Re: glibc test time regression after gawk update

Hi.

So, first, I gather that the fix for gensub did the trick and
that the printf test suite runs correctly, but it's just somewhat slower?

What happens if you use `configure --enable-O3' to build gawk? Do the times
get better?

For the moment, I am not going to hold up the next release.  If you
can tell us how to run the tests ourselves, I'll ask Mike Haertel
to do some profiling and take a look at the performance issue.

Thanks,

Arnold

"H.J. Lu" <[email protected]> wrote:

> On Fedora 44/x86-64 compiled with -O3,
>
> 1.  With gawk 5.4:
>
> commit 2a5799a85c452a903cc334cb90a8e5abcd84faaf (HEAD ->
> gawk-5.4-stable, origin/gawk-5.4-stable)
> Author: Arnold D. Robbins <[email protected]>
> Date:   Sun Jul 5 11:44:01 2026 +0300
>
>     Update generated files.
>
> $ time make test t=stdio-common/tst-printf-format-f-ldouble-e
> ...
> real 0m15.996s
> user 0m18.359s
> sys 0m0.335s
>
> 2. With gawk 5.3:
>
> commit 16ba2113002eedad16b944cac3439eb309648144 (HEAD ->
> gawk-5.3-stable, origin/gawk-5.3-stable)
> Author: Andrew J. Schorr <aschorr-wEaNgGkE20pmvX9EvAWxPKAgWQW2ebYtAL8bYrjMMd8@public.gmane.org>
> Date:   Thu Jul 24 10:47:40 2025 -0400
>
>     Disable lint warning when a filename is numeric but came from user input.
>
> $ time make test t=stdio-common/tst-printf-format-f-ldouble-e
> ...
> real 0m9.955s
> user 0m12.328s
> sys 0m0.304s
>
> gawk 5.4 is much slower.