Provide users with a means to query for the currently allowed tokens.
Could be used for autocompletion for instance.
* data/skeletons/yacc.c (yyexpected_tokens): New, extracted from
yysyntax_error_arguments.
* examples/c/calc/calc.y (PRINT_EXPECTED_TOKENS): New.
Use it.
We use a different format to check parse.error custom. Compute the
"verbose" one from it instead of forcing the test author to provide
the various formats of expected error messages.
* tests/calc.at (_AT_CHECK_CALC_ERROR): Handle this transformation
when needed.
Simplify callers.
* tests/local.at (AT_ERROR_CUSTOM_IF, AT_ERROR_VERBOSE_IF)
(AT_ERROR_SIMPLE_IF): New.
(AT_YYERROR_DEFINE(c)): Generate yyreport_syntax_error.
* tests/calc.at (_AT_CHECK_CALC_ERROR): Accept custom error messages
as additional test case.
Use it.
Add a new test case for %define parse.error custom.
When parse.error is custom, let users define a yyreport_syntax_error
function, and use it.
* data/skeletons/bison.m4 (b4_error_verbose_if): Accept 'custom'.
* data/skeletons/yacc.c: Implement it.
* examples/c/calc/calc.y: Experiment with it.
That allows users to cover more cases, such as easily filtering some
arguments they don't want to expose. But they now have to call
yysymbol_name explicitly.
* data/skeletons/yacc.c (yysyntax_error_arguments, yysyntax_error):
Deal with symbol numbers instead of symbol names.
Isolate a function that returns the list of expected and unexpected
tokens. It will be exposed to users willing to customize their error
messages.
* data/skeletons/yacc.c (yyparse_context_t): New.
(yyerror_message_arguments): New, extracted from yysyntax_error.
When not empty, AT_PARSE_PARAMS was guaranteed to end with a comma.
Remove the trailing comma, so that we can use AT_PARSE_PARAMS at the
end of the arguments, not only at the beginning.
* tests/local.at: here.
Unfortunately, m4_append relies on the macro not being defined whereas
we would have preferred it to check for emptiness. So use
m4_define/m4_undefine instead of m4_pushdef/m4_popdef.
Let's have C be the reference, and match it elsewhere. Maybe C is too
verbose and some adjustments are needed, but then that would be done
in another batch of patches.
* data/skeletons/lalr1.cc: Print the stack once we popped after
YYERROR, and before emptying the stack at the end of parsing.
Currently the C and C++ parse traces differ in the order in which the
stack is displayed: bottom up in C, top down in C++. Let's stick to
the C order.
* data/skeletons/stack.hh (stack::iterator, stack::const_iterator)
(begin, end): Be forward, not backward.
Provide the users with a public API to get the name of the tokens. A
thin wrapper around yytname.
* data/skeletons/yacc.c (yysymbol_name): New.
Use it.
I was hoping it would help us catch warnings when char is
unsigned (see 78bb152a63), but it does
not seem to help. It's a pity that the compiler is the same all over
the place, I would have preferred testing others.
* .travis.yml: here.
Supporting YYERROR_VERBOSE via cpp is a nuisance: m4 is in charge of
handling alternatives. When adding more options for %define
parse.error, supporting both CPP and M4 is too complex. Anyway,
YYERROR_VERBOSE was deprecated long ago.
* data/skeletons/yacc.c: Use m4 only to handle verbose/simple error
messages.
I would like to offer new ways to build the error message. As a first
step, let's simplify yysyntax_error whose first loop does two things
at the same time: (i) collect the tokens to be reported in the error
message, and (ii) accumulate their sizes and possibly return
"overflow". Let's pull (ii) in a second step.
Then test 525 (regression.at:1193: parse.error=verbose overflow)
failed. This test checks that we correctly report "memory overflow"
when the error message is too large. However the test is mistaken: it
is triggered in a place where there are five (large) expected tokens,
so anyway we would not display them, so there is no (memory) overflow
here! Transform this test to (i) check that indeed there is no
overflow, and (ii) create syntax_error3 which does check the intended
behavior, but with four expected tokens.
* data/skeletons/yacc.c (yysyntax_error): First compute the list of
arguments, then compute yysize.
* tests/regression.at (parse.error=verbose overflow): Enhance and fix.
GCC's -Wchar-subscripts may report issues on platforms where char is
unsigned. Unfortunately the current CI does not reproduce the
problem. But that would allow contributors to report issues if the
warning appears somewhere.
See 139d065594.
Problem reported by Andy Fiddaman in:
https://lists.gnu.org/r/bug-bison/2019-12/msg00021.html
* configure.ac (warn_common): Add -Wchar-subscripts.
Another breakage revealed by vcsn.
* data/skeletons/c++.m4 (yytranslate_): Do not hard code "yy" and
"parser", both can be changed by the user.
Actually, since we are in the parser itself, there's really no need to
qualify the type.
* data/skeletons/glr.c (YYASSERT): Rename as...
(YY_ASSERT): this, for consistency with yacc.c, and also to emphasize
the fact that this is not for the end user (YY_ prefix).
* tests/glr-regression.at: Define parse.assert.
Now that we use small integral types, possibly unsigned (e.g.,
unsigned char), to store state numbers, using -1 to denote an empty
state (i.e., a state that stores no semantical value) is very
dangerous: it will be confused with state 255, which might be
non-empty.
Rather than allocating a larger range of state numbers to keep the
empty-state apart, let's use the number of a state known to store no
value. The initial state, numbered 0, seems to fit perfectly the job.
Reported by Frank Heckenbach.
https://lists.gnu.org/archive/html/bug-bison/2019-11/msg00016.html
* data/skeletons/lalr1.cc (empty_state): Be 0.
It is not used. And its implementation was wrong when api.token.raw
was defined, as it was still mapping to the external token numbers,
instead of the internal ones. Besides it was provided only when
api.token.constructor is defined, yet always declared.
* data/skeletons/c++.m4 (by_type::token): Remove, useless.
Reported by Frank Heckenbach.
https://lists.gnu.org/archive/html/bug-bison/2019-11/msg00016.html
The cast is needed when yytranslate_'s argument type is token_type,
i.e., when api.token.constructor is defined.
373. types.at:138: testing lalr1.cc api.value.type=variant api.token.constructor ...
======== Testing with C++ standard flags: ''
../../tests/types.at:138: bison --color=no -fno-caret -o test.cc test.y
../../tests/types.at:138: $CXX $CXXFLAGS $CPPFLAGS $LDFLAGS -o test test.cc $LIBS
stderr:
test.cc:966:16: error: result of comparison of constant 257 with
expression of type 'yy::parser::token_type'
(aka 'yy::parser::token::yytokentype') is always true
[-Werror,-Wtautological-constant-out-of-range-compare]
else if (t <= user_token_number_max_)
~ ^ ~~~~~~~~~~~~~~~~~~~~~~
1 error generated.
It is because it is expected that when api.token.constructor is
defined, only symbol constructors will be used, that yytranslate_ then
takes a token_type. But it is wrong: we still allow literal
characters in this case, as demonstrated by test 373 for instance.
%define api.value.type variant
%define api.token.constructor
%token <std::pair<int, int>> '1' '2';
[...]
static yy::parser::symbol_type yylex ()
{
static char const input[] = "12";
int res = input[toknum++];
typedef yy::parser::symbol_type symbol;
if (res)
return symbol (res, std::make_pair (res - '0', res - '0' + 1));
else
return symbol (res);
}
So let yytranslate_ always take an int, which makes the cast truly
useless.
* data/skeletons/c++.m4, data/skeletons/lalr1.cc (yytranslate_): here.
The C++ implementation of LAC did not skip the $undefined token,
probably because it was not exposed. Expose it, and use clearer
names.
* data/skeletons/c++.m4: Don't define undef_token_ in yytranslate_,
but...
* data/skeletons/lalr1.cc (yy_undef_token_): here.
Use a more precise type to define yy_undef_token_ and yy_error_token_.
Unfortunately we move from a compile-time value defined via an enum to
a static const member. Eventually we should make it constexpr.
Make LAC implementation more alike yacc.c's one.
* data/skeletons/lalr1.d, data/skeletons/lalr1.java: Don't expose
yyuser_token_number_max_ and yyundef_token_. Do as in C++: scope them
into yytranslate_, and only when api.token.raw is not defined.
(yyterror_): Rename as...
(yy_error_token_): this.
* data/skeletons/lalr1.d (token_number_type): New.
Use it.
Can't be done in the Java backend, as Java does not have type aliases.
* data/skeletons/lalr1.d, data/skeletons/lalr1.java (yytoken_number_):
Remove, useless.
Was used in ancient C skeletons to support YYPRINT, long obsoleted by
%printer.
It is not used at all. We will remove it also from yacc.c, but
later (see TODO).
* data/skeletons/lalr1.cc, data/skeletons/lalr1.d,
* data/skeletons/lalr1.java (yyerrcode_):
Remove.
Reported by Frank Heckenbach.
https://lists.gnu.org/archive/html/bug-bison/2019-11/msg00016.html
* data/skeletons/c++.m4 (b4_yytranslate_define): Don't use yyeof_ as
if it had two different types.
It is used once against the input argument, which is the value
returned by yylex, which is an "external token number", typically an
int. It is also used as output type, an "internal symbol number".
It turns out that in both cases we mean "0", but let's keep yyeof_
only for the case "internal symbol number", i.e., _after_ conversion
by yytranslate.
This frees us from one cast.
The current code for yysyntax_error for %define parse.error verbose is
fishy (given that YYEMPTY is -2, invalid argument for yytname[]):
static int
yysyntax_error ([...])
{
YYPTRDIFF_T yysize0 = yytnamerr (YY_NULLPTR, yytname[yytoken]);
[...]
if (yytoken != YYEMPTY)
A nearby comment reports
The only way there can be no lookahead present (in yychar) is if
this state is a consistent state with a default action. Thus,
detecting the absence of a lookahead is sufficient to determine
that there is no unexpected or expected token to report. In that
case, just report a simple "syntax error".
So it _is_ possible to call yysyntax_error with yytoken == YYEMPTY,
albeit quite difficult when meaning to, so virtually impossible by
accident (after all, there was never a bug report about this).
I failed to produce a test case, but Joel E. Denny provided me with
one (added to the test suite below). The yacc.c skeleton fails on
this, and once fixed dies on a second problem. The glr.c skeleton was
also dying, but immediately of this second problem.
Indeed we were not allocating space for the error message's final \0.
This was hidden by the fact that we only had error messages with at
least an unexpected token displayed, so with at least one "%s" in the
format string, whose size (2) was included (incorrectly) in the final
size of the message (where the %s have been replaced by the actual
content).
* data/skeletons/glr.c, data/skeletons/yacc.c (yysyntax_error):
Do not invoke yytnamerr on YYEMPTY.
Clarify the computation of the length of the _final_ error message,
with the NUL terminator but without the '%s's.
* tests/conflicts.at (Syntax error in consistent error state):
New, contributed by Joel E. Denny.
Having a file named "exception" is risky: the compiler might use that
file in #include.
Reported by 马俊 <[email protected]>.
* tests/local.at (AT_SKIP_IF_EXCEPTION_SUPPORT_IS_POOR): Generate
'exceptions', not 'exception'.
String literals as tokens serve two distinct purposes: freeing from
having to implement the keyword matching in the scanner, and improving
error messages. Most of the time both can be achieved at the same
time, but on occasions, it does not work so well.
We promote their use for error messages. We will also still support
the former case, but it is _not_ the recommended approach.
* doc/bison.texi (Tokens from Literals): Clearly state that we don't
recommend looking up the token types in the list of token names.
String literals, which allow for better error messages, are (too)
liberally accepted by Bison, which might result in silent errors. For
instance
%type <exVal> cond "condition"
does not define “condition” as a string alias to 'cond' (nonterminal
symbols do not have string aliases). It is rather equivalent to
%nterm <exVal> cond
%token <exVal> "condition"
i.e., it gives the type 'exVal' to the "condition" token, which was
clearly not the intention.
Introduce -Wdangling-alias to catch this.
* src/complain.h, src/complain.c: Add support for -Wdangling-alias.
(argmatch_warning_args): Sort.
* src/symtab.c (symbol_check_defined): Complain about dangling
aliases.
* doc/bison.texi: Document it.
* tests/input.at (Dangling aliases): New test.
On
%token TOKEN1
%type <ival> TOKEN1 TOKEN2 't'
%token TOKEN2
%%
expr:
bison -Wyacc gives
input.y:2.15-20: warning: POSIX yacc reserves %type to nonterminals [-Wyacc]
2 | %type <ival> TOKEN1 TOKEN2 't'
| ^~~~~~
input.y:2.29-31: warning: POSIX yacc reserves %type to nonterminals [-Wyacc]
2 | %type <ival> TOKEN1 TOKEN2 't'
| ^~~
input.y:2.22-27: warning: POSIX yacc reserves %type to nonterminals [-Wyacc]
2 | %type <ival> TOKEN1 TOKEN2 't'
| ^~~~~~
The messages appear to be out of order, but they are emitted when the
error is found.
* src/symtab.h (symbol_class): Add pct_type_sym, used to denote
symbols appearing in %type.
* src/symtab.c (complain_pct_type_on_token): New.
(symbol_class_set): Check that %type is not applied to tokens.
(symbol_check_defined): pct_type_sym also means undefined.
* src/parse-gram.y (symbol_decl.1): Set the class to pct_type_sym.
* src/reader.c (grammar_current_rule_begin): pct_type_sym also means
undefined.
* tests/input.at (Yacc's %type): New.
As an extension to POSIX Yacc, Bison's %type accepts tokens.
Unfortunately with string literals as implicit tokens, this is
misleading, and led some users to write
%type <exVal> cond "condition"
believing that "condition" would be associated to the 'cond'
nonterminal (see https://github.com/apache/httpd/pull/72).
* doc/bison.texi: Promote %nterm rather than %type to declare the type
of nonterminals.
* README: A few fixes.
Explain how to install color support.
* README-hacking: Rename as...
* README-hacking.md: this, and convert to Markdown.
Improve typography.
Improve explanations about update-test.
Akim Demaille