[committed] c: Default to -std=gnu23

[Joseph Myers]

On Fri, 22 Nov 2024, Sam James wrote:

> > Some estimate of the build failure would also be helpful.  Is it the
> > same as, e.g. the GCC 12 to GCC 13 update, or is it igher?
> 
> It's pretty large so far. Between 12 and 13, the main issues were:
> libstdc++ transitive include changes and a small number of -Werror
> breakages.

What's the comparison to the GCC 13-to-14 update like?  It's GCC 14 that 
made implicit int, implicit function declarations and some other such 
things into errors by default.  I think of both those things and 
unprototyped function declarations (and old-style function definitions, 
and defining your own bool rather than using _Bool / <stdbool.h>) as 
indicative of old code that hasn't been very well maintained over the 
years.

Both the GCC 14 changes and this change required large numbers of changes 
to the GCC testsuite, but many of those would have been because automatic 
test reduction tools are liable to reduce tests to the minimum number of 
tokens accepted by the compiler, which tended to result in unprototyped 
function declarations, implicit int and implicit function declarations 
when the compiler accepted such things, rather than indicating that the 
original code from a bug report was actually using those features before 
reduction.

(When fixing the testsuite I tended to err on the side of using 
-std=gnu17, not because the use of unprototyped functions was expected to 
be in any way essential for most tests, but because it's harder to figure 
out years later whether in fact the test cares about functions being 
unprototyped or not.  That is not of course the best fix for most code 
outside the compiler testsuite - actually using correct prototypes would 
be better for most application code that now gets errors for unprototyped 
declarations.)

-- 
Joseph S. Myers
josmyers@redhat.com