https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63831
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #16 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
That looks wrong. You really can't keep the __has_attribute__(...) in the
preprocessed stream, it should either be not recognized at all (for languages
other than C/C++ - like for CLK_ASM or when the callback is NULL (think about
Fortran preprocessing; that is done as if it was C89, but of course the
callback is not defined and expanding __has_attribute__ doesn't make much sense
in that case).
And, for C/C++ preprocessing without compilation, i.e. c-ppoutput.c, you should
IMHO just do very similar if not the same thing as in c-lex.c. If something
needs initializing and has not been initialized for the attribute lookup, just
initialize it lazily.