Conversation
ubuntu 22.04 was failing to run electrum.AppImage with the following error: dlopen(): error loading libfuse.so.2 AppImages require FUSE to run. Fix by installing libfuse2
|
test-liquid (cache miss): 7 minutes |
👍
What was the rational for using separate jobs instead of separate steps? It can run more quickly because the jobs run in parallel, however it seems that the savings aren't significant. Running the tests before this PR took 2m 51s, while with this PR it takes 1m 19s for the last test to finish. (with cache in both cases) However, in our CI setup prior to this PR, it seems that a significant time was spent loading the cache - 1m 26s vs 13s with the new So with just a better caching strategy we could get most of the speedup gains of this PR, even without using separate parallel jobs. The most that could be saved by running in parallel is the difference between the serial execution time of 1m 11s and the last-to-finish test execution time of 45s -- a maximum of 26 seconds. The drawback I see for separate jobs is that it complicates the workflow file, requires repeating the setup actions for each job and maintaining changes in multiple places. Reusing the setup actions would reduce this PR's workflow file by 21 LoC (~40%) and make maintenance easier. (The setup duplication also requires more compute resources overall -- but I guess we don't really care about that since GH is footing the bill)
We were using an outdated version of However I'm a bit wary of using unofficial actions if there's no compelling reason to. Just last month there was a supply-chain attack against a widely used GitHub Action (CVE-2025-30066). Does |
|
It was more about fixing the broken CI than anything else. |

Updates the CI workflow: