status: what should be the default behavior and distinction to diff?

[dmpetrov]

git status and git diff have different syntax (diff is based on checksums, status - not) and semantics (diff is based on history, status - on workspace and index).

dvc diff and dvc status have different syntax with similar semantics because dvc diff is not doing an actual content diff. So, these commands could be potentially "merged" into a single one.

Even if we decide to keep both I'd still prefer to see a status-like output for dvc diff when a file was not added to the repo by dvc add.

[jorgeorpinel]

There's definitely perceived overlap, which may be confusing to new users accustomed to git status and git diff which don't really have such an overlap.

I played with both commands and came up with this comparative summary:

• dvc status always works for all projects (even non-Git). It always compares DVC-file data vs. actual files in the workspace, and reports this info grouped by DVC-file.
  Main use: Know when to run dvc add or dvc repro, and on which targets.
• The new dvc diff only works for Git+DVC repos. It should compare HEAD vs local changes by default I think (similar to dvc status) but you can pick any 2 Git refs A..B (as long as A comes before B). Reports found differences in added/deleted/modified groups + count summary.
  Main use: Compare DVC repo versions

So clearly their motivations and purposes are different right now. But still, they are somewhat confusing given possible overlap and similar output. My suggestion would be to keep both but consider redesigning dvc diff (at some point).

I think dvc status is OK because the Git analogy is pretty good.

Conclusion

dvc diff is the tricky one: on the one hand we have the confusion explained above; on the other, it takes the same 2 arguments as dvc diff (2 commits) AND it has the same purpose (compare repo/object versions) so the Git analogy is also good. For this reason I don't see a strong argument to change it soon.

Perhaps reordering the output of dvc diff by data file instead of by type of change would make it feel less like a "status" report, but we just redesigned the output in #3229...

I think the real solution will be when/if #770 is addressed (some sort of semantic data format plugins).

[dmpetrov]

I don't see a strong argument to change it soon.

Agree.

I think the real solution will be when/if #770 is addressed (some sort of semantic data format plugins).

💯 Both will be needed with content diff

Related: #1406

[jorgeorpinel]

OK I guess we can close this one then?

[jorgeorpinel]

Both will be needed

Ah, maybe not!

[jorgeorpinel]

Another distinction between status and diff is that the former checks untracked workspace changes (same as Git working tree) while the latter checks tracked but uncommitted changes in the workspace (including Git staged DVC-file and .gitignore for ex.) So that's another reason to close this one, I think?

We're working on just improving the docs to clarify all this. See treeverse/dvc.org#953 (review).