@Icicle from supertypes are not persisted through state

[frankiesardo]

@Icicle from supertypes are not persisted through state

[fluxtah]

Sorry my bad :( my fault the generator did not generate the super type call first time round

[pushbit]

It appears that this issue has returned in 2.3.2. My SubType$$Icicle methods are called, but the SuperType$$Icicle methods are not.

[frankiesardo]

That sounds weird. For me both the sample project with superclass and the tests are fine. Can you copy/paste the generated classes? SubType$$Icicle should have a call to SuperType$$Icicle in both save and restore instance state.
Gradle > Clean your project may also help. Maybe you were using another version and the generated classes have not been updated.

[pushbit]

I have cleaned the project and the SubType$$Icicle class is generated without calls to the SuperType$$Icicle methods. Could it be because SuperType is in an Android library project and not in the same project/package as SubType? I will try to verify this theory later today or tomorrow when I have time.

[pushbit]

This does appear to be the issue. When SuperType and SubType are in the same project, SubType$$Icicle is generated with calls to SuperType$$Icicle. But if SuperType is moved to a library project (which SubType's project uses), then SubType$$Icicle no longer contains the calls to SuperType$$Icicle.

I have tried adding both the library jar and src/ to the annotation processor factory path, but it did not help. Any idea why the processor is not seeing SuperType in the library project?

[frankiesardo]

I see. The problem is that only the annotations of one project are processed at time, so the Processor can't tell if the class inherits from an annotated parents.

A solution could be to scan all parent's fields and see if there is an @Icicle annotation and thus call the generated file, but it seems that getAnnotationMirrors is not working properly across projects. Plus, this will considerably slow down the build time.

I'll have a look at it. I'm reopening the issue in the meanwhile.

[frankiesardo]

@pushbit I just pushed a 2.3.2-SNAPSHOT on Maven. This is a feature that it makes sense to support, regardless of the drop in performance. Early benchmarks shows that is at least 2x slower to walk the hierarchy asking for every field if it has an @Icicle annotation, but seems to be no other way to retrieve this information from another project. Persisting something on the filesystem can easily get out of hand.

Still, I can probably cut some slow nested loop by keeping track of unsuccessful paths.

Let me know if this fixes your issue.

[pushbit]

2.3.3-SNAPSHOT fixes the issue and I don't notice any build slowdown, thanks!

[frankiesardo]

Glad it helped. We're still talking about an order of hundreds of milliseconds, so the build should still be pretty fast. I added a log at the end of the processing that prints how long it took, if anybody notice a slow down they can share their timing.