On 5/7/24 18:40, Andrea Bolognani wrote:
On Tue, May 07, 2024 at 01:58:07PM GMT, Daniel P. Berrangé wrote:
> On Tue, May 07, 2024 at 02:37:05PM +0200, Michal Prívozník wrote:
>> On 5/7/24 12:11, Daniel P. Berrangé wrote:
>>> FYI, I'd really *not* splitting out the removal and addition into
>>> separate jobs. If you remove 20.04 and add 24.04 in the same
>>> commit, then git shows the rename and we get a tiny diff so we
>>> can see the interesting changes.
>>
>> Yeah, and that's how I've started. But then I realized I needed to bump
>> glib version and Ubuntu 20.04 doesn't have it, but without the bump
>> Ubuntu 24.04 build fails. But maybe Fedora and AlmaLinux can be done
>> this way.
>
> I'd suggest just temporarily disabling -Werror warnings, or disabling
> UBSAN. That way you can
>
> * squash the forthcoming warning
> * update all the distros new/old in one go
> * update glib & re-enable the warning
Is this dance really necessary? We obviously care about bisectability
of the code itself, but IMO it's fine if the CI pipeline stops making
sense for a bit in the middle of a series, as long as things are once
again working by the end of it.
Well, I've picked the best of both worlds in my v2: patches that could
be squashed together are, and that left me with just ubuntu where I
needed to update glib in between. But now that I think about it more,
maybe I could just have one patch where Ubuntu 20.04 is replaced with
24.04 WITHOUT ASAN/UBSAN; then bump glib version and finally enable
ASAN/UBSAN on Ubuntu 24.04; Well, I should have thought of that couple
of hours ago, before I sent v2.
Michal