Merge train: infer candidate from state

rudymatela commented 2 years ago

... in preparation for #77 (partly closes: #77)

The changes in this PR do not change Hoff behaviour, the high level functionality should be the same.

Instead of maintaining a global candidate, we infer it from the state always. This is in preparation for merge trains (#77), where we will have multiple builds running in parallel and thus multiple candidates.

The ProjectState datatype is now simpler and easier to maintain (before we could risk it being inconsistent);
Some code paths are now simpler, as we don't have to do the work maintaining the global candidate;
There is a new explicit integration status Promoted to signify that the PR branch is now part of master, before this was signified by being Integrated ... BuildSucceeded but not being the global candidate.

A linear search on the number of open PRs is imposed. I believe this is not a problem because:

the number of PRs should be at most a few dozen, examples:
- there are only 5 open PRs on channable/hoff currently;
- a dozen open PRs seems to be a reasonable average (:wink:);
- a busy project (:wink:) should have between 100 and 200 open PRs in which a linear search is still reasonable;
- ... and we could still have way more than that without suffering significant performance impact (1000 or 10000 open PRs).
we already perform linear search for other operations, the candidate was just a special case in where we don't. If we were to actually get rid of linear search altogether, we perhaps should have a more generic solution that covers all cases. (If linear search were to be a problem, we would already be suffering from it.)
with #77, we are bound to perform a linear search on the candidate PRs anyway.

Notes

This is an alternative to PR #130. Here, instead of changing the datatype of integrationCandidate, we simply remove it and try to infer from the rest of the state every time.

rudymatela commented 2 years ago

c'mon GitHub, I thought you could handle the renaming of branches... apparently not.

rudymatela commented 2 years ago

c'mon GitHub, I thought you could handle the renaming of branches... apparently not.

So I tried renaming the branch from junk/merge-train/infer-integration-candidate by removing the junk/ prefix, signifying that this was no longer an experiment. GitHub does not let me do it without opening a new PR. So I'll leave the junk/ prefix then even though it should be not there in spirit.

rudymatela commented 2 years ago

@fatho Thanks for the review.

Since I think I have addressed your comments, I am going to merge and tag now.

I will leave the deploy for later today at around 19:00 (if there are no items in the build queue).

@OpsBotPrime merge and tag

OpsBotPrime commented 2 years ago

Pull request approved for merge and tag by @rudymatela, rebasing now.

OpsBotPrime commented 2 years ago

Rebased as c3d7e54d61afcb9f1ce94951c8646dcf1fad4c0b, waiting for CI …

OpsBotPrime commented 2 years ago

@rudymatela Sorry, I could not tag your PR. The previous tag v0.25.6 seems invalid

rudymatela commented 2 years ago

@rudymatela Sorry, I could not tag your PR. The previous tag v0.25.6 seems invalid

Oohh yes, of course. #117 will make you support it.

channable / hoff

Merge train: infer candidate from state #131

Notes