Modifying multiple commits at once, e.g. deleting a file from each

[BatmanAoD]

Is there a way to apply the same change to multiple commits at once, e.g. to delete a file from each commit in some revset?

[martinvonz]

No. I've considered commands for some of that. Deleting paths is easy. Adding a file with constant content across a bunch of commits is also easy. Other cases seem harder. It's the UX that's hard, not the implementation. We should probably collect some use cases first.

[PhilipMetzger]

Agreed, it basically should be jj run 'rm <file>' -r <revs>.

[BatmanAoD]

In the particular instance I ran into today, I found that I had accidentally committed new files to my branch quite a few commits ago. In a similar instance that I run into frequently, my unit tests modify tracked files (...which, yes, I would like to fix). In both of these cases, I essentially just want to restore the state that exists in some other commit, so if restore supported multiple --to targets, that would be mostly sufficient.

There's one other wrinkle, though. As far as I can tell, restore --from <foo> --to <bar> will delete added files, but restore --from <foo> --to <bar> <PATH> will not; so even with multiple --to targets, I wouldn't have been able to delete the specific files that way.

[ilyagr]

There's one other wrinkle, though. As far as I can tell, restore --from --to will delete added files, but restore --from --to will not

If you can reproduce this, it sounds like a bug or at the very least a missing feature to me: either the default behavior should allow deleting files, or there should be an option to allow it. In your example, is PATH a directory with multiple files or just one file?

[BatmanAoD]

It appears I was mistaken, sorry; I had dome some quick testing of this but I guess I got confused somewhere. Looks like restore --from <foo> --to <bar> <PATH> does delete <PATH> if it's missing from <foo>.

[ilyagr]

Good to know, thanks for checking! And if you can find a case where it doesn't work, let us know.

[ilyagr]

I will close this discussion in favor of #4757 (choosing more or less randomly)