Re: FYI, 2.73 slower than 2.72

On Thu, Mar 19, 2026, at 10:41 AM, Akim Demaille wrote:
>> Le 19 mars 2026 =C3=A0 15:36, Paul Eggert <[email protected]> a =C3=A9=
crit :
>> On 2026-03-18 22:01, Nick Bowler wrote:
>>> On Wed, Mar 18, 2026 at 12:57:53PM -0400, Zack Weinberg wrote:
>>>> Do you happen to remember why we concatenate confdefs.h and the
>>>> test program
>>> A long time ago C compilations did use #include "confdefs.h" but it
>>> was changed to resemble the current method in version 2.55 (2002).
>>> No explicit rationale is given but it was probably done this way to
>>> improve the logging around compilation failures.
>>
>> Thanks, then let's ask Akim.
> Wow...  That's not recent :)
>
> I believe the answer is in
> https://lists.gnu.org/archive/html/autoconf-patches/2002-10/msg00072.h=
tml.

So, in short, yes, this was done so that "failed program" dumps in
config.log would be more informative.

I don't think it makes sense to change anything without a bunch more
benchmarking, and certainly not for 2.73, but if we do conclude that
going back to #include "confdefs.h" would make sense, we could retain
the improved logging by dumping out _both_ confdefs.h and the test
program when a compilation fails.

zw