exp run: add --no-save flag

[dberenbaum]

The first step could be adding a flag (--no-save?) to make the current dvc exp run behave like repro so that we have one consolidated command we could use when we are ready to deprecate the others.

Originally posted by @dberenbaum in #7866 (comment)

[dberenbaum]

What's our goal here? Do we want to have the same behaviour as dvc repro in dvc exp run --no-save? Or, is it just --no-save variant of dvc exp run? Ideally, they should be the same, but there are going to be subtle behaviour differences.

Originally posted by @skshetry in #7866 (comment)

[dberenbaum]

@skshetry Copying your comment here since #7866 has a lot of other noise.

What are the differences implementation-wise since they do both eventually call reproduce, right? One concern with basing it on exp run is additional overhead. If this is not doing any of the git tracking of the experiment and is always running in the workspace (no queueing or cloning to temp dirs), what overhead is left?

From #6791 (comment), there are a few options that aren't supported in exp run:

• --no-run-cache
• --no-commit
• --glob

I can't recall the issue with trying to support glob, but if it's too much of a mess to support it properly, I think we could drop it. The other two seem reasonable to keep only with --no-save, although I doubt either is commonly used.

[daavoo]

If we add --no-save and just make it an if that calls repo.reproduce, I don't see the need for any discussion. There is no overhead and all flags could be supported

[dberenbaum]

Another important distinction to preserve is that dvc repro can run with no_scm.

[dberenbaum]

@skshetry I've been playing with using exp run for everything, and TBH it feels annoying a lot of the time when you don't actually need an experiment. Even with --no-save, I would prefer not to have to add that flag all the time. So I don't think we even need to prioritize this flag.

What is critical is to make repro and exp run as similar as possible so we can say that they are functionally equivalent except for the additional overhead of saving, queuing, etc. without users needing to worry about small differences between them.

For now, it seems that these are the known discrepancies to address then:

From #6791 (comment), there are a few options that aren't supported in exp run:

* `--no-run-cache`

* `--no-commit`

* `--glob`

I can't recall the issue with trying to support glob, but if it's too much of a mess to support it properly, I think we could drop it. The other two seem reasonable to keep only with --no-save, although I doubt either is commonly used.