Bump ubuntu image in ci to 22.04#725
Conversation
|
Out of curiosity why not use |
|
The images in Ubuntu ship with the Android Native Development Kit (NDK), and historically changes in this NDK has not always been smooth with respect to uniffi. Using In this case the NDK goes up again, and that has me nervous I'm not gonna lie. I'm running the Android build here. We'll see if it succeeds. |
|
Gotcha. Would it save future headache to have |
|
Actually small correction: I think we also had issues with the |
Yeah in theory it would. I sort of like having it all build/run/test on the same image so it's cleaner to think through issues if there are any that are related to the CI images, but also they haven't caused any other issues than the one mentioned above really. So yeah I'm not feeling too strongly one way or the other. In this case my local ndk is 25.X so I'm looking at bumping that to the 27 version that's the LTS version mentioned here. I won't merge until I've figured this thing out... |
7329d4b to
24e2be2
Compare
|
I'm saying all of this but in fact they still change what's in the images even the images don't change! So you can pin the version of Ubuntu all you want but I remember once the Python wheels failed for some obscure reason and it was that the build tools included in the 20.04 image had actually been updated... forget about reproducibility it's just a circus out there with the GitHub CI images lol. Mind you their purpose is probably not build/publish systems but still. |
b409b1f to
749c62f
Compare
|
@rob I'm actually adding a fix to the live tests here (which have been failing for a few weeks) just to try to get it all in one PR and not have you review a bunch of PRs. I merged it on this branch just so I can actually run all live tests on it and make sure we fixed the issues. In Python it was just the imports that were wrong, and in Kotlin it's that we now use new descriptors (for quicker sync) and the CHANGE_FORBIDDEN option creates problems because we currently send "out" the sats instead of cycling them back to the external keychain, and we only had one output on the external keychain (my initial send to the wallet), and so now all owned outputs by the wallet are on the change keychain. For now I just removed this flag on the builder. I have also tested locally the Android build with the default NDK version that ships with the |
|
Strange the python linter didn't catch that unknown import. Oh well. Since the philosophy is to keep all versions the same in CI to rule out failures due to OS mismatch, I think this all makes sense. ACK 749c62f |
|
For some reason the Swift live tests fail (https://github.com/bitcoindevkit/bdk-ffi/actions/runs/14357615687/job/40254293039#step:4:624) but also the logs show that they all pass? I ran all live tests locally and if everything is green I'll merge. |
See #724.
Fixes #724.