Use VirtualField for associating netty listener with wrapper

[laurit]

Probably resolves #5265
To support removing netty listeners we need to be able to find the wrapper that we register instead of the original listener from the original listener. Currently we use a static weak cache for it. Removal of the elements from this cache depends on the gc, which might confuse users who inspect the heap dump and make them suspect that there is a memory leak.

[trask]

I didn't follow the connection to skipping transforming lambdas of injected helper classes?

[laurit]

@trask I should have added a comment for this. As far as I understand we don't transform helper classes, so not transforming their lambdas should really be the expected behaviour. Transforming lambdas of helper classes can fail because some bytebuddy matcher tries to resolved the class file of containing class of lambda, which does not exist, because it is an injected class. We make tests fail when there is a transformation failure. Adding a virtual field for GenericFutureListener causes such failures.

[trask]

oh, i c, and non-lambda injected classes are skipped already because they live under "io.opentelemetry.javaagent.shaded" which we skip in GlobalIgnoredTypesConfigurer

would it work to move the isHelperClass check up to GlobalIgnoredTypesConfigurer?

[laurit]

oh, i c, and non-lambda injected classes are skipped already because they live under "io.opentelemetry.javaagent.shaded" which we skip in GlobalIgnoredTypesConfigurer

would it work to move the isHelperClass check up to GlobalIgnoredTypesConfigurer?

I thought it was because the injection happens during transformation and transformation isn't re-entered. Only helper classes from library modules have "shaded".
Using GlobalIgnoredTypesConfigurer would require some kind of heuristics as you need to configure it before classes are actually injected. We would need to use something like class name starts with "io.opentelemetry.javaagent." and ends with "$Lambda*"

[trask]

oh of course, I forgot that GlobalIgnoredTypesConfigurer is just used to build up the trie 👍

[trask]

oh yes, and this:

I thought it was because the injection happens during transformation and transformation isn't re-entered

and I missed that the check isn't whether the lambda class itself is a helper class, but rather it's whether the lambda's targetClass is a helper class.

btw, what is the targetClass? is it the "outer" class where the lambda lives? or the target interface of the lambda?

thx 😅

[laurit]

btw, what is the targetClass? is it the "outer" class where the lambda lives? or the target interface of the lambda?

thx 😅

targetClass is the class that declares the lambda. Not the best name. It has the same value as https://github.com/openjdk/jdk11/blob/37115c8ea4aff13a8148ee2b8832b20888a5d880/src/java.base/share/classes/java/lang/invoke/AbstractValidatingLambdaMetafactory.java#L54