IssueTracker

We've had several R8 bugs in the past caused by rolling in R8 revisions where R8's test suite had caught the same bug (they have CI but no CQ).

We should roll versions of R8 that are not known to be bad :P

Details from bug 404791127, where this last happened:

Thanks for the pointer. Around once a week we cut a new dev-version of R8 from the main branch, which then lands on the active R8 branch (currently 8.11). The latest commit on the active R8 branch should always pass all tests.

After a new dev-version is created we release it to Android Studio, the Android Platform and google3.

It seems that you can fetch the R8 branches and their most recent commit here: https://r8-review.googlesource.com/projects/r8/branches?r=[1-9].*. You would then need to sort the results according to their semantic version.

You can also take it from:

wget https://storage.googleapis.com/r8-releases/raw/latest-dev/r8lib.jar -O /tmp/latest.jar && unzip -c /tmp/latest.jar r8-version.properties | grep version.sha | cut -d "=" -f 2

if you don't mind downloading a r8lib.jar to get it (we can also archive an additional file just with the hash and version next to this)

We can also list the bucket via:

gsutil.py ls -l gs://r8-releases/raw/8.11.5-dev

For now, it's probably best to continue building from source since we add a couple of proguard -keeps. This will also let us switch back to building from main if there are patches we are eager to test out.