Race condition in the dependencies of the osgiBundle task

[lefou]

The race is very well analyzed in this issue, so please read that issue completely.

• community_build_a fails intermittently with corrupt scalatest artifacts scala/scala3#11604

The issue was introduces as part of this change

• Use dependencies as JARs for force proper manifests (fixes #60) #64

Some known documented occurence of this issue are

• community_build_a fails intermittently with corrupt scalatest artifacts scala/scala3#11604
• update sbt-osgi to 0.9.6 akka/akka#31437

The applied workaround in these issues was to downgrade to sbt-osgi 0.9.4, but said version does not work with Java 17 and newer.

Here are some solutions:

1. Fix the race. This requires very thorough understanding of sbt, which I lack
2. Revert Use dependencies as JARs for force proper manifests (fixes #60) #64, which will result in less accurate generated manifests
3. Introduce a new config setting to fine control how accurate we analyze dependencies to generate proper Import-Package statements. This is effectively a way to switch Use dependencies as JARs for force proper manifests (fixes #60) #64 on and off. We could default to off. Also, it should come with some documentation.

[mdedetrich]

@romainreuillon Should I assign you this issue? I noticed that you made some PR's on this topic so you seem to know the most about this problem

[romainreuillon]

I think this PR: 65ad38f fixes this issue doesn't it?

Can you try @lefou?

[lefou]

I thought that too, but if I understood @mdedetrich correctly, sbt-osgi 0.9.8 (which contains the fix) does not work with Pekko, whereass the specially released sbt-osgi v0.9.4-INVALID-CEN-JAR-PATCH does.

[mdedetrich]

@romainreuillon I have already merged that PR and made a release of sbt-osgi (v0.9.8) and it didn't solve the issue (at least for Pekko)

[mdedetrich]

I thought that too, but if I understood @mdedetrich correctly, sbt-osgi 0.9.8 (which contains the fix) does not work with Pekko, whereass the specially released sbt-osgi v0.9.4-INVALID-CEN-JAR-PATCH does.

Maybe you can try and replicate it on your end, if it did end up solving the issue for you maybe we have multiple issues (and/or I didn't test correctly in which case I would do it again).

[lefou]

@mdedetrich Can you post the error and/or give a link the the build logs?

[mdedetrich]

The core issue is here apache/pekko#757 but the error I get is apache/pekko#745 (comment), i.e. there appears to be multiple classes in the produced jar's?

Let me know if I need to provide additional info

[lefou]

Thanks. As stated previously in the various Akka vs. OSGi tickets, I no longer have any sbt project that uses Akka and OSGi. I still have OSGi projects but I build them with Mill now, where this issues does not occur, as Mill has a more proper cache and output concept than sbt. I do not plan to work on these issue again, but I wanted to point out potential related issues as I felt kind-of responsible, as it was my PR that introduced the race condition. I'm happy that one is fixed. I now think the multiple classes issue in unrelated.

[lefou]

I think we can close this issue and assign it to version/milestone 0.9.8.

[lefou]

Maybe, it's a good idea to re-add the initial test case I added to #61, which then got reverted in #63 + #64, apparently due to the race condition.

[mdedetrich]

Okay, I will do a bit more digging and/or wait for @romainreuillon response (if he has one) to confirm that we are dealing with different separate issues or a rather a single one.

[mdedetrich]

Maybe, it's a good idea to re-add the initial test case I added to #61, which then got reverted in #63 + #64, apparently due to the race condition.

Ill get to this during the week

[romainreuillon]

Indeed I have not encountered this kind of error message yet.

[mdedetrich]

Maybe, it's a good idea to re-add the initial test case I added to #61, which then got reverted in #63 + #64, apparently due to the race condition.

@lefou PR made at #97 and it passes

[mdedetrich]

@lefou @romainreuillon So shall I close this issue since its been confirmed that the problem Pekko is experiencing specifically is another issue and this issue is actually solved?

[lefou]

@mdedetrich Yeah, this issue is fixed now. Thank you for re-adding the tests.