Le 17/10/2026 à 13:03, Jochen Sprickerhof a écrit :
Hi Paul,
* Paul Gevers <[email protected]> [2026-10-17 09:41]:
On Fri, 11 Sep 2026 15:08:21 +0200 =?UTF-8?Q?Fabian_Gr=C3=BCnbichler?=
<[email protected]> wrote:
On Fri, Sep 11, 2026, at 2:22 PM, Sylvestre Ledru wrote:
Le 11/09/2026 à 14:16, Jochen Sprickerhof a écrit :
How can i work around this?
By reverting the change. I am not aware of a way to make PGO >>
reproducible.
this probably warrants some broader discussion..
I think the only way to implement PGO and reproduciblity is to treat
the profiles as allowed generated input to the build (i.e., lift the
restriction of "all pre-generated parts of the upstream source must be
regenerated as part of the build" for them).
So, how to move forward? This feels a bit like the community behind
reproducible builds should paper out a general solution to deal with
PGO in binary distributions, which is unlikely to happen overnight. Do
we agree on that? If so, I think we should let rust-coreutils migrate
now.
Well, the other option is to revert enabling PGO. But as reproducible
builds are still kind of new I would be fine with letting rust-coreutils
migrate for trixie and discuss other options after the release.
Given the important performance wins, I am not interested in disabling PGO.
What should be the procedures to get an exception for this? thanks!
Just an email to release management ?
Which also means that we should ensure "this probably warrants some
broader discussion". I have the impression that's somewhat already
happening, but I've only seen it on IRC. Can somebody link to broader
discussion (if it exists already)?
There is:
https://lists.reproducible-builds.org/pipermail/rb-general/2026-September/004153.html
I proposed a PR
https://github.com/bmwiedemann/theunreproduciblepackage/tree/master/pgo
to document this a bit more :)
S
