Add prepare artifacts job; fixup AzDO fetch script #1679
Conversation
| "$(nonportableSourceBuiltStageDir)" \ | ||
| "$(portableSourceBuiltStageDir)" \ | ||
| -iname 'Private.SourceBuilt.Artifacts.*.tar.gz' \ | ||
| -exec tar -xf {} -C "$(allSourceBuiltStageDir)" \; |
There was a problem hiding this comment.
What about PackageVersions.props file? If last one in wins, it will not reflect the combined files. It seems like this needs to be coupled with #1638
There was a problem hiding this comment.
This is (the general process of) how @crummel made the previously-source-built tarball we're using now, and I don't see any PVPs, so I assume it worked out... I'm not sure.
There was a problem hiding this comment.
Yeah, it's not strictly correct, but it works out because we don't use runtime package versions for anything right now. If we're productizing it, it might be worth looking into a bit. There's probably two things we care about:
- getting all package versions into PackageVersions.props: I think something like adding unique lines from each individual PackageVersions.props to the output PackageVersions.props would work decently well here and probably shouldn't screw up too much. This currently doesn't matter but would cover us if we ever needed to depend on, e.g., a runtime package's version.
- not having any conflicting package versions: in theory, this shouldn't happen. Package names and versions should be consistent across RIDs, and I don't think we've seen any real-life cases where this has gone wrong. I can't think offhand of a cheap way to detect this other than a bunch of
sednonsense, but I think it would be worth filing an issue for a future improvement.
I don't think either of these are blockers and #1638 should obviate them anyway.
|
Do you have the tarball from before you disabled for pull requests? I checked for the build but couldn't find it, I was just going to make sure it looked reasonable. |
|
I don't (I don't think the build completed), but this is the output for 3.1.106: Thanks for checking. |
|
osx job failure (libicu not found in smoke-test) tracked by #1685, seems impossible to be related to this PR's changes. Ignoring for merge. |
Add a simple prepare-artifacts job to take over manual work preparing previously-source-built tarballs.
Fix up the AzDO fetch script: simplify and document for the scenario it's actually used for: downloading onto fresh machines for manual testing.
I worked on this to avoid manual steps for 3.1.106, and polished it a little more for this PR for next time.
Fixes #1676