bug#63556: 29.0.90; Use of _Generic breaks Emacs build on GCC <4.9

[Paul Eggert]

On 5/18/23 04:11, Po Lu wrote:

> Would anyone mind if I made configure check for that?

There would be problems in doing that. GCC and Clang both have problems 
with _Generic (suggested by this original bug report), and some of them 
have to do with which warnings you've enabled. The configure-time check 
would run without the warnings whereas the build would run with them.

This is partly why intprops-internal.h has given up on _Generic with GCC 
and Clang; see its ifdefs. I assumed that these problems didn't apply to 
their simple use in elogb, but apparently I assumed incorrectly.


> Btw, couldn't the macro be defined to something along the lines of this
> when the compiler only supports C99?
>
>    #define foo(expression)                        \
>      (sizeof (expression) == sizeof (signed char)  \

That wouldn't work on unusual platforms that have padding bits. Of 
course we could simply refuse to port to targets like that; still, I'd 
be reluctant to do that just for this little issue.


> The problem of using `long long' on systems with less than 64
> significant bits in their words.  I don't think that's a particularly
> good idea -- once such uses pile up, Emacs will gradually become slower
> and slower on such systems.

I couldn't measure the slowdown in this case. Until it's measurable I 
wouldn't worry about it. To some extent I put in that _Generic 
originally more as documentation than as an actual attempt to speed 
things up.