Commit hashes different between splitsh-lite and git subtree split, part II

[kdambekalns]

This is named like #20, but a different bug, as far as we know.

The commit neos/flow-development-collection@882be01 results in neos/flow@4d563e6 right now, using the subtree split. Splitting using splitsh it ends up in kdambekalns/flow@79cae3e. The latter looks exactly like the commit from the development collection…

Looking further, there is no difference between the results, but the hashes differ. To maximise confusion, GitHub actually shows a difference, locally Git says it's the same (https://github.com/neos/flow/compare/3.0...kdambekalns:3.0-splitsh?expand=1 vs. git diff 3.0..3.0-splitsh after cloning kdambekalns/flow)

The split was done using splitsh-lite --prefix=TYPO3.Flow --origin=origin/3.0 --target=refs/heads/3.0-splitsh

[kdambekalns]

To top it off, we created a PR from the "difference" at kdambekalns/flow#1 that showed differences in 20 files. We wanted to have fun and merged that, resulting in no changes (see kdambekalns/flow@f5d167d) - which we expected, since the changes the PR wanted to do, actually existed in the base branch.

After the merge, the fork is now 23 commits ahead of the original, but again (as expected), no differences: https://github.com/neos/flow/compare/3.0...kdambekalns:3.0?expand=1

So, this might even be a bug in git and/or git-subsplit, and splitsh actually gets it right.

[kdambekalns]

No bug in GitHub, it’s caused by a “two dot” vs “three dot” diff. But the difference between the results is real.

[kdambekalns]

Anyone feels like being able to help with finding out the cause for this?

[dxops]

I think you have the same case like we discussed some time ago in #26 (comment)

[cebe]

sounds similar to yiisoft/yii2#15785 (comment)

[kdambekalns]

I am more or less convinced this cannot be changed and is the result of changes "to git itself". Thanks everyone for their input!

We'll most likely just accept the different hashes, composer should iron out any glitches for us (by preferring tags over hashes) except in the rarest edge-cases. 🤞