{}, since this breaks the common use of `YYDPRINTF ((...));' if a
single statement is desired (e.g. before `else'). Work around GCC
warnings by surrounding corresponding calls with {} if needed.
(yyhasResolvedValue): Remove unused function.
(yymergeOptionSets, yyresolvStack): Use `continue;' for empty
loop body.
(yyreportSyntaxError): Renamed from yyreportParseError.
(yyrecoverSyntaxError): Renamed from yyrecoverParseError.
All uses changed.
"$CC -E foo.h" is allowed, as this doesn't work with the Portland
Group compiler. Instead, use "$CC -E bar.c". Include the .h
file twice in the grammar, as an extra check.
* tests/input.at (Torturing the Scanner): Surround the
backslash-newline tests with "#if 0", to make it less likely that
we'll run into compiler bugs. Bring back solitary \ inside
comment, but add a closing comment to work around HP C bug. Don't
test backslash-newline in C character constant. This should fix
the input.at bug reported by Nelson H. F. Beebe in
<http://mail.gnu.org/pipermail/bug-bison/2002-November/001893.html>.
"#if 0", to make it less likely that we'll run into compiler bugs.
Bring back solitary \ inside comment, but add a closing comment to
work around HP C bug. Don't test backslash-newline in C character
constant. This should fix the input.at bug reported by Nelson
H. F. Beebe in
<http://mail.gnu.org/pipermail/bug-bison/2002-November/001893.html>.
them with auto vars.
Switch to new escape scheme, where @ is the escape character uniformly.
Abort if a stray escape character is found. Avoid unbounded input
buffer when parsing non-escaped text.
escapes other than \\ and \'; this simplifies the code.
(<SC_STRING>): Likewise, for \\ and \".
(<SC_COMMENT,SC_LINE_COMMENT,SC_STRING,SC_CHARACTER,SC_BRACED_CODE,
SC_PROLOGUE,SC_EPILOGUE>): Escape $ and @, too.
Use new escapes @{ and @} for [ and ].
(prepare_symbols): Quote tokens for M4.
(prepare): Don't insert output_infix, output_prefix,
output_parser_name, output_header_name; this is now down by scan-skel.
Insert skeleton as a C string.
(user_actions_output, symbol_destructors_output,
symbol_printers_output): Quote filenames for C and M4.
(b4_input_suffix, b4_output_parser_suffix, b4_output_parser_name,
b4_output_header_suffix, b4_output_header_name, b4_header_guard):
Remove, since they couldn't handle arbitrary characters in file
names.
Remove use of "#ifdef b4_header_guard", since it
mishandled funny characters in file names, and anyway it isn't
needed any more.
(YYSTYPE_IS_TRIVIAL): Define when the .h file would.
(b4_input_suffix, b4_output_parser_suffix, b4_output_parser_name,
b4_output_header_suffix, b4_output_header_name, b4_header_guard):
Remove, since they couldn't handle arbitrary characters in file
names.
Use YYSLP_NEEDED instead of b4_header_guard.
<http://mail.gnu.org/pipermail/bug-bison/2002-November/001815.html>.
* data/glr.c (yyis_pact_ninf, yyis_table_ninf): New macros, which
evaluate to 0 if it's impossible for NINF to be in the respective
table.
(yygetLRActions, yyrecoverParseError): Use them.
* src/scan-gram.l (unexpected_end_of_file): Fix bug: columns were
counted in the token inserted at end of file. Now takes
location_t *, not location_t, so that the location can be
adjusted. All uses changed.
* tests/regression.at (Invalid inputs): Adjust wording in
diagnostic to match the new behavior.
* tests/torture.at (AT_DATA_TRIANGULAR_GRAMMAR,
AT_DATA_HORIZONTAL_GRAMMAR, AT_DATA_LOOKAHEADS_GRAMMAR,
AT_DATA_STACK_TORTURE): Replace `assert (x);' with `if (! (x))
abort ();'. This reduces the runtime of the "Many lookaheads"
test from 27.6 to 2.7 minutes on a 440 MHz Ultrasparc III running
GCC 3.2.
AT_DATA_LOOKAHEADS_GRAMMAR, AT_DATA_STACK_TORTURE): Replace `assert
(x);' with `if (! (x)) abort ();'. This reduces the runtime of the
"Many lookaheads" test from 27.6 to 2.7 minutes on a 440 MHz
Ultrasparc III.
All uses removed.
* src/scan-gram.l: Remove stack option. We no longer use the
stack, since the stack was never deeper than 1; instead, use the
new auto var c_context to record the stacked value.
Remove nounput option. At an unexpected end of file, we now unput
the minimal input necessary to end cleanly; this simplifies the
code.
Avoid unbounded token sizes where this is easy.
(unexpected_end_of_file): New function.
Use it to systematize the error message on unexpected EOF.
(last-string): Now auto, not static.
(YY_OBS_FREE): Remove unnecessary do while (0) wrapper.
(scanner_last_string_free): Remove; not used.
(percent_percent_count): Move decl to just before use.
(SC_ESCAPED_CHARACTER): Return ID at unexpected end of file,
not the (never otherwised-used) CHARACTER.
never deeper than 1; instead, use the new auto var c_context to record
the stacked value.
Remove nounput option. At an unexpected end of file, we now unput
the minimal input necessary to end cleanly; this simplifies the
code.
Avoid unbounded token sizes where this is easy.
(unexpected_end_of_file): New function.
Use it to systematize the error message on unexpected EOF.
(last-string): Now auto, not static.
(YY_OBS_FREE): Remove unnecessary do while (0) wrapper.
(scanner_last_string_free): Remove; not used.
(percent_percent_count): Move decl to just before use.
(SC_ESCAPED_CHARACTER): Return ID at unexpected end of file,
not the (never otherwised-used) CHARACTER.
Initialize to zero if the compiler is being picky.
(INITIAL): Clear braces_level instead of incrementing it.
(SC_BRACED_CODE): Treat <% and %> as { and } when inside C code,
as POSIX 1003.1-2001 requires.
Use consistent diagnostics for reporting type name clashes.
Quote the types with <>, for consistency with Yacc.
* tests/input.at (Type Clashes): Adjust to diagnostic changes.
to check for invalid inputs.
Use mbsnwidth to count columns correctly. Account for tabs, too.
Include mbswidth.h.
(YY_USER_ACTION): Invoke extend_location rather than LOCATION_COLUMNS.
(extend_location): New function.
(YY_LINES): Remove.
Handle CRLF in C code rather than in Lex code.
(YY_INPUT): New macro.
(no_cr_read): New function.
Scan UCNs, even though we don't fully handle them yet.
(convert_ucn_to_byte): New function.
Handle backslash-newline correctly in C code.
(SC_LINE_COMMENT, SC_YACC_COMMENT): New states.
(eols, blanks): Remove. YY_USER_ACTION now counts newlines etc.;
all uses changed.
(tag, splice): New EREs. Do not allow NUL or newline in tags.
Use {splice} wherever C allows backslash-newline.
YY_STEP after space, newline, vertical-tab.
("/*"): BEGIN SC_YACC_COMMENT, not yy_push_state (SC_COMMENT).
(letter, id): Don't assume ASCII; e.g., spell out a-z.
({int}, handle_action_dollar, handle_action_at): Check for integer
overflow.
(YY_STEP): Omit trailing semicolon, so that it's more like C.
(<SC_ESCAPED_STRING,SC_ESCAPED_CHARACTER>): Allow \0 and \00
as well as \000. Check for UCHAR_MAX, not 255.
Allow \x with an arbitrary positive number of digits, as in C.
Check for overflow here.
Allow \? and UCNs, for compatibility with C.
(handle_symbol_code_dollar): Use quote_n slot 1 to avoid collision
with quote slot used by complain_at.
(Symbols): Document that trigraphs have no special meaning in Bison,
nor is backslash-newline allowed.
(Actions): Document that trigraphs have no special meaning.
Check for close failures.
* src/files.h (xfclose): Return void, not int, since it always
returned zero.
* src/files.c (xfclose): Likewise. Report I/O error if ferror
indicates one.
* src/output.c (output_skeleton): Use xfclose rather than fclose
and ferror. xfclose now checks ferror.
* data/glr.c (YYLEFTMOST_STATE): Remove.
(yyreportTree): Use a stack-based leftmost state. This avoids
our continuing battles with bogus warnings about initializers.
Paul Eggert