Unexpected shas for Merge Queues

[Fitzbert-Fitz]

Hi there,
as also explained here #100 we encounter massive building overhead due to the sha calculation.
What we would expect:
assume you have two PRs queued in the merge queue. Tha latest main build is successful. We would assume that Queue positon '#1' would derive latest main(successful) as base sha. Position '#2' in the queue would derive position '#1' as base sha since it assumes that position '#1' will pass the queue as well.

But what we see is that the base sha is always the git merge-base / the common ancestor. Meaning Position '1#' in the queue builds all changes since the branch was created not taking into account that builds on main might have been successful in the meantime. The same for position '#1' in the queue. See below:

Looking at the code we do not understand why git merge-base is used here. It might be a good fall back in case you do not find successful commits on main. Maybe miss ordered inside the if else statement?

nx-set-shas/find-successful-workflow.ts

Lines 42 to 62 in 76907e7

	if (
	(["pull_request", "pull_request_target"].includes(eventName) &&
	!github.context.payload.pull_request.merged) ||
	eventName == "merge_group"
	) {
	try {
	const mergeBaseRef = await findMergeBaseRef();
	const baseResult = spawnSync(
	"git",
	["merge-base", `origin/${mainBranchName}`, mergeBaseRef],
	{ encoding: "utf-8" }
	);
	BASE_SHA = baseResult.stdout;
	} catch (e) {
	core.setFailed(e.message);
	return;
	}
	} else {
	try {
	BASE_SHA = await findSuccessfulCommit(
	workflowId,

[mandarini]

Hi there @Fitzbert-Fitz ! Thanks for filing an issue! Do you think the changes suggested by @paulo9mv in the PR linked above would solve your issue? Do have a suggested implementation that would make things smoother?

[andersonba]

up!

[scottkgaran]

I feel like @paulo9mv's solution should work. I can take a stab at testing it.

[paulo9mv]

I will re-open that PR, since after using this solution for a quite time, it solved my problem and behaved as expected.
On the merge_group event the expected behaviour is that base SHA is the previous commit on the queue (or the head commit on main branch, if queue is size 1)

PR opened: #145