cli: implement enough of jj fix to run a single tool on all files

[hooper]

Our immediate motivation is to have something basic that will work with a code formatter multiplexing wrapper script thingy inside of Google. The goal is to get to feature parity with hg fix, minus a couple of bad parts.

The main difference to look out for here was mentioned in Martin's PR, which is that for efficiency we want to loop over file IDs instead of (file ID, commit ID) pairs. That comes with some complications to error reporting.

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

[PhilipMetzger]

just minor stuff and a question

[thoughtpolice]

Just as a note, I think it would be nice if:

1. jj fix took a fileset to express which paths to actually touch
2. It would be nice if the user had some way of specifying the command per fileset in some way, rather than having a single blob which is expected to differentiate between all possible input files.

Maybe a map in the config, something like:

[fix]
"glob:**/BUILD" = "fix-bazel-files" # this is a command in $PATH
"glob:**/*.rs" = ["rustfmt", "..."] # include more arguments

Which would allow a more fine-grained approach.

Probably neither of these need to be handled on a first pass, though, and are relatively simple additions.

[martinvonz]

• jj fix took a fileset to express which paths to actually touch

Yes, that's coming in a later PR. I don't think @hooper has started working on it, but we're planning to add it, and I think it's trivial to add.

2. It would be nice if the user had some way of specifying the command per fileset in some way, rather than having a single blob which is expected to differentiate between all possible input files.

Also planned :) hg fix supports that. I think the plan is to do something similar.

[martinvonz]

btw, non-googlers: don't worry, i'm just helping review this but i'll leave approval for a non-googler

[hooper]

Also for non-Googlers, a lot of the context is in the documentation for the prior art in Mercurial: https://www.mercurial-scm.org/doc/hg.1.html#fix

[PhilipMetzger]

Maybe a map in the config, something like:

[fix]
"glob:**/BUILD" = "fix-bazel-files" # this is a command in $PATH
"glob:**/*.rs" = ["rustfmt", "..."] # include more arguments

Which would allow a more fine-grained approach.

In my opinion, fix is actual the perfect command for TOML tables, so you even could name the formatting/fixing action. So a eventual --verbose flag can print something like ran clang-format on <file>.cc, <file>.h.

This would fix some issues, which the Mercurial version has.

[fix]
# "patch" could mean accept the patch it generates and directly apply it.  
clang-format = { bin = "git-clangformat", files = "glob:**/*.h,**/*.cc", patch = true } 
rustfmt = { bin = "rustfmt", files = "glob:**/*.rs", args = "..." }
buildifier-lint = { bin = "buildifier", files = "glob:**/*.bzl,BUILD", args = "--lint" }
buildifier-fmt = { bin = "buildifier", files = "glob:**/*.bzl,BUILD" }

[PhilipMetzger]

Another round of mostly minor points, it looks good enough to land for me anyway.

[PhilipMetzger]

LGTM, if you create the issue and address my doc nits. I'd wait for a final approval from Waleed though, as he definitely is better qualified.