On Tue, 14 Oct 2025, Alejandro Colomar wrote:
> +/* Parse a _Maxof or _Minof expression. */
> +
> +static struct c_expr
> +c_parser_maxof_or_minof_expression (c_parser *parser, enum rid rid)
> +{
> + const char *op_name = (rid == RID_MAXOF) ? "_Maxof" : "_Minof";
> + struct c_expr expr;
> + struct c_expr result;
> + location_t expr_loc;
> + gcc_assert (c_parser_next_token_is_keyword (parser, rid));
> +
> + location_t start;
> + location_t finish = UNKNOWN_LOCATION;
> +
> + start = c_parser_peek_token (parser)->location;
> +
> + pedwarn_c23 (start, OPT_Wpedantic,
> + "ISO C does not support %qs before C2Y", op_name);
If you're proposing something as an extension, it would be pedwarn, not
pedwarn_c23; pedwarn_c23 is only relevant if actually accepted into C2Y.
> + c_inhibit_evaluation_warnings++;
> + if (c_parser_next_token_is (parser, CPP_OPEN_PAREN)
> + && c_token_starts_compound_literal (c_parser_peek_2nd_token (parser)))
> + {
> + /* Either operator ( type-name ) or operator unary-expression
> + starting with a compound literal. */
All this logic is massively overcomplicated for something that only
accepts a type name and not an expression at all. Don't copy irrelevant
logic from other syntax constructs that's not needed here. Just parse a
type name in parentheses (and thus you don't need to handle storage class
specifiers that might appear for a compound literal either, for example).
(And if the feature did end up supporting expressions as well, then
refactoring things so that the logic for "unary expression or
parenthesized type name" doesn't need replicating for all of
sizeof/countof, alignof (as an extension) and maxof/minof would be a good
idea, though maybe complicated by the various evaluation and error
handling logic.)
--
Joseph S. Myers
[email protected]
