--config command-line option

[ilyagr]

What do you think about having a --config TOML_STRING global JJ option? This could replace the --color option, you would do --config ui.color=always instead. You could also do --config ui.diff-editor=vimdiff. You could also merge in an extra TOML file with --config "$(cat extra-config.toml)" on bash.

The downside of TOML is that it often requires line breaks, which are inconvenient on the command-line. However, that can be ameliorated by allowing repetition of --config options.

It could also be shortened to -c, assuming that global options are in a separate namespace from the subcommand options.

This came up as I was thinking about whether the resolve command discussed in #18 needs a --tool option.

[martinvonz]

Yes, I agree that we should do that. Both Git and Mercurial do it. I just haven't gotten around to it.

The option would go in the GlobalArgs struct.

One thing that complicates it a bit is aliases. I know that Git lets you define an alias that adds another alias, so you can do git -c alias.foo='-c alias.bar=status bar' foo. I'm not sure how useful that is, but it seems that we should ideally support it.

Oh, another thing that may complicate it is the planned support for reading configs from .jj/config.toml (or something like that). Since we have a -R flag for specifying a repo, it seems like there is another level of recursion like for aliases. For example, let's say you're in ~/repo1 and run jj foo, and there's a ~/repo/.jj/config.toml that says alias.foo = -R ~/repo2 foo, I suppose that means we should go to that repo and load the repo-level config from there (which could lead to infinite recursion if that repo's config says alias.foo = -R ~/repo1 foo). I checked what Git does in this case (the simple case without mutual recursion). It says fatal: alias 'foo' changes environment variables..

I don't know if there are any security issues to think about.

[ilyagr]

Crazy idea: what if jj didn't have general aliases? The shell or clap seem better positioned to deal with these issues.

I would very much like aliases inside revsets or for log output formats, but these specific cases don't raise nearly as many issues. Not having general aliases would make it more difficult and more important to carefully design those, as well as configuration for default revset patterns or formats. But, perhaps, that's worthwhile.