resolve: try to resolve all conflicted files in fileset

[scott2000]

If many files are conflicted, it would be nice to be able to resolve all conflicts at once without having to run jj resolve multiple times. This is especially nice for merge tools which try to automatically resolve conflicts without user input, but it can also be used for regular merge editors like VS Code.

Checklist

If applicable:

• I have updated CHANGELOG.md
• I have updated the documentation (README.md, docs/, demos/)
• I have updated the config schema (cli/src/config-schema.json)
• I have added tests to cover my changes

[arxanas]

I recently learned that some commands support filesets. Is jj resolve one of those? Could we support this use-case by extending the fileset language, rather than adding a flag specifically to jj resolve? (Does jj resolve 'all()' work today?)

[martinvonz]

We actually already support filsets in this command. It's just that the command would always stop after the first conflict before this PR (so even jj resolve 'all()' would just pick the first conflict it runs into).

[scott2000]

I recently learned that some commands support filesets. Is jj resolve one of those? Could we support this use-case by extending the fileset language, rather than adding a flag specifically to jj resolve? (Does jj resolve 'all()' work today?)

Yes, it already does support filesets, and the default is all() currently already. The issue is that it always picks a single file from the fileset, rather than using the whole fileset. The new --all flag just tells it to use the whole fileset, rather than picking a single file from the fileset. We could maybe make it so that if the user explicitly passes a fileset, it uses the whole fileset? I'm not sure if anyone relies on the existing behavior though.

Edit: nevermind, looks like @martinvonz already responded!

[arxanas]

We actually already support filsets in this command. It's just that the command would always stop after the first conflict before this PR (so even jj resolve 'all()' would just pick the first conflict it runs into).

Is this because existing merge editor work on a single file at a time, so we have no choice but to resolve conflicts individually? (Is there a warning in the case that subsequent conflicts are ignored?)

For uniformity, should we have an all: modifier (like for revsets) for filesets?

• Assuming that it's a limitation of the merge editor interface, and you don't inadvertently want to resolve multiple files at once, perhaps it makes sense to warn before trying to resolve conflicts for multiple files, in the same way that we warn for certain multiple-revision operations?
• I could be convinced either way, but by default, I would be inclined to keep the DSLs similar.

[scott2000]

If we want the default behavior of picking a single file to be expressable with filesets, then we'd probably want a first(fileset) function to pick a single file from a fileset. I'm not sure if that would make the implementation of filesets more complicated though.

TBH, I'm not sure I've ever actually wanted the default behavior of only resolving a single, seemingly-random file though. Usually, I want to resolve every conflict, or I want to resolve a conflict in a specific file I specify. So maybe we could just change jj resolve to just resolve every file in the fileset and require the user to name a specific file if they only want to resolve a single file?

[arxanas]

If we want the default behavior of picking a single file to be expressable with filesets, then we'd probably want a first(fileset) function to pick a single file from a fileset. I'm not sure if that would make the implementation of filesets more complicated though.

Makes sense to me, although implementing that is its whole own matter.

TBH, I'm not sure I've ever actually wanted the default behavior of only resolving a single, seemingly-random file though. Usually, I want to resolve every conflict, or I want to resolve a conflict in a specific file I specify. So maybe we could just change jj resolve to just resolve every file in the fileset and require the user to name a specific file if they only want to resolve a single file?

In both your workflow and my workflow, it sounds like resolving a single arbitrary file is probably not desirable:

• In my own workflow, I typically get scared and confused when I try to finish up and I wasn't expecting the tool to launch more editors.
  • It used to be particularly difficult to resolve in cases where I was using a CLI editor to resolve conflicts, so there was no way to just go back to my terminal and type ctrl-c to cancel the whole process.
  • These days, I know how to make Vim exit with an error code, but there may be other users might share my old workflow of trying to trying to quit and then press ctrl-c really fast before the caller can launch another editor 👀.
• Thererfore: I would prefer an error in the case that jj resolve would operate on multiple files, while you would prefer to just start resolving multiple files, but neither of us would want the current behavior.
• But, in fairness: I don't use jj resolve that much vs just opening the conflicted files in my editor and letting them get auto-resolved, so I'll defer to anyone who specifically uses jj resolve.

[scott2000]

These days, I know how to make Vim exit with an error code, but there may be other users might share my old workflow of trying to trying to quit and then press ctrl-c really fast before the caller can launch another editor 👀.

I think in practice, this shouldn't be an issue because we treat an unmodified or blank output file as an error, so it would stop launching more editors if someone closes the editor without making any changes. That being said, it does feel a bit strange to launch the editor multiple times without the user requesting it explicitly, so it might be nice to not launch multiple editors by default anyway.

At some point, I'm thinking that we probably will want to support merge tools which can merge whole directories at once instead of being invoked with each file individually.

[martinvonz]

So maybe we could just change jj resolve to just resolve every file in the fileset and require the user to name a specific file if they only want to resolve a single file?

I'm also for that.

[martinvonz]

Oh, @ilyagr might also have thoughts. He wrote most of the current implementation.