Bison emits strings to translate in the generated code, for builtin
tokens. So they appear only in generated parsers, which are not
shipped, so they are not in the src tree, so we cannot use them in our
POTFILE.
Except src/parse-gram.c, which is in the source tree. And even in the
git repo. But to avoid useless diffs in the repo, we do not keep the
src/parse-gram.c _with_ the #lines. This is done in a dist-hook which
regenerates src/parse-gram.c when we run "make dist".
Unfortunately, then, update-po traverses the whole tree and sees that
the location of the strings to translate in src/parse-gram.c have
changed, so the bison.pot is to be updated. And that is not possible
in the "make dist" which is run within "make distcheck"
(not the one preparing the dist for distcheck, the one run by
distcheck to check that a distributed tarball can build a tarball)
because then the src tree is read-only.
So let's not put src/parse-gram.c in the POTFILE, and expose these
strings to gettextize by hand.
* src/i18n-strings.c: New.
* po/POTFILES.in: Add it, and remove src/parse-gram.c.
Some people have been using that symbol. Some even have #defined it
themselves.
https://lists.gnu.org/r/bison-patches/2020-04/msg00138.html
Let's provide backward compatibility, having it point to YYUNDEF, so
that an error message is generated.
* data/skeletons/yacc.c (YYERRCODE): New, at the exact same location
it was defined before.
These are internal details. `type_get ()` is still there to ensure
backward compatibility, `kind ()` being the modern way.
* data/skeletons/c++.m4 (by_type, by_type::type): Rename as...
(by_kind, by_kind::kind_): this.
Adjust dependencies.
From
public interface Lexer {
/* Token kinds. */
/** Token number, to be returned by the scanner. */
static final int YYEOF = 0;
/** Token number, to be returned by the scanner. */
static final int YYERRCODE = 256;
/** Token number, to be returned by the scanner. */
static final int YYUNDEF = 257;
/** Token number, to be returned by the scanner. */
static final int BANG = 258;
...
/** Deprecated, use b4_symbol(0, id) instead. */
public static final int EOF = YYEOF;
to
public interface Lexer {
/* Token kinds. */
/** Token "end of file", to be returned by the scanner. */
static final int YYEOF = 0;
/** Token error, to be returned by the scanner. */
static final int YYerror = 256;
/** Token "invalid token", to be returned by the scanner. */
static final int YYUNDEF = 257;
/** Token "!", to be returned by the scanner. */
static final int BANG = 258;
...
/** Deprecated, use YYEOF instead. */
public static final int EOF = YYEOF;
* data/skeletons/java.m4 (b4_token_enum): Display the symbol's tag in
comment.
* data/skeletons/lalr1.java: Address overquotation issue.
* examples/java/calc/Calc.y, examples/java/simple/Calc.y: Use YYEOF,
not EOF.
On an invalid character literal such as "'\777'" we used to produce
two errors:
input.y:2.9-12: error: invalid number after \-escape: 777
input.y:2.8-13: error: empty character literal
Get rid of the second one.
* src/scan-gram.l (STRING_GROW_ESCAPE): New.
* tests/input.at: Adjust.
I'm quite pleased to see that the tricky case of glr.c was already
prepared by the changes to support syntax_error exceptions. Better
yet, it is actually syntax_error that becomes a special case of the
general pattern: make yytoken be YYERRCODE.
* data/skeletons/glr.c (YYFAULTYTOK): Remove the now useless (Basil)
Faulty token.
Instead, use the error token.
* data/skeletons/lalr1.d, data/skeletons/lalr1.java: When computing
the action, first check the case of the error token.
* tests/calc.at: Check cases for the error token symbols before and
after it.
* data/skeletons/yacc.c (yyparse): When the scanner returns YYERRCODE,
go directly to error recovery (yyerrlab1).
However, don't keep the error token as lookahead, that token is too
special.
* data/skeletons/lalr1.cc: Likewise.
* examples/c/bistromathic/parse.y (yylex): Use that feature to report
nicely invalid characters.
* examples/c/bistromathic/bistromathic.test: Check that.
* examples/test: Neutralize gratuitous differences such as rule
position.
* tests/calc.at: Check that case in C only.
The other case seem to be working, but that's an illusion that the
next commit will address (in fact, they can enter endless loops, and
report the error several times anyway).
* examples/c/bistromathic/parse.y: here.
* examples/c/bistromathic/bistromathic.test: Check it.
Included a stupid case where the error is actually ignored.
When the user ctrl-d the line, we left the cursor not at col 0.
Let's fix that.
This revealed a few short-comings in the testing framework.
* examples/test (run): Also display the diffs.
And support -n.
* examples/c/bistromathic/bistromathic.test
* examples/c/bistromathic/parse.y
We will not keep YYERRCODE anyway, it causes backward compatibility
issues. So as a first step, let all the skeletons use that name,
until we have a better one.
* data/skeletons/bison.m4, data/skeletons/glr.c,
* data/skeletons/glr.cc, data/skeletons/lalr1.cc,
* data/skeletons/lalr1.d, data/skeletons/lalr1.java,
* data/skeletons/yacc.c, doc/bison.texi, tests/headers.at,
* tests/input.at:
here.
* etc/bench.pl.in: Don't force parse.error=detailed
Use a simpler way to display the pseudo %bison directive.
(&bench_with_gbenchmark): Give details about the compiler.
We could try to avoid the weird "#if 1", but then the indentation of
the inner #if would be wrong. Let' keep it this way.
* data/skeletons/yacc.c: here.
Also, avoid sticking the comment to the directive.
Currently it was only using stubs. Let's actually translate the
strings using gettext.
* examples/c/bistromathic/local.mk: Define LOCALEDIR, BISON_LOCALEDIR
and link with libintl.
* examples/c/bistromathic/parse.y: Use them.
Remove useless includes.
Take ENABLE_NLS into account.
(error_format_string): New.
(yyreport_syntax_error): Rewrite to rely on a format string, which is
more appropriate for internationalization.
* examples/c/bistromathic/Makefile: We no longer use Flex.
We need readline and intl.
* doc/bison.texi: Point to bistromathic for a better option for
internationalization.
* po/POTFILES.in: Add bistromathic.
In C/C++, N_ is a no-op. Define it if the user didn't.
Suggested by Frank Heckenbach.
https://lists.gnu.org/r/bug-bison/2020-04/msg00010.html
* src/output.c (prepare_symbol_names): Rename has_translations as
has_translations_flag.
* data/skeletons/bison.m4 (b4_has_translations_if): New.
* data/skeletons/java.m4 (b4_trans): Use it.
* data/skeletons/glr.c, data/skeletons/lalr1.cc, data/skeletons/yacc.c
(N_): Provide a default definition.
* src/output.c (prepare_symbol_names): Don't play tricks with the
symbols, it's quite too late.
(has_translations): Move to...
* src/symtab.c: here.
(symbols_pack): Use it to enable translation for special symbols.
On macOS, wc -l always prepends the result with a tab, even when fed
by stdin. But anyway, we should have used `grep -c -v`, which appears
to be portable according to Autoconf's "Limitations of Usual Tools"
section.
Reported by Denis Excoffier.
https://lists.gnu.org/r/bug-bison/2020-04/msg00009.html
* tests/calc.at (_AT_CHECK_CALC): Use grep's -c instead.
macOS' version of readline does not repeat stdin on stdout in
non-interactive mode, contrary to the current version of GNU readline.
* examples/test: Add support for strip_prompt.
* examples/c/bistromathic/bistromathic.test (strip_prompt): Set it
when needed.
Early exit when needed.
symbol_type::token () was removed: it returned the token kind of a
symbol. To do that, one needs to convert from the symbol kind to the
token kind, which requires a table.
This broke some users' unit tests for scanners, see
https://lists.gnu.org/r/bug-bison/2020-01/msg00001.htmlhttps://lists.gnu.org/r/bug-bison/2020-03/msg00020.htmlhttps://lists.gnu.org/r/help-bison/2020-04/msg00005.html
Instead of making this possible again, let's check the symbol's kind
instead. So give proper access to a symbol's kind.
That feature existed, undocumented, as 'type_get()'. Let's rename
this as 'kind()'.
* data/skeletons/c++.m4, data/skeletons/glr.cc,
* data/skeletons/lalr1.cc (type_get): Rename as...
(kind): This.
(type_get): Install a backward compatibility alias.
* doc/bison.texi (Complete Symbols): Document symbol_type and
symbol_type::kind.
* data/skeletons/c++.m4: Define the old names in terms on the new
ones, instead of the converse.
* doc/bison.texi (C++ Parser Interface): Be more extensive about
token_kind_type.
Not all the symbols have a fixed symbol code. UNDEF's one is fixed:
-2.
* data/skeletons/glr.c, data/skeletons/lalr1.cc, data/skeletons/lalr1.d,
* data/skeletons/yacc.c: here.
Why didn't I think about this before??? symbolName should be a method
of SymbolKind.
* data/skeletons/lalr1.java (YYParser::yysymbolName): Move as...
* data/skeletons/java.m4 (SymbolKind::getName): this.
Make the table a static final table, not a local variable.
Adjust dependencies.
* doc/bison.texi (Java Parser Interface): Document i18n.
(Java Parser Context Interface): Document SymbolKind.
* examples/java/calc/Calc.y, tests/local.at: Adjust.
* doc/bison.texi (C++ Parser Context): New.
* data/skeletons/lalr1.cc (parser::yysymbol_name): Rename as...
(parser::symbol_name): this.
(A Complete C++ Example): Promote LAC, now that we have it.
Promote parse.error detailed over verbose.
* examples/c++/calc++/calc++.test, tests/local.at: Adjust.
* NEWS (Deep overhaul of the symbol and token kinds): New.
* doc/bison.texi: Promote YYEOF over "0" in scanners.
(Token Decl): No longer show YYEOF here, it now works by default.
(Token I18n): More details about YYEOF here.
(Calc++): Just use YYEOF.
I have been hesitating a lot before doing it ---after all the user
must not use this kind, so what's the point of showing it in
yytoken_kind_t. And eventually I chose to play it safe with the
typing system and make it possible to use yytoken_kind_t for all the
tokens, even the "empty token".
* data/skeletons/c.m4: Give an id and a tag to YYEMPTY.
(b4_token_enums): Define YYEMPTY.
* data/skeletons/c++.m4 (b4_token_enums): Define YYEMPTY.
* data/skeletons/glr.c, data/skeletons/glr.cc, data/skeletons/yacc.c:
(YYEMPTY): Remove.
Use b4_symbol(-2, id) instead.
The user should think of yypcontext fields as accessible only via
yypcontext_* functions. So let's rename yyexpected_tokens to reflect
that.
Let's _not_ rename yyreport_syntax_error, as the user may define this
function, and is not allowed to access directly the fields of
yypcontext_t: she *must* use the "accessors". This is comparable to
the case of C++/Java where the user defines
parser::report_syntax_error, not parser::context::report_syntax_error.
* data/skeletons/glr.c, data/skeletons/yacc.c (yyexpected_tokens):
Rename as...
(yypcontext_expected_tokens): this.
Adjust dependencies.
The name "$end" is nice in the report, in particular it avoids that
pointed-rules (aka items) be too long. It also helps keeping them
"standard".
But it is bad in error messages, we should report "end of file" (or
maybe "end of input", this is debatable). So, unless the user already
defined the alias for the error token herself, make it "end of file".
It should even be translated if the user already translated some
tokens, so that there is now no strong reason to redefine the $end
token.
* src/output.c (prepare_symbol_names): Issue "end of file" instead of
"$end".
* data/skeletons/lalr1.java (yytnamerr_): Remove the renaming hack.
* build-aux/update-test: Accept files with names containing a "+",
such as c++.at.
* tests/actions.at, tests/c++.at, tests/conflicts.at,
* tests/glr-regression.at, tests/regression.at, tests/skeletons.at:
Adjust.
Yet, don't change the structure identifier to avoid introducing
conflicts in Vincent Imbimbo's PR (which, amusingly enough, is about
conflicts).
* src/symtab.c: here.
* tests/diagnostics.at, tests/input.at: Adjust.
yy::parser features a parse() function, not a yyparse() one.
* data/skeletons/lalr1.cc (yyreport_syntax_error)
(context::yyexpected_tokens): Rename as...
(report_syntax_error, context::expected_tokens): these.
Currently EOF is handled in an adhoc way, with a #define YYEOF 0 in
the implementation file. As a result, the user has to define her own
EOF token if she wants to use it, which is a pity.
Give the $end token a visible kind name, YYEOF. Except that in C,
where enums are not scoped, we would have collisions between all the
definitions of YYEOFs in the header files, so in C, make it
<api.PREFIX>EOF.
* data/skeletons/c.m4 (YYEOF): Override its name to avoid collisions.
Unless the user already gave it a different name.
* data/skeletons/glr.c (YYEOF): Remove.
Use ]b4_symbol(0, [id])[ instead.
Add support for "pre_epilogue", for glr.cc.
* data/skeletons/glr.cc: Remove dead code (never emitted #undefs).
* data/skeletons/yacc.c
* src/parse-gram.c
* src/reader.c
* src/symtab.c
* tests/actions.at
* tests/input.at
* data/skeletons/bison.m4 (b4_symbol_token_kind): Give a definition to
$undefined.
(b4_token_visible_if): $undefined has an id.
* src/output.c (prepare_symbol_definitions): Stop lying: $undefined
_is_ a token.
* tests/input.at: Adjust.
There are people out there that do use YYERRCODE (the token kind of
the error token). See for instance
https://github.com/borbolla-automation/SPC_Machines/blob/3812012bb782bfdfe7b325950a35cd337925fcad/unixODBC-2.3.2/Drivers/nn/yylex.c.
Currently, YYERRCODE is defined by yacc.c in an adhoc way as a #define
in the *.c file only. It belongs with the other token kinds.
YYERRCODE is not a nice name, it does not fit in our naming scheme.
YYERROR would be more logical, but it collides with the YYERROR macro.
Shall we keep the same name in all the skeletons? Besides, to avoid
collisions in C, we need to apply the api prefix: YYERRCODE is
actually <PREFIX>ERRCODE. This is not needed in the other languages.
* data/skeletons/bison.m4 (b4_symbol_token_kind): New.
Map the error token to "YYERRCODE".
* data/skeletons/yacc.c (YYERRCODE): Don't define it, it's handled by...
* src/output.c (prepare_symbol_definitions): this.
* tests/input.at (Redefining the error token): Check it.
* data/skeletons/bison.m4 (b4_symbol_kind): Dispatch on the UNDEF
token number rather than its name.
* data/skeletons/c++.m4, data/skeletons/c.m4, data/skeletons/java.m4:
Comment changes.
A forthcoming commit (tokens: properly define the "error" token kind)
revealed a problem in the C++ generated headers: they are not
self-contained. With this file:
%language "c++"
%define api.value.type variant
%code {
static int yylex (yy::parser::semantic_type *lvalp);
}
%token <int> X
%%
exp:
X { printf ("x\n"); }
;
%%
void
yy::parser::error (const std::string& m)
{
std::cerr << m << '\n';
}
static
int yylex (yy::parser::semantic_type *lvalp)
{
static int const input[] = {yy::parser::token::X, 0};
static int toknum = 0;
return input[toknum++];
}
int
main (int argc, char const* argv[])
{
yy::parser p;
return p.parse ();
}
the generated header fails to compile cleanly (foo.cc just #includes
the generated header):
$ clang++-mp-9.0 -c -Wundefined-func-template foo.cc
In file included from foo.cc:1:
bar.tab.hh:550:12: warning: instantiation of function 'yy::parser::basic_symbol<yy::parser::by_type>::basic_symbol' required here, but no definition is available
[-Wundefined-func-template]
struct symbol_type : basic_symbol<by_type>
^
bar.tab.hh:436:7: note: forward declaration of template entity is here
basic_symbol (basic_symbol&& that);
^
bar.tab.hh:550:12: note: add an explicit instantiation declaration to suppress this warning if 'yy::parser::basic_symbol<yy::parser::by_type>::basic_symbol' is explicitly instantiated
in another translation unit
struct symbol_type : basic_symbol<by_type>
^
1 warning generated.
* data/skeletons/c++.m4 (b4_public_types_define): Move the
implementation of the basic_symbol move-ctor to...
(b4_public_types_define): here, its declaration.
* tests/headers.at (Sane headers): Use a declared token so that the
corresponding token constructor is declared. Which triggers the
aforementioned issue.
Akim Demaille