Build: Use actions/cache

[hoffie]

Describe the bug

We should use actions/cache to improve build speeds and resiliency against temporary flakiness on external dependencies.

The following things to check come to mind:

• aqt: Can keep downloaded archives, but won't re-use them. Caching the whole Qt directories for that reason now.
• choco: Hardcoding a cacheLocation now
• brew: not used
• pip: Different paths for Windows/Mac
• Android SDK & NDK
• brew if added via Switch to create-dmg #2207 Tracked in Build: Cache Mac's brew install (create-dmg) #2412

Depends on #2276 to avoid caching a bad state.

To Reproduce

Random flakiness such as Qt download failures.

Expected behavior

Dependencies should be cached. Temporary download failures should not break the build.

Screenshots

https://github.com/jamulussoftware/jamulus/runs/4936344409?check_suite_focus=true

Caught ArchiveConnectionError, terminating installer workers
Failure to connect to https://download.qt.io/online/qtsdkrepository/windows_x86/desktop/qt5_5152/qt.qt5.5152.win64_msvc2019_64/5.15.2-0-202011130602qtremoteobjects-Windows-Windows_10-MSVC2019-Windows-Windows_10-X86_64.7z.sha1: ReadTimeout

Additional context

https://docs.github.com/en/actions/advanced-guides/caching-dependencies-to-speed-up-workflows
https://github.com/actions/cache

[hoffie]

My current PoC is at hoffie@cache-autobuild-deps. Will submit a PR after some more testing and history cleanups.

Relevant runs:
Initial run: https://github.com/hoffie/jamulus/runs/4948136692?check_suite_focus=true
Cached run: https://github.com/hoffie/jamulus/actions/runs/1749769513

[hoffie]

As expected, some rather small performance improvements kick in:

Job	Prepare (uncached)	Prepare (cached)	Restore cache	Save cache	Cache size	Total (master)	Total (1st)	Total (2nd)
Android	7m 38s	1m 16s	0m 53s	1m 32s	1.886 MB	35m 51s	49m 0s	31m 52s
MacOS (Artifacts)	0m 28s	0m 0s	0m 11s	0m 5s	111 MB	8m 28s	6m 8s	5m 51s
Linux	1m 34s	1m 21s	-	-	-	11m 57s	13m 55s	10m 57s
Windows	2m 25s	0m 27s	1m 49s	1m 10s	294 MB	25m 27s	32m 53s	27m 21s

Note: I have included the Linux job despite there being no changes. This shows that there is some variance in runtime duration.

I had tried to calculate absolute/relative improvements, but the improvements by caching would have been much larger (17 minutes for Android) than what could be explained by the shorter Prepare step duration. Therefore, I assume that run time is highly influenced by external factors on Github Actions Runners. So, take those numbers with a grain of salt, they're only from two runs.

[pljones]

Is this closed by #2284? I wasn't sure from the comments.

[hoffie]

Is this closed by #2284? I wasn't sure from the comments.

Yes, it was correctly auto-closed by #2284. The remaining to do is covered by #2412. I'll add it to the PR description to clarify. Everything else should be covered.

Umm: I missed that the version pinning for brew/create-dmg happens in a different file. So, in theory, this should have been part of this PR. #2412 will have it anyway.