Fix: <https://github.com/google/ksp/pull/2164>

# ksp

[baxter]

Fix: https://github.com/google/ksp/pull/2164

[Zac Sweers]

Darn you just missed the 1.0.26 release. Hopefully a fast follow is possible

[baxter]

I honestly thought it would have been a much more drastic change for the fix.

[Ting-Yuan Huang]

Merged. Thanks for the fix!

[Zac Sweers]

how soon is too soon for 1.0.27 🥺

[Ting-Yuan Huang]

My rough estimation is 2-4 weeks.

[Ting-Yuan Huang]

How urgent do you need this? I think this bug has been around from the beginning and only discovered recently?

[Zac Sweers]

we've been dealing with a lot of memory pressure and leaks in ksp and kotlinc that pablo and jason have been fixing recently, would just be really nice to have them all available now that the last one is merged

[baxter]

@Zac Sweers just as an FYI, it looks like the ksp leak was producing ~1k files / run on NowInAndroid. The Kotlin file leak only produces around 400 files / run.

[Ting-Yuan Huang]

Wow. OK, let's try to do a release in 2 weeks.

[baxter]

And this would depend on when garbage collection runs too, since the file descriptors would be forced closed when they are collected.

[baxter]

@Zac Sweers, are you dealing with the file leaks in MacOS or on another OS?

[Ting-Yuan Huang]

I was reading this and it reminds me about some test failures that only exist on Windows. If I'm reading URLClassLoader.close() correctly, the only benefit on Linux / Mac is releasing the jars / classes earlier, right?

[baxter]

For MacOS, it'll also eliminate issues with hitting the file descriptor limit, since they made it difficult to increase it.

[baxter]

https://developer.apple.com/forums//thread/735798