On Thu, Jul 21, 2022 at 10:39:55PM +0200, Fabio Valentini wrote:
On Wed, Jul 20, 2022 at 4:56 PM Dan Horák <dan(a)danny.cz>
> On Wed, 20 Jul 2022 16:48:02 +0200
> Miro Hrončok <mhroncok(a)redhat.com> wrote:
> > On 15. 07. 22 23:48, Ben Cotton wrote:
> > > == Summary ==
> > > Move the minimal architecture level for IBM Z hardware (the s390x
> > > architecture) to z13 to benefit from the new features and better
> > > performance in the newer CPU.
> > I suppose, considering the mass rebuild is today, that this change only makes
> > sense for Fedora 38. Or do we really want to change this without rebuilding
> > everything?
> it still makes sense to me, we can change default in the kernel, next
> glibc build will pick it, so the majority of the perf improvements will
> be available. And if something should go wrong (although I don't expect
> that), then the individual package builds will get the change and it
> will be easier to fix them incrementally, rather having a big(er) batch
> from the rebuild at once.
I'd argue that this should've been submitted as a System-Wide Change.
If I understand correctly, this change would affect every package that
is built on s390x?
As such, it would've been way too late for Fedora 37, as it should
also have been ready before the mass rebuild.
And I'd also argue that not having this change ready before the mass
rebuild, and only pushing selective builds is not a good idea.
If there is fallout from this change, it would be better to catch it
at or before the mass rebuild.
Otherwise, if the switch is made *after*, potential breakage could
well happen over the whole lifetime of Fedora 37 (i.e. after GA and
until EOL), as more and more packages are rebuilt, which would be a
completely unpredictable source of issues.
Note, in terms of software compatibility with newer CPU baseline,
this has already been widely tested given that RHEL-8 built with
the z13 baseline and RHEL-9 built with z14 baseline.
I do agree though that this feels like the kind of change that
makes sense todo all in one go at a mass rebuild, rather than
Given Fedora has lagged waaaaaay behind RHEL in its s390x baseline
for years, I'm not seeing why there's a need to rush this in Fedora
late in the F37 cycle, rather than co-ordinating it for the F38
release to leverage a mass rebuild.