Unit & Integration Testing utilities

[vorburger]

Unit & Integration testing ADK related code could perhaps benefit from some simple common shared utilities?

This isn't about some sort of full blown "evaluation" toolkit, but just along the lines of "I want to be able to have to write less code to write a simple (integration) test for something like the https://google.github.io/adk-docs/get-started/quickstart/#create-multitoolagentjava".

core/src/test/java/com/google/adk has some utility classes a bit like that. But they are not part of a published Maven artifact, and intended for testing the ADK itself, not for its users.

The RunLoop which is used in the LangChain4jIntegrationTest (from @glaforge) is a similar kind of basically generic utility.

I have started to create more like that here, which can be used e.g. like this.

Is there perhaps any interest in accepting contributions for "upstreaming" (a subset of) that into this repo?

Or maybe I should publish my own "ADK testing library", and it could be linked to?

I'm interested in feedback and thoughts from other ADK users about this.

[xinhuagu]

IMHO, using utilities to make Java test code more concise is always a good move. Compared to the adk-python code, the Java version does feel a bit more verbose...

[vorburger]

core/src/test/java/com/google/adk has some utility classes a bit like that.

More specifically, moving the 3 (and 2 inner) classes in com.google.adk.testing, or just some of, maybe even start with e.g. the TestLlm which is very similar to [my] MockModel), into a separate //testlib (with com.google.adk:google-adk-testlib GAV) [inspired by guava-testlib] is IMHO likely a sensible small start which can then be expanded upon for this.

I'll see if I can make time for this one day.

[vorburger]

Note to my future self: ADK's TestLlm as-is has as "design flaw" which makes impossible to use its getLastRequest() in its Supplier<Flowable<LlmResponse>> responsesSupplier... I've hacked around this in https://github.com/enola-dev/enola/pull/1578/files, but perhaps later as part of this issue we can create a clean solution (I'm not quite sure yet how though).