-
Notifications
You must be signed in to change notification settings - Fork 10.5k
[Windows] Support Nightly Toolchains for main and release/5.10 #69484
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Comments
Currently blocked on #69483 |
Currently blocked on getting this PR merged #69555 |
Job created for main branch: https://ci-external.swift.org/job/swift-main-windows-toolchain/20/ |
Job failed to find
Found the issue, and started new build: https://ci-external.swift.org/job/swift-main-windows-toolchain/21/console |
@compnerd We are seeing following test failing, it might be config issue. However, if you can look into this would be helpful. Thanks! https://ci-external.swift.org/job/swift-main-windows-toolchain/21/console
|
@shahmishal super exciting! Just looking through a couple of things stand out:
With the switch to build.ps1, this shouldn't be needed (build.ps1 will properly setup the VS Dev environment to support building the SDK for all the architectures). Something that might help simplify the Jenkins configuration.
Odd, I'm not sure where this is coming from. We should investigate this later.
This looks right ... and if this is failing, I doubt it's a configuration issue and more likely this is an issue in build.ps1. I'll need a little bit to dig into this a bit further to try to understand what is going on. |
There seem to be a few failures due to |
If it helps, here is the build step:
|
I have a fix for this. |
It appears that a number of the failures are referencing I wonder if there was something that we had done before that results in the build tree being linked or something? |
Hmm, I wonder if it is the |
Bad Run:
Good Run:
This seems to be the diff: - lit.py: swift\test\lit.cfg:289: note: using swift: t:\1\bin\swift.exe
- lit.py: swift\test\lit.cfg:385: note: Using resource dir: t:\1\lib\swift
+ lit.py: swift\test\lit.cfg:385: note: Using resource dir: ..\lib\swift
- lit.py: swift\test\lit.cfg:2253: note: Using platform module dir: t:\1\lib\swift\windows
+ lit.py: swift\test\lit.cfg:2253: note: Using platform module dir: ..\lib\swift\windows This does match the majority of the failures. Now the question is, why is the resource dir and platform module dir path different? |
The missing
along with https://github.com/apple/swift/blob/main/test/lit.cfg#L273-L298 seems to indicate that |
@compnerd Let me know if I missed setting an env |
@shahmishal I don't see off hand any missing environment variables. The only ones that should be concerning are the ones that determine the flags (which repositories to test, whether to package or not, build type), but those seem to be setup. I would say that removing the Just to double check my assumption - the image is identical to the PR testing one right? |
Hmm ... @tristanlabelle pointed out that I misread the code: swift/test/lit.cfg at main · apple/swift It seems to imply that there is a |
@compnerd Can you try running this locally?
|
@shahmishal everything now makes sense - there is an environment variable |
this is used to pick up the correct SHA, what is the best way to unset variable? |
Depends on the shell. With |
Going to try |
New build started: https://ci-external.swift.org/job/swift-main-windows-toolchain/25/console |
@shahmishal if there is a trailing space, I think it might fail as it will set it to the string containing a space, there should be nothing after the |
The trailing space is setting the string to " " rather than deleting the environment variable. |
main branch built: https://ci-external.swift.org/job/swift-main-windows-toolchain/26/ |
Awesome! The 5.10 cherry-pick is at #69608 |
release/5.10 job created: https://ci-external.swift.org/job/swift-5.10-windows-toolchain/ |
I thought this would delete the virtual drive:
|
Yes, that removes the "virtual drive", but does not remove the contents that were mapped there. |
I also have
|
Interesting - and the download doesn't say that the toolchain installer is invalid. So, it could be that it is failing to pass the signature check :/ I don't think that I have a way to diagnose this without actually interacting with the host. |
Can we add verbose mode to the signature check? |
msiexec_failure.log
|
Thats the current script, let me know if I missed something |
I was thinking of adding the |
I wonder if the builders are running as different users.
|
#69852 adds the logging for future use as well. |
@shahmishal seems that the last re-run succeeded |
@compnerd sent link in private for review for nightly main toolchain. |
@shahmishal - reviewed, seems ready to go to me! |
|
@shahmishal - it seems that the main branch nightlies do not auto-update? There was a snapshot from 11/27, but the other platforms are on 12/7. |
@shahmishal any progress on the auto-updated nightlies? It seems that the Windows target is out of date by a month. |
ping again on this @shahmishal - the Windows snapshot on main is now ~2m out of date and we still do not have a 5.10 release snapshot. This really risks the 5.10 release as we will need extra time to validate the Windows releases. |
@compnerd I am looking into getting this configured, due to other tasks I was not able to get to this. Hopefully, I will be able to focus on this soon. |
main failed:
cc: @compnerd |
Ah, right, we dropped the |
New build started: https://ci-external.swift.org/job/swift-main-windows-toolchain/34/console |
Can we also cherry-pick this to release/5.10? |
Sure, let me look into that! |
Nightly 5.10 released on swift.org - swiftlang/swift-org-website#488 |
Trying to pull in the changes for |
Swift.org now has builds from all the release branches, so this can be closed? |
Windows toolchain needs to be built on ci-external.swift.org for main and release/5.10 branch.
The text was updated successfully, but these errors were encountered: