Timothée Ravier

We've merged the CI fixes. Can you rebase this one? Thanks

The problem with copying the entire `EFI` directory is that we don't know what's there, and there might a lot of other vendor directories (`windows`, `ubuntu`, etc.) that we don't...

That looks ok but the logic feels a little bit difficult to follow. I might have missed something however so maybe my understanding is not correct. I'm thinking about the...

> if changes are EFI/fedora/subdir/newgrub.x64 and EFI/fedora/grub.x64, the temp dir will be EFI/fedora/subdir.tmp and EFI/fedora.tmp With `EFI/fedora/subdir/newgrub.x64` changed, we should copy `EFI/fedora` to `EFI/fedora.tmp`. That should cover all the subdirs...

I'm not sure I understand what you mean by "which comes first". I think the idea is more to think of it as the first subdir in the `EFI` folder...

> Sorry for the confusion, if the changes are `EFI/fedora/subdir/newgrub.x64, EFI/fedora/grub.x64`, copy to `EFI/fedora/subdir.tmp` first, then comes `EFI/fedora.tmp`, need to remove `EFI/fedora/subdir.tmp` first, just copy to `EFI/fedora.tmp`. If we have...

The goal is to make the operation atomic at the `vendor` (i.e. `fedora`, `BOOT`, etc.) directory level, not at the `EFI` or any other sub-directory level.

(I've not looked at the code again yet)

> make a test to verify the case where the update fails in the middle, right between the two rename exchange calls. We need to make sure that we can...

That might need a special "fail" mode flag for testing however so might not be desirable.