Potential workaround for #3820

[qwwdfsad]

Replace the specific place where ARFU gets misexecuted by specific Android toolchain

Fixes #3820

[dkhalanskyjb]

Let's hope this works!

[rossbacher]

Is there any eta for a merge/release or a snapshot? We can actually see this in our beta release and could verify a fix that way.

[qwwdfsad]

@rossbacher we do plan to release it around the next week or two.
About a potential snapshot -- @dkhalanskyjb WDYT?

[dkhalanskyjb]

Sure, we can publish 1.8.1-Beta in the next few days. Maybe even today.

Also, maybe we should document the procedure for publishing our libraries locally. There should be no reason to wait for a new release just to try out new changes.

[rossbacher]

I managed to figure out how to build and publish a fork (1.8.0 + this PR) to our internal artifactory on Friday. We did have this out on beta since then and so far no crashes (where we had about 2 crashes on the same # sessions on the previous beta that just had 1.8.0) so it looks like as if it is fixed. 🤞

[rossbacher]

Also, maybe we should document the procedure for publishing our libraries locally. There should be no reason to wait for a new release just to try out new changes.

This would be great, my initial approach was to build it and manually publish the artifacts to our artifactory (with manually patched .pom files from the previous release, but I realized that at least two artifacts we need (core and bom) did not seem to get built when you just run the build target.

In the end I just ended up (on my fork) changing https://github.com/Kotlin/kotlinx.coroutines/blob/master/buildSrc/src/main/kotlin/Publishing.kt#L60 to our repo URL and setting my credentials for libs.sonatype.user and libs.sonatype.password and running the publish target, which worked like a charm.

[aasitnikov]

We've rolled out an update that includes a fix from this PR a week ago. Yesterday, we had 1,400 crashes from 4.5 million users on the previous version without a fix, and zero crashes from 80,000 users on the current version with the fix. So far, it looks like this workaround is working.