all: fix history of dev.regabi and dev.typeparams branches

[mdempsky]

@cherrymui noticed that https://go-review.googlesource.com/c/go/+/277032 (merge master into dev.regabi) was actually committed as a regular commit (one parent ref), rather than a merge commit (two parent refs). It's also been subsequently merged into dev.typeparams.

We should rewrite the history of these two branches so that "git bisect" later on will continue to work reliably in the future. Filing as an issue since my understanding is this is the Go release team's responsibility.

/cc @dmitshur @griesemer @dr2chase

[griesemer]

FWI, I rewound (checked out) dev.regabi locally just before the most recent merge, merged again, and then cherry-picked the handful of changes on top. It's 5min of work (the merge is clean except for two files, cmd/compile/internal/ssa.go and cmd/compile/internal/builtin.go - so make a copy of these files beforehand and just replace them during the merge).

The main issue is that we will force rewrite history and invalidate clients. But I think it's ok if we communicate beforehand.

I will happily do the dev.typeparams branch as well if we agree that this is what we should do. Since there's a lot master changes that went into the merge, being able to dissect on each individual change (rather than the one huge CL) would be good, in my mind.

(Thanks to Cherry for finding this.)

[FiloSottile]

This happens with some regularity, we should probably reprioritize #26201.

I think we should rewrite the history. Squashed merges often confuse git when doing later merges and will break blame.

Force pushes are destructive operations, so they require a Release or Security team member to obtain administrator privileges. Please reach out internally on Monday (or through my phone number if it needs to happen sooner).

[griesemer]

cc: @findleyr who is also working on dev.typeparams.
cc: @rsc in case he has pending changes

[aclements]

cc @golang/osp-team

[toothrot]

I can take care of this.

[toothrot]

dev.regabi has been updated:

8ce37e4110 (HEAD -> dev.regabi, sso-direct/dev.regabi) [dev.regabi] cmd/compile: fix noopt builder
7e17b46c58 [dev.regabi] cmd/compile/internal/types: add IsScalar query method
2b76429eb0 [dev.regabi] cmd/compile: refactor type initialization code into helper
9c5241e520 [dev.regabi] cmd/compile: remove unnecessary String methods
267975dc47 Merge branch 'master' into dev.regabi
6c64b6db68 (master) cmd/compile: don't constant fold divide by zero

It looks like @rsc had landed some changes this morning while I was working on this. We'll coordinate on getting those back on the branch.

[mdempsky]

It looks like the rewrite merged 6c64b6d, rather than 6d2b335 (what @dr2chase 's CL was attempting to merge). I think this should be fine (they're only a few commits apart), but thought I'd point out.

[mdempsky]

I've confirmed though that 6d2b335 merges cleanly on top of 267975d (@toothrot 's merge), and produces a tree identical to 2a1cf9d (@dr2chase 's merge). So the merge conflict resolutions appear to have been faithfully reproduced.

Thanks @toothrot and all!

Edit: Steps to reproduce yourself:

git checkout 267975d             # checkout @toothrot's merge
git merge -q -m merge 6d2b335    # merge up to the same commit @dr2chase merged
git diff 2a1cf9d                 # diff against @dr2chase's merge; should be empty output

[dmitshur]

Thank you for verifying that @mdempsky.

[toothrot]

Thanks for verifying, @mdempsky!

@griesemer and myself have corrected history for the dev.typeparams branch, and @griesemer is working on getting the changes that had previously landed re-submitted. I will leave this issue open until we've caught up.

[griesemer]

With commit 8ec9e89, the dev.typeparams branch has caught up again.