Remove all of the experiments in the workspace

[karajan1001]

fix #5676

1. remove all the experiments in workspace
2. add tests for it

• ❗ I have followed the Contributing to DVC checklist.

• 📖 If this PR requires documentation updates, I have created a separate PR (or issue, at least) in dvc.org and linked it here.
  Remove all experiments dvc.org#2722

Thank you for the contribution - we'll try to review it as soon as possible. 🙏

[pmrowla]

This will remove all experiments in the user's entire git repository. exp remove --workspace should only remove experiments that are based on the user's current HEAD (and not experiments based on other git commits)

#5676 (comment)

so we can't use the general exp_refs() iterator here, it should be:

        head = repo.scm.get_ref()
        exp_refs_by_baseline(head)

[dberenbaum]

I think it's fine to leave the behavior as is and rename the flag to --all. That fits the requirement here and is easier to explain.

[pmrowla]

@karajan1001 can you please rename the flag to --all? If it is left as --workspace it will be confusing, since all other DVC commands use --workspace to mean HEAD + whatever is in the actual working directory

[karajan1001]

@pmrowla , @dberenbaum
I think the feature --all is not as useful as --workspace? Can add it sometime later?

[dberenbaum]

I think which is more useful depends. Most of the time, I would guess users won't notice a different between them. If they have experiments in other commits, they might want to keep those, but I think it's just as likely that they have forgotten about them or don't know they still exist and would be happy to drop them.

After the discussion in this ticket, I prefer --all because:

1. It's easier to explain and understand. It's not obvious what is a "workspace" experiment, and users might want to confirm that these are the experiments they want to remove.
2. It parallels dvc exp list. If we eventually want to do something like dvc exp remove --remote, having this parallel dvc exp list seems useful since it's easy to see what experiments will be deleted.
3. Similarly, in the future we could add the --rev option like in dvc exp list. This is probably more useful than the --workspace/--all-branches/--all-tags/--all-commits flags since --rev is more granular (and --rev HEAD can do the same as --workspace), and I don't see --all-branches or --all-tags ever being useful in dvc exp remove.

[karajan1001]

[pmrowla]

Also, don't forget that this will need a docs PR

[pmrowla]

I'm not actually sure if -A/--all should affect queued experiments, since we have a separate option for that (--queue). It seems to me that -A should only affect "full" experiments that have already been reproduced

maybe a question for @dberenbaum

[dberenbaum]

🤔 It seems simpler to explain if it includes the queue, but it's probably more useful if it doesn't since we already have an option to remove the whole queue. Going back to the discussion above, dvc exp list ignores queued experiments, and I think it makes sense to do the same here.

[karajan1001]

@dberenbaum , Same with -w/--workspace, do not affect queued ones?

table	workspace based	all based
committed	1	2
queued	3	4

currently --queue removes 3 and 4. After next change --all removes 1 and 2, and workspace? removes 1 and 3 or only 1

[dberenbaum]

Sorry, I thought we were getting rid of --workspace?

[karajan1001]

I keep both of them for now.

[dberenbaum]

I would recommend keeping one, and I prefer --all so we don't have to get into the confusing corner cases like the one you mentioned above. I think when you document the features, you will find it easier if you don't have to document --workspace 😄 .

As mentioned in the other discussion above, I think you can add --rev later if functionality similar to --workspace is needed.

[pmrowla]

LGTM on the engineering side, but should hold off on merging until @dberenbaum double checks the final usage

[dberenbaum]

Looks great! Sorry for the lack of clarity on the syntax. Hopefully keeping it simple for now will be helpful, especially when we start expanding usage like in #6006.