You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As part of investigating dotnet/arcade#10311, I realized that there seems to be some kind of change in MSBuild, where build tasks that call Process.Start fail due to non-zero exit code. Digging in, all the errors were from scripts being run getting "broken pipe" type errors on Linux. I swapped out the entire "bad" .NET Core runtime used into the "good" SDK and did not get a repro; this leads me to believe (along with the only repro beign in a build task) that this is an MSBuild bug. I've leveraged a custom version of the .NET Arcade task that hit the original issue as a repro.
Steps to Reproduce
(Assuming the files from repro.zip are copied to c:\scratch\hosting-bug-repro)
To see the expected behavior:
Happy SDK 7.0.100-rc.1.22377.7 does not repro the problem (Version 7.0.22.36704 @Commit: aafa91036e1efd3d4dcb67eeb261cb6d8f774685)
Happy SDK has MSBuild version 17.4.0-preview-22366-04+d2871ca13
docker run -v c:/scratch/hosting-bug-repro:/repro -it mcr.microsoft.com/dotnet-buildtools/prereqs:ubuntu-18.04-20220808192747-20889cf
cd /repro && ./dotnet-install.sh -version 7.0.100-rc.1.22377.7
/root/.dotnet/dotnet build ./repro.proj
Sad SDK 7.0.100-rc.1.22379.1 repros the problem (Version 7.0.22.37802 @Commit: 97749c034e3a94c03c2ebfef94fa6aa4b257488b)
Sad sdk has MSBuild version 17.4.0-preview-22378-04+827c1bf9c
To reproduce the bug:
docker run -v c:/scratch/hosting-bug-repro:/repro -it mcr.microsoft.com/dotnet-buildtools/prereqs:ubuntu-18.04-20220808192747-20889cf
cd /repro && ./dotnet-install.sh -version 7.0.100-rc.1.22379.1
@MattGal can you try rerunning your impacted build with the environment variable DOTNET_CLI_DO_NOT_USE_MSBUILD_SERVER set to true? There was a bug on linux in the new MSBuild server functionality that was addressed in #7860 (though that may not have flowed all the way yet).
@MattGal can you try rerunning your impacted build with the environment variable DOTNET_CLI_DO_NOT_USE_MSBUILD_SERVER set to true? There was a bug on linux in the new MSBuild server functionality that was addressed in #7860 (though that may not have flowed all the way yet).
Yup, setting this true/false works like an on/off switch for the repro... feel free to close this as dupe of that.
Great! Well, not great, but I'm glad you're unblocked. Can you make a note to try reverting that change in like a week? We'd like the MSBuild Server daemon to be the default as much as possible once the kinks are worked out.
Can you make a note to try reverting that change in like a week?
In this case the changes never got merged to take a newer SDK, so given there's a subtle msbuild bug in this particular SDK I think all we have to do is wait for the fixed version.
Issue Description
As part of investigating dotnet/arcade#10311, I realized that there seems to be some kind of change in MSBuild, where build tasks that call Process.Start fail due to non-zero exit code. Digging in, all the errors were from scripts being run getting "broken pipe" type errors on Linux. I swapped out the entire "bad" .NET Core runtime used into the "good" SDK and did not get a repro; this leads me to believe (along with the only repro beign in a build task) that this is an MSBuild bug. I've leveraged a custom version of the .NET Arcade task that hit the original issue as a repro.
Steps to Reproduce
(Assuming the files from repro.zip are copied to c:\scratch\hosting-bug-repro)
To see the expected behavior:
Happy SDK 7.0.100-rc.1.22377.7 does not repro the problem (Version 7.0.22.36704 @Commit: aafa91036e1efd3d4dcb67eeb261cb6d8f774685)
Happy SDK has MSBuild version 17.4.0-preview-22366-04+d2871ca13
Sad SDK 7.0.100-rc.1.22379.1 repros the problem (Version 7.0.22.37802 @Commit: 97749c034e3a94c03c2ebfef94fa6aa4b257488b)
Sad sdk has MSBuild version 17.4.0-preview-22378-04+827c1bf9c
To reproduce the bug:
repro.zip
Expected Behavior
Build succeeds as many times as run
Actual Behavior
errors like this:
The text was updated successfully, but these errors were encountered: