-
Notifications
You must be signed in to change notification settings - Fork 4.9k
Update CoreClr, CoreFx, External, ProjectNTfs, Standard to beta-25019-02, beta-25018-01, beta-25016-01, beta-25017-00, beta-25017-01, respectively (master) #16252
Conversation
6197358
to
be2cd1a
Compare
be2cd1a
to
0e33c48
Compare
0e33c48
to
27ab987
Compare
@jkotas, we're seeing random failures and seg faults across various Unixes when updating to the new CoreCLR, all appear to be stemming from the System.Collections.Tests tests: |
cc: @danmosemsft |
@stephentoub do you repro these locally? If not I can try. |
I haven't tried. |
OK, will do. |
Oop, I see it doesn't repro on Ubuntu, which is what I have (and indeed it doesn't). |
@ianhays since it's collections -- do you have a mac to try this locally? |
Sure do. I'll see if I can repro it. |
This probably has an obvious answer, but how does one map from the nuget versions back to commit ranges? |
The only way I know is
This is obviously painful, and you have to repeat it to get a range. Perhaps @stephentoub or @ianhays have a better way - there really ought to be a better way. |
In this case it gives me a range from |
Or maybe @MattGal knows a better way. |
Looks like the 15-02 version passed (see #16168) , so perhaps focus on the range after that? |
It didn't, actually. That accidentally got merged during the Jenkins outage when no CI legs were being run. I reverted it because I was then seeing this failure sporadically across the repo. Which itself suggests that instead of focusing on the range after that, you could likely focus on the range before that :) |
Ah -- so the last good version was indeed 25014-04... |
27ab987
to
e17f2f7
Compare
@danmosemsft I do not know a better way to do this than checking the packages themselves. The only other place I know that we really have this information is in the build definitions output themselves, which takes looking in a few different places on VSO to infer. |
No luck yet. Ran them in a loop about 50 times on OSX and didn't get any failures from this PR. |
e17f2f7
to
44195aa
Compare
You may want to try checked build of CoreCLR; and if it still does not repro - also run it with GC stress on. It is likely a regression from one of the recent JIT changes. |
I saw the older CI build fail on this on Windows and I was able to repro this locally after 4 tries:
Will see if I can pin this down. |
Guessing maybe the "update" job on CoreCLR has to go through first? Will keep an eye on it. |
Thanks, Andy. Once a new coreclr build is available, this PR will be updated to it automatically. Looks like the last one was just before the revert. |
There are arm64 build failures in CoreCLR, might that block the update process? |
The 25019-01 build of CoreCLR just finished and should have the fix. |
0635e21
to
ae6eea9
Compare
XML tests now. This is not showing up in the regular CI or Jenkins.
|
We picked up more than just the jit fix with 25019-01. Suspect this latest set of failures might be dotnet/coreclr#9187. @benaadams does this seem possible? |
@AndyAyersMS looks like it could be, it goes through |
Making revert PR 😢 |
Revert PR dotnet/coreclr#9671 |
Will investigate seperately |
Thanks @benaadams . Perhaps our coverage of the Encoder code is insufficient (as opposed to indirect coverage through XML tests) |
There's a follow up issue https://github.com/dotnet/coreclr/issues/9668 for matching the implementations with corefxlabs which would require a change to how the fallbacks worked; so it looks like the tests in this area would need to be improved before that was attempted. |
I x-ref'd 9197 initially, not 9187, sorry if that confused anyone. Fixed with edit. |
Hopefully CoreCLR is back on track now with dotnet/coreclr#9197 reverted via dotnet/coreclr#9671. If I understand the official build schedule, the -02 build should kick off just after midnight PST, so with any luck we will have a successful update around 3AM tomrrow. |
…-02, beta-25018-01, beta-25016-01, beta-25017-00, beta-25017-01, respectively
ae6eea9
to
a3864d4
Compare
Seems less likely that this latest failure is a CoreCLR issue. |
Old failing log here, let's see if it repros. @dotnet-bot test Innerloop OSX Release Build and Test |
That failure looks like: |
Thanks, all. I'm assuming this is good now. |
…ependencies Update CoreClr, CoreFx, External, ProjectNTfs, Standard to beta-25019-02, beta-25018-01, beta-25016-01, beta-25017-00, beta-25017-01, respectively (master) Commit migrated from dotnet/corefx@1c9dd7d
No description provided.