Hi all,
Forgive the top post and please see below/previous messages for
previous updates.
TL;DR: I plan to merge mesa-updates into master today-ish (well,
tomorrow for me at this point).
I've been checking in with Efraim who's been very helpful at trying to
nudge along substitute coverage on non-x86_64 platforms. Unfortunately
looks like we have plateaued a bit on, e.g., aarch64. We haven't been
getting stats from QA for this round, and Berlin looks good for what
it covers (x86) but other architectures are down from what we can
tell.
I don't think there are any fundamental failures at this point but
just lots of "missing derivation" errors (I've restarted so many
manually for x86_64/i686) and builds not completing without restarts.
Or unknown reasons. Given the few weeks I've given this and the risk
of just perpetually doing rebuilds to keep catching up (with then more
updates to push) I think it would be best to merge to master. Mesa and
other bits will continue to move forward as well, so I think it is
time so we can be somewhat timely.
I'd rather not without complete substitute coverage, but given recent
build farm difficulties, and the tools we do have for users (pinning,
weather checks, etc.) I think it is best to call this branch so we can
move on. Gnome has some updates that will need (re)building as well as
trying to move forward with core-updates now too.
This is a case where having some better sense of our users and actual
substitute needs/wants would be helpful (yes, Guix survey!) as well as
recognizing our current infrastructure limits. Here's another vote for
prioritizing infrastructure and making sure QA lives and expands.
Feel free to object to this merge timing, though with the relative
silence in each previous message I take it I can make a call here.
Thanks everyone and hope 2024 is off to a good start! Enjoy the new
mesa with curl and xwayland security updates (no new grafts!).
John
On Thu, Jan 04, 2024 at 12:09 AM, John Kehayias wrote:
Toggle quote (82 lines)
> Hi Efraim and guix-devel
>
> On Mon, Dec 25, 2023 at 08:44 AM, Efraim Flashner wrote:
>
>> On Fri, Dec 22, 2023 at 09:19:27AM +0200, Efraim Flashner wrote:
>>> On Thu, Dec 21, 2023 at 09:18:50PM +0000, John Kehayias wrote:
>>> > Hi all,
>>> >
>>> > On Mon, Dec 18, 2023 at 12:57 AM, John Kehayias wrote:
>>> >
> [snip]
>>> >
>>> > I haven't seen QA process this branch, so I'm just going with what I
>>> > see on Berlin. From the branches overview it shows about 61% last I
>>> > saw, compared to 72% for master. Unfortunately, non x86 architectures
>>> > are usually better covered by Bordeaux, but I don't know where to get
>>> > a sense of that coverage. For what it is worth, Efraim has manually
>>> > built xorgproto and mesa at least on powerpc64le, riscv64, without
>>> > issues.
>>>
>>> I had berlin build for powerpc64le and that went without any problems.
>>> Locally I built for riscv64 and powerpc and those both built fine. I
>>> ran into an issue locally with curl on aarch64 and test 1477(?) which is
>>> weird since it's supposed to be skipped but I'm sending it through
>>> again. Haven't started armhf yet.
>>>
>>> > Coverage on x86_64 and i686 seems good from what I can tell. I also
>>> > don't think there are any other branches ready to merge, and would
>>> > like to give them time to rebuild once these changes hit.
>>> >
>>> > Any thoughts on when to merge?
>>> >
>>> > Thanks everyone!
>>> > John
>>
>
> Coming back to this point, seems Berlin is doing better with building
> but I don't see mesa-updates on QA so I'm not sure about non
> x86_64/i686-linux coverage. Anyone have any thoughts?
>
> I don't know that I've seen real new failures, as still lots of
> "missing derivation" or other transient issues that resolve on forcing
> a rebuild.
>
> I don't want to merge to master and have issues with substitute
> coverage, but do have to call it at some point or will end up keep
> scheduling/waiting for rebuilds to happen anyway.
>
> Thoughts?
>
>> I've been having trouble with curl on aarch64 again. Looking at this
>> snippet from the build log:
>>
>> test 1477...[Verify that error codes in headers and libcurl-errors.3 are in sync]
>>
>> 1477: stdout FAILED:
>> --- log/1/check-expected 2023-12-22 10:53:51.658667071 +0000
>> +++ log/1/check-generated 2023-12-22 10:53:51.658667071 +0000
>> @@ -1 +0,0 @@
>> -Result[LF]
>>
>> - abort tests
>> test 1475...[-f and 416 with Content-Range: */size]
>> --pd---e--- OK (1247 out of 1472, remaining: 00:45, took 5.310s, duration: 04:11)
>> test 1474...[HTTP PUT with Expect: 100-continue and 417 response during upload]
>> --pd---e--- OK (1246 out of 1472, remaining: 00:48, took 22.794s, duration: 04:29)
>> Warning: test1474 result is ignored, but passed!
>> ...
>> TESTFAIL: These test cases failed: 1477
>>
>> It looks like 1474 is passing locally and the ~1474 is telling the test
>> suite to ignore the result. If that's how ~<number> is interpreted then
>> I'd suggest that 1477 is failing hard enough that it's taking the test
>> suite with it, not merely ignoring the result. I'll continue poking it
>> but right now I'm starting to like the hurd plan of disabling the test
>> instead of merely ignoring the result.
>
> Thanks for looking at this Efraim. Looks like a good chunk of the curl
> rebuilds did get through, did it look okay on aarch64 and anywhere
> else you checked?
>
> John
