APICompat failures on Linux

[jkoritzinsky]

APICompat is causing failures in PRs that don't change library code on Linux.

/__w/1/s/src/libraries/shims/ApiCompat.proj(116,5): error : CannotRemoveAttribute : Attribute 'System.ComponentModel.TypeConverterAttribute' exists on 'System.ComponentModel.MarshalByValueComponent' in the contract but not the implementation.

Example: #40404

[ghost]

Tagging subscribers to this area: @safern, @ViktorHofer
See info in area-owners.md if you want to be subscribed.

[safern]

Finally I was able to get a repro on my private branch that is publishing the artifacts in question, now to investigate: 🎉

https://dnceng.visualstudio.com/_apis/resources/Containers/4784658/libraries_bin_Linux_arm64_Release?itemPath=libraries_bin_Linux_arm64_Release%2Flibraries_bin_Linux_arm64_Release.tar.gz

[jkotas]

@safern @ericstj @danmosemsft This is taking down a large number of PRs. Can we prioritize getting it fixed?

[am11]

Made a similar fix for SuppressGCTransitionAttribute, by listing it in eng/DefaultGenApiDocIds.txt. Not sure if it is applicable to TypeConverterAttribute.

[safern]

@safern @ericstj @danmosemsft This is taking down a large number of PRs. Can we prioritize getting it fixed?

This is my top priority, I added logging to APICompat and ran multiple builds from a branch using APICompat... now looking at the logs and trying to find the reason behind it. It seems like a race condition... if I can't fix this today I can baseline the errors temporarily until I found the fix.

Made a similar fix for SuppressGCTransitionAttribute, by listing it in eng/DefaultGenApiDocIds.txt. Not sure if it is applicable to TypeConverterAttribute.

We don't want to do this as that means that we don't check those attributes in between ref and implementation, so we would loose coverage for individual libraries.

[danmoseley]

. if I can't fix this today I can baseline the errors temporarily until I found the fix.

@safern what do you tihnk about putting up a PR to do this while you work on it? I think you could potentially direct-submit it if it's just an edit to the baseline and you build locally to check, since the risk is outweighed by the cost of these breaks.

[danmoseley]

As a general rule if something hits CI it's good to immediately suppress the problem if we can, while we work on it. (?)

[safern]

Yeah I guess that makes sense.

[ViktorHofer]

@safern what's the status on this?

[Anipik]

I didnt see this one reproing any more on official builds.

[ViktorHofer]

Ok so let's close this for now. Thanks.

[safern]

This is not reproing because I baselined the errors. We should still fix this. Will re-open, but I will mvoe to 6.0.0 and remove blocking-clean-ci.

[safern]

The errors are no longer baselined since we moved APICompat to run against 5.0.0 and we haven't gotten a repro (really weird). Closing.