Windows os errors during build/test cleanup

[bollwyvl]

As always, thanks for rattler-build!

I have now seen issues during both the build and test phases where windows is unable to proceed due to (presumably) file permissions and/or dangling processes:

On conda-forge/granian-feedstock#42:

 ╰─────────────────── (took 3 minutes)
  all tests passed!
Error: 
  × failed to remove directory `D:\bld\bld\rattler-build_granian-with-
  │ all_1739238925\work\test`: Access is denied. (os error 5)

I tried using some python tricks (psutil and its lovely child tracking), but it was no good. I just skipped the test on windows and moved on.

On conda-forge/plantuml-feedstock#62:

 │ │ Archive written to 'D:\bld\win-64\plantuml-1.2025.1-he596248_0.conda'
 │ │
 │ ╰─────────────────── (took 4 seconds)
 │
 ╰─────────────────── (took 9 minutes)
Error: 
  × failed to remove directory `D:\bld\bld\rattler-
  │ build_plantuml_1739725916\build_env`: The process cannot access the file
  │ because it is being used by another process. (os error 32)

This is of course harder to work around, as I can't just skip the build.

While leaking disk/memory is of course A Bad Thing, I'd really rather be able to opt into getting a warning... which I could duly ignore on windows CI until it really failed due to resource exhaustion.

[bollwyvl]

update: I got plantuml to build with an explicit call to gradlew -stop 😅

Nice gains, too:

• 22m 28s with conda-build
• 14m 39s with rattler-build

[wolfv]

I believe conda-build has some retry/backoff/rename strategy for this kind of failure mode.

We should probably look into something like this. We also had to implement some similar facilities on unix because Go produces a cache that doesn't belong to the user and then isn't easily removable.