This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug c++/84609] [8 Regression] internal compiler error: in cp_parser_abort_tentative_parse, at cp/parser.c:28960 (cp_parser_member_declaration())


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84609

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The:
23637             /* The following code wants to know early if it is a
bit-field
23638                or some other declaration.  Attributes can appear before
23639                the `:' token, but are hopefully rare enough that the
23640                simplicity of the tentative lookup pays off.  */
23641             if (cp_next_tokens_can_be_attribute_p (parser)
23642                 || (token->type == CPP_NAME
23643                     && cp_nth_tokens_can_be_attribute_p (parser, 2)
23644                     && (named_bitfld = true)))
23645               {
23646                 cp_parser_parse_tentatively (parser);
23647                 if (named_bitfld)
23648                   cp_lexer_consume_token (parser->lexer);
23649                 cp_parser_attributes_opt (parser);
23650                 token = cp_lexer_peek_token (parser->lexer);
23651                 is_bitfld = cp_lexer_next_token_is (parser->lexer,
CPP_COLON);
23652                 cp_parser_abort_tentative_parse (parser);
23653               }
code assumes that during parsing of the attribute(s) we don't commit the
tentative parsing and it can be unconditionally undone, which is not the case
here, because in:
#0  cp_parser_commit_to_tentative_parse (parser=0x7ffff7ff6bd0) at
../../gcc/cp/parser.c:28976
#1  0x00000000009d267a in cp_parser_parameter_declaration
(parser=0x7ffff7ff6bd0, template_parm_p=false, parenthesized_p=0x7fffffffc26f)
    at ../../gcc/cp/parser.c:21547
#2  0x00000000009d1ed0 in cp_parser_parameter_declaration_list
(parser=0x7ffff7ff6bd0, is_error=0x7fffffffc317) at ../../gcc/cp/parser.c:21305
#3  0x00000000009d1d20 in cp_parser_parameter_declaration_clause
(parser=0x7ffff7ff6bd0) at ../../gcc/cp/parser.c:21228
#4  0x00000000009bca81 in cp_parser_lambda_declarator_opt
(parser=0x7ffff7ff6bd0, lambda_expr=<lambda_expr 0x7fffefdb4040>)
    at ../../gcc/cp/parser.c:10517
#5  0x00000000009bbe71 in cp_parser_lambda_expression (parser=0x7ffff7ff6bd0)
at ../../gcc/cp/parser.c:10172
#6  0x00000000009b1217 in cp_parser_primary_expression (parser=0x7ffff7ff6bd0,
address_p=false, cast_p=false, template_arg_p=false, 
    decltype_p=false, idk=0x7fffffffc8ec) at ../../gcc/cp/parser.c:5257
#7  0x00000000009b539c in cp_parser_postfix_expression (parser=0x7ffff7ff6bd0,
address_p=false, cast_p=false, member_access_only_p=false, 
    decltype_p=false, pidk_return=0x0) at ../../gcc/cp/parser.c:7026
#8  0x00000000009b89aa in cp_parser_unary_expression (parser=0x7ffff7ff6bd0,
pidk=0x0, address_p=false, cast_p=false, decltype_p=false)
    at ../../gcc/cp/parser.c:8318
#9  0x00000000009b9ac6 in cp_parser_cast_expression (parser=0x7ffff7ff6bd0,
address_p=false, cast_p=false, decltype_p=false, pidk=0x0)
    at ../../gcc/cp/parser.c:9086
#10 0x00000000009b9bc3 in cp_parser_binary_expression (parser=0x7ffff7ff6bd0,
cast_p=false, no_toplevel_fold_p=false, decltype_p=false, 
    prec=PREC_NOT_OPERATOR, pidk=0x0) at ../../gcc/cp/parser.c:9187
#11 0x00000000009ba9a0 in cp_parser_assignment_expression
(parser=0x7ffff7ff6bd0, pidk=0x0, cast_p=false, decltype_p=false)
    at ../../gcc/cp/parser.c:9482
#12 0x00000000009b758c in cp_parser_parenthesized_expression_list
(parser=0x7ffff7ff6bd0, is_attribute_list=1, cast_p=false, 
    allow_expansion_p=false, non_constant_p=0x0, close_paren_loc=0x0,
wrap_locations_p=false) at ../../gcc/cp/parser.c:7760
#13 0x00000000009d99f3 in cp_parser_gnu_attribute_list (parser=0x7ffff7ff6bd0)
at ../../gcc/cp/parser.c:25047
#14 0x00000000009d9839 in cp_parser_gnu_attributes_opt (parser=0x7ffff7ff6bd0)
at ../../gcc/cp/parser.c:24965
#15 0x00000000009d9772 in cp_parser_attributes_opt (parser=0x7ffff7ff6bd0) at
../../gcc/cp/parser.c:24922
#16 0x00000000009d74d9 in cp_parser_member_declaration (parser=0x7ffff7ff6bd0)
at ../../gcc/cp/parser.c:23649

we in:
21531         /* After seeing a decl-specifier-seq, if the next token is not a
21532            "(", there is no possibility that the code is a valid
21533            expression.  Therefore, if parsing tentatively, we commit at
21534            this point.  */
21535         if (!parser->in_template_argument_list_p
21536             /* In an expression context, having seen:
21537   
21538                  (int((char ...
21539   
21540                we cannot be sure whether we are looking at a
21541                function-type (taking a "char" as a parameter) or a cast
21542                of some object of type "char" to "int".  */
21543             && !parser->in_type_id_in_expr_p
21544             && cp_parser_uncommitted_to_tentative_parse_p (parser)
21545             && cp_lexer_next_token_is_not (parser->lexer, CPP_OPEN_BRACE)
21546             && cp_lexer_next_token_is_not (parser->lexer,
CPP_OPEN_PAREN))
21547           cp_parser_commit_to_tentative_parse (parser);

commit to tentative parse.  Not really sure what to do here though, can we just
reject statement-expressions and lambdas in attribute arguments?
I guess for the non-standard attributes we can do anything we want, but for
standard attributes like alignas we don't really have a choice, right?
If we could restrict it, we could do something similar to
parser->in_template_argument_list_p when parsing attribute arguments.

Or I can just try to implement what I talked about when submitting the patch:
"As I said on IRC, I hope [[/__attribute__/alignas early is rare enough that
the tentative parsing shouldn't be a big deal, if it is, we could add some
cheaper function that allows us to skip over attributes (return a peek offset
after the attributes given a starting peek offset)."

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]