Chris Adams wrote:
> Can you stop grinding your axe against a decision you don't agree with?
> You're just "guessing" with zero evidence.
I have also seen zero evidence of the contrary (i.e., that the size change
is *not* significant), which should have been a prerequisite for accepting
the change. All the talk in the FESCo ticket was only about speed, and my
requests for size data have been ignored (not even turned down, just ignored
altogether!), which makes it look to me like somebody has something to hide.
We need to be much stricter on size increases! In Fedora 9 (when the xz
compression for live images was introduced, which made it smaller than
Fedora 7 or 8), the x86_64 KDE Spin was 729272320 bytes. In Fedora 38
Branched, it is now 2418429952 bytes. That is about 3.32 times as much!
> If you look at the bugs, they date back well before the mass rebuild, so
> there's no justification to guess it's the frame pointer change.
I can tell you my justification: I have experience with the TIGCC cross
toolchain. Enabling -fomit-frame-pointer by default was one of the changes
that helped decrease the size of basically all executables there. (It turns
out that the handful extra stack adjustment instructions added up to much
less code overall than all the unnecessary link/unlk instructions for the
frame pointer.) Of course, that was m68k and we are talking about x86_64
here, which is why I was asking for data on Fedora on x86_64. But the guess
is NOT completely out of the blue.
And the analysis of the Python performance regression on Fedora x86_64 also
pointed to a size increase of a function (overflowing some cache line) as
the likely cause. So that, too, is evidence of size increases.
Vít Ondruch wrote:
> Generally, it seems that after mass rebuild, our package set is bigger by
> ~0,36%:
>
> ~~~
>
> Size of upgraded packages: 147.51 GiB
>
> Size change of upgraded packages: 542.39 MiB
> ~~~
So that is at least SOME data. Thanks for that. Though it does not include
the packages that had already been rebuilt between when the frame pointer
change was introduced and the mass rebuild. Several frequently updated
packages (and some that just happened to be updated in that time window) are
in that boat.
> But hard to tell what is the reason. If the GCC 13 or frame pointers or
> something else. I for one would be interested in such analysis.
I do not see what else it would be if not frame pointers.
Kevin Kofler
_______________________________________________
devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: https://lists.fedoraproject.org/archives/list/devel@xxxxxxxxxxxxxxxxxxxxxxx
Do not reply to spam, report it: https://pagure.io/fedora-infrastructure/new_issue
