Fix Symbol$CompletionFailure crash when building CFG

[msridhar]

Fixes #6396

It turns out the crash occurred inside the javax.lang.model.util.Elements#getAllMembers method, which was called to find the implementation of close() to invoke for a resource in a try-with-resources block. I don't know the exact cause, but it looks like that method tries to resolve all members in all superclasses of the argument, which is leading to a crash with this specific third-party jar file. I suspect the root cause of the crash is somewhere else. (Maybe in the Scala compiler? The jar seems to include Scala-generated classes.) But, as a workaround, it seems that if we search for the method more directly, only looking for members with the name "close", the crash is avoided. I'm not 100% thrilled with this fix, but it's the best I can come up with at this point to avoid this crash.

[mernst]

Thanks!

[mernst]

@msridhar Something in this pull request significantly degrades run-time performance. I've run CI 3 times and wasn't able to complete within the timeout any of those times. Could you look into improving that? Maybe use the original code, but catch the exception and if the exception is caught, use your new code.

[msridhar]

@mernst I'm puzzled by the CI failure. It seems the only job failing is junit_jdk17, but junit_jdk11 and junit_jdk21 are both fine. And, I can run ./gradlew :checker:test on JDK 17 on my machine and I see no degradation in performance. Can I somehow increase logging on CI for the junit_jdk17 job to get a better idea of what could be happening, maybe by adding --info to the Gradle command? I can try poking at it by using the old code and catching the exception if it fails, but even if that works somehow (I am skeptical it will), I'd like to understand the root cause.

[mernst]

@msridhar I didn't see anything obviously non-performant in your code. Feel free to add any logging you want to this pull request; we can remove or comment it out before merging. I just triggered a new build (as opposed to re-running the failed build).

[mernst]

I'm seeing the failure on the master branch of my fork (https://github.com/mernst/checker-framework/commits/master/). I'm running some other tests now.

[smillst]

I think this call is potentially costly, which is why the Checker Framework doesn't use Names very often. So you could try making it a field.

[msridhar]

@mernst @smillst could I get a re-review on this one? No significant changes since the last approval