Unconditionally writing the config causes unnecessary (?) cargo cache busting

[david-boles]

Problem

Every time I rebuild my dioxus app, all crates that indirectly depend on manganis-macro are rebuilt.

Expected behavior

Only crates that have changes (e.g. my app) should be rebuilt.

Context

Running with cargo logging shows:

$ CARGO_LOG=cargo::core::compiler::fingerprint=info dx build --platform web
...
   0.193984191s  INFO prepare_target{force=false package_id=manganis-macro v0.3.0-alpha.1 target="build-script-build"}: cargo::core::compiler::fingerprint: fingerprint dirty for manganis-macro v0.3.0-alpha.1/RunCustomBuild/TargetInner { ..: custom_build_target("build-script-build", "/home/deb/.cargo/registry/src/index.crates.io-6f17d22bba15001f/manganis-macro-0.3.0-alpha.1/build.rs", Edition2021) }
   0.193991000s  INFO prepare_target{force=false package_id=manganis-macro v0.3.0-alpha.1 target="build-script-build"}: cargo::core::compiler::fingerprint:     dirty: FsStatusOutdated(StaleItem(ChangedFile { reference: "/home/deb/git/my-crate/rust/target/debug/build/manganis-macro-24dd372a8b03c9b1/output", reference_mtime: FileTime { seconds: 1726897562, nanos: 349187625 }, stale: "/home/deb/.cargo/assets/config.toml", stale_mtime: FileTime { seconds: 1726897797, nanos: 944207436 } }))
...

Which I surmise is because the config is written unconditionally, regardless of whether it already contains the config we want:

manganis/common/src/config.rs

Lines 56 to 65 in 63ee74f

	pub fn save(&self) {
	let current = Self::current();
	if current == *self {
	return;
	}
	let config = toml::to_string(&self).unwrap();
	let config_path = config_path();
	std::fs::create_dir_all(config_path.parent().unwrap()).unwrap();
	std::fs::write(config_path, config).unwrap();

I assume this resultant behavior isn't intended?

In the bigger picture, I don't fully understand the rational behind this global config file that gets re-written every build. That seems ripe for race conditions when you have parallel builds on the same machine? I definitely don't have the full picture here but I wonder if it would make sense to have the cli pass in a path to a temporary file containing the config through an environment variable? Well, I guess not temporary temporary since it would also need a stable path across builds.