TODO: Attempt to rebase human commits instead of failing immediately

[balsoft]

update-daemon should ideally try to rebase "human" commits on top of the automatic update commit. It should work in most cases (when humans didn't change the flake.lock file), and would make the workflow much smoother.

[gromakovsky]

Example: https://github.com/serokell/tezos-infra/pull/78

[notgne2]

We used to have some form of rebasing before this commit 91d1f66 - which removed it as iirc it was very broken, but old implementation probably still worth referencing

[sancho20021]

We used to have some form of rebasing before this commit 91d1f66 - which removed it as iirc it was very broken, but old implementation probably still worth referencing

What were the main reasons why old implementation was broken, does anybody remember?)

[balsoft]

It didn't handle conflicts at all, resulting in broken state.

[sancho20021]

Ok, thanks.

I'm trying to understand the current behavior of update procedure, am I right that:

1. If the latest commit in remote update branch is done by someone else, updating fails;
2. If there were changes in default (master) branch and remote update branch exists, update will just reset local repository to remote update and won't try to synchronize with master, which will cause conflicts?

So, if the stated above is correct,

• in the first situation I just need to reset to the latest human commit?
• in the second I need to rebase update branch on top of master?

[sancho20021]

Btw what is the logic for aborting update in case of human commits in the update branch? What errors will happen if we remove this check?

[balsoft]

If the latest commit in remote update branch is done by someone else, updating fails;

Yes

If there were changes in default (master) branch and remote update branch exists, update will just reset local repository to remote update and won't try to synchronize with master, which will cause conflicts?

It resets the update branch to master every time, regardless of changes in master. I.e. it disregards the remote update branch state. It's basically

git checkout master
git pull
git checkout -B $UPDATE_BRANCH
nix flake update
git push --set-upstream origin $UPDATE_BRANCH --force-with-lease

[balsoft]

What errors will happen if we remove this check?

No errors will happen, but human commits may potentially be orphaned, and we presume that human commits are imporant and worth keeping, even at the cost of failing to update.

I guess it might kinda make sense to simply remove this check and let update-daemon force-push over human commits, and then let humans do the rebasing manually, but it would be annoying.