On Tue, Jul 20, 2021 at 01:47:01PM -0600, Martin Sebor wrote:
> > Addressing that is for another day.
>
> David Malcolm (CC'd) has a patch attached to pr63326 to issue
> a warning to point out that #pragmas are treated as statements
> that would help prevent this type of a bug. David, do you still
> plan to submit it?
That patch doesn't look correct.
c_parser_pragma and cp_parser_pragma is already told if it appears
in a context for which treating the pragma as standalone statement changes
the behavior and in contexts where it doesn't - pragma_stmt stands
for the problematic ones, pragma_compound for the correct ones (there are
other values for namespace scope, class scope etc.).
OpenMP/OpenACC pragmas shouldn't be touched, those already do the right
thing the standard asks for, for the remaining ones there should be a
warning for the pragma_stmt cases.
Jakub
