprint_generate): Use it.
* data/lalr1.cc (YYLLOC_DEFAULT): Stop relying on the initial
value of the result.
(error_start_): Replace with...
(error_range_): this location array.
This allows to replace code relying on the implementation of
locations by portable code.
* data/yacc.c (yylerrsp): Replace with...
(yyerror_range): this.
Every time a token is popped, update yyerror_range[0], to have an
accurate location for the error token.
* data/glr.c (YY_LOCATION_PRINT): New.
(yyprocessOneStack): Fix an invocation of YY_SYMBOL_PRINT:
deference a pointer.
* tests/actions.at (_AT_CHECK_PRINTER_AND_DESTRUCTOR): No longer
report the location in %printers.
* src/scan-skel.l: Instead of abort, report error messages to ease
understanding skeleton scanning failures.
* data/yacc.c, data/lalr1.c, data/glr.c: Use YY_SYMBOL_PRINT
to report Shifts instead of ad hoc YYDPRINTF invocations,
including for the error token.
* data/lalr1.cc (symprint_): Output the location.
* tests/actions.at (_AT_CHECK_PRINTER_AND_DESTRUCTOR): In C++, don't
output the location within the %printer.
Activate GLR tests, at least to make sure they compile properly.
They still don't pass though.
* tests/calc.at: Adjust expect verbose output, since now "Entering
state..." is on a different line than the "Shifting" message.
Bison directive from the Bison file to the invocation of this
macro, so that these directives are passed to
AT_BISON_OPTION_PUSHDEFS to get correct help macros.
Use these helping macros (e.g., AT_LOC, AT_VAL and so forth).
Move the AT_SETUP/AT_CLEANUP outside, to report as test title
the extra Bison directives instead of the whole series.
Change the grammar so that there are recoverable errors, and
unrecoverable errors. Now we can have the parser give up before
consuming the whole input. As a result we now can observe that
the lookahead is freed when needed.
Change the parser source to parse argv[1] instead of a hard coded
string.
Simplify yylex, and give a value and location to EOF.
Simplify some invocations of AT_CHECK_PRINTER_AND_DESTRUCTOR that
passed directives already coded in the file.
Add some tests to check the location of "error".
For some tests, the C++ parser is correct, and not yacc.c.
For other tests, they provide different, but unsatisfying, values,
so keep the C++ value so that at least one parser is "correct"
according to the test suite.
(Actions after errors): Remove, this is subsumed by the
AT_CHECK_PRINTER_AND_DESTRUCTOR series.
invoked, yydestruct the lookahead.
* tests/calc.at (Calculator $1): Update the expected lengths of
traces: there is an added line for the discarded lookahead.
* doc/bison.texinfo (Destructor Decl): Some rewording.
Define "discarded" symbols.
AT_CHECK_PRINTER_AND_DESTRUCTOR): New argument
UNION-FLAG. All callers changed.
(_AT_CHECK_PRINTER_AND_DESTRUCTOR): Don't assume %union.
Use type char, not unsigned int, when declaring an array of char;
this lets us remove a cast.
(Printers and Destructors): Add non-%union test cases.
