[romtsn]

PR Stack (Robolectric CI)

• test(android): Speed up Robolectric tests in CI #6543
• test(android): Use fewer Robolectric SDK levels in core and replay #6544
• ci: [Robolectric CI 2] Cache Robolectric android-all jars in the build job #6545

---

📜 Description

Caches ~/.m2/repository/org/robolectric in the Build job:

• Every run restores the cache (actions/cache/restore) before make preMerge.
• Only main saves it (actions/cache/save), and only when there was no exact hit. PRs read the entry that main saved, like setup-gradle does for the Gradle caches. Otherwise each PR would save its own ~700 MB entry that only that PR can use.

The cache key is a hash of the Robolectric version line in libs.versions.toml and every sdk = [...] / sdk= in the test sources. So the key changes only when the Robolectric version or the set of SDK levels changes, not on each dependency bump. restore-keys allows a partial hit, so a new SDK level downloads only its own jar.

💡 Motivation and Context

Robolectric downloads the android-all-instrumented jars at test runtime into ~/.m2/repository. setup-gradle caches only ~/.gradle, so each CI run that executes Robolectric tests downloads 100-200 MB per SDK level. With the levels from #6544, that is ~713 MB for core and replay.

💚 How did you test it?

A temporary benchmark workflow (not part of this PR) ran only the core and replay Robolectric tests. Each job ran cold runs (jars deleted first) and warm runs in turns on the same runner, so the runner hardware does not affect the comparison. Mean of 3 rounds per job:

Job	Jars	core cold → warm	replay cold → warm
with #6544, Intel Xeon 6973P	713 MB	117.0s → 112.3s (+4.7s)	71.0s → 67.3s (+3.7s)
with #6544, AMD EPYC 7763	713 MB	161.8s → 157.5s (+4.3s)	94.7s → 92.3s (+2.4s)
without #6544, AMD EPYC 9V74	1244 MB	163.0s → 152.2s (+10.8s)	91.1s → 83.1s (+8.0s)
without #6544, AMD EPYC 9V74	1244 MB	169.0s → 162.9s (+6.1s)	92.9s → 89.6s (+3.3s)

So the cache saves about 4-5s per run with #6544, and up to ~11s without it. Runners download from Maven Central at about 150-370 MB/s, so the download is short. curl of the same 6 jars takes about 3.5s.

Note: a first comparison of two Build runs of this PR (cache miss vs. hit) showed 2.5 min less, but most of that was runner variance. All test tasks in that run were 20-30% faster, including the Spring tests, which do not use Robolectric.

📝 Checklist

• I added GH Issue ID & Linear ID
• I added tests to verify the changes.
• No new PII added or SDK only sends newly added PII if sendDefaultPII is enabled.
• I updated the docs if needed.
• I updated the wizard if needed.
• Review from the native team if needed.
• No breaking change or entry added to the changelog.
• No breaking change for hybrid SDKs or communicated to hybrid SDKs.
• Public API changes reviewed by another Mobile SDK team member or implemented according to the develop docs spec.

🔮 Next steps

The entry is ~750 MB of the repo's 50 GB Actions cache. If cache pressure becomes a problem, remove restore-keys so that old jars do not stay in the entry after a Robolectric upgrade.

🤖 Generated with Claude Code

⚠️ Merge this PR using a merge commit (not squash). Only the collection branch is squash-merged into main.

#skip-changelog

[sentry]

📲 Install Builds

Android

🔗 App Name	App ID	Version	Configuration
SDK Size	io.sentry.tests.size	8.63.0 (1)	release

⚙️ sentry-android Build Distribution Settings

[romtsn]

closing in favour #6553