Cache

[jfkonecn]

closes #7517

Still a work in progress

[smores56]

Good start, excited to see caching of build artifacts for the first time since Roc was created!

[smores56]

On Unix, this should be used first when looking for the cache root, then $HOME/.cache/roc after that

[jfkonecn]

I made this a separate function because I didn't want the tests to break because someone set XDG_CACHE_HOME. I wanted to mock it out, but it's probably making the code too confusing so I'll refactor.

I'm sure we can figure out something better to solve the problem.

[smores56]

very minor nitpick: I think generally Zig uses error.NameOfError when creating an error variant from a custom error union.

[jfkonecn]

I added it so that I can pass a callback function for saving to the file system in saveToCacheInternal. I wanted to add the callback so that I can mock it out during testing.

[smores56]

I don't think I communicated well here, I meant I more see error.FailedToMakeCacheFolder instead of SaveCacheFileError.FailedToMakeCacheFolder

[smores56]

Two naming things:

• Roc is implied throughout this whole compiler, so no need to prefix names with it except in cases of ambiguity
• Zig prefers to nest related "has-a" data like the above CompilerCacheFile: https://ziggit.dev/t/nesting-struct-types-or-not/6750

An example of how I might nest it would be:

pub const CacheFile = union(enum) {
    compiler: Compiler,
    build,
    packages_src,
    packages_build,

    pub const Compiler = struct {
        version_name: []const u8,
        binary: []const u8,
    };
};

[bhansconnect]

@jfkonecn can you interactive rebase to either squash the commits or at least give them minorly meaningful names? A bunch of commits just with the name a aren't meaningful and nice to merge.

[jfkonecn]

Yeah, I was going to do that when I'm ready for an official review, but if you prefer I do that now I can do that as well.

[bhansconnect]

but if you prefer I do that now I can do that as well

Doesn't matter when. Just was noting in general.

[jfkonecn]

@smores56 this is a nitpick, but for getPackageRootAbsDir the allocator comes last in the arguments. It seems like the zig standard library generally has the allocator come first. I was thinking about sticking to that convention for this PR.

[smores56]

You're right, I was trying to write "Roc" code in spite of Zig standard patterns. That's been corrected in #7657

[jfkonecn]

Addressed with #7759