cli: lazily emit cargo:rerun-if-changed to avoid spurious rebuilds

[ipetkov]

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/updated tests to cover my changes

[ipetkov]

Hmm test suite passed for me on the initial run, but now that I have this merged into another branch I see the test_global_opts::test_version failing consistently 🤔

thread 'test_global_opts::test_version' (3658728) panicked at cli/tests/test_global_opts.rs:61:5:
    `jj version` output: "jj 0.32.0-dirty\n".
    Sanitized: "jj ?.??.?-?irty\n"
    Expected one of: ["jj ?.??.?\n", "jj ?.??.?-????????????????????????????????????????\n"]
    stack backtrace:
       0: __rustc::rust_begin_unwind
       1: core::panicking::panic_fmt
       2: runner::test_global_opts::test_version
       3: runner::test_global_opts::test_version::{{closure}}
       4: core::ops::function::FnOnce::call_once
    note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.

[ipetkov]

Maybe another option to hardcoding NIX_JJ_GIT_HASH = self.rev or "dirty"; would be to change it to NIX_JJ_GIT_HASH = self.rev or ""; in the devshell, and then tweak the build script to not ignore when $NIX_JJ_GIT_HASH is set but empty? Not fully sure the implications of doing that (would this break version output for Nix builds?) so I could use a pointer on whether there's a preferable way to resolve this or just live with the spurious rebuilds and close this PR

[yuja]

cc @thoughtpolice as I have no expertise on the nix part.

[yuja]

The changes in build.rs look good. I can't review the nix change, so I'll leave it to someone else.

[martinvonz]

The changes in build.rs look good. I can't review the nix change, so I'll leave it to someone else.

Maybe @thoughtpolice or @emilazy can help?

[ipetkov]

Hmm test suite passed for me on the initial run, but now that I have this merged into another branch I see the test_global_opts::test_version failing consistently 🤔

Ended up adding an allowance for version strings ending with a -dirty suffix for debug builds, please let me know if this is not something we want to have!

[ipetkov]

The changes in build.rs look good. I can't review the nix change, so I'll leave it to someone else.

I've dropped the Nix specific changes from this PR, I'm just as happy hard-coding export NIX_JJ_GIT_HASH=dirty in my .envrc which does what I want (avoiding unnecessary rebuilds when I amend commits).

With that said, @yuja do you feel comfortable approving the rest of the build.rs changes?

[yuja]

I don't know how nix sets -?irty suffix, the other changes look good, thanks.