RFC: Rename LibGit2.Oid to LibGit2.GitHash

[ararslan]

This addresses one of the points in #19839: renaming Oid to GitHash in LibGit2. I only did this one rename in this PR, since the diff is fairly large as it is.

cc @simonbyrne

[simonbyrne]

Fantastic thanks. However there is one other thing that has since occurred to me that we may want to change.

Command line git accepts both full and shortened hashes whenever they're unique. LibGit2 implements this by providing:

• git_oid_fromstrn which pads the rest of the hash with zeros
• various git_*_lookup_prefix functions (e.g. git_object_lookup_prefix) which only looks at the first few hex digits of the hash

Currently this is exposed in a somewhat cumbersome manner. e.g. here you need to manually provide the length of the hash.

My current thinking is to define 2 types:

• GitHash as defined in this PR
• GitShortHash as

immutable GitShortHash <: AbstractGitHash
    hash::GitHash
    len::Csize_t
end

Additionally:

• GitHash(::String) would throw an error if the string is too short.
• Define a macro githash_str which would create the appropriate object depending on the input length.

[ararslan]

I've implemented bits and pieces of what you outlined, Simon. It's turning into a somewhat bigger job than I had intended this PR to be, however, so is there any chance I could do that as a separate PR? I think that alone will produce a sizable diff, and for those changes I'd want to get more focused feedback on design aspects, etc. Does that seem reasonable? If not I'll just keep chipping away at it here.

[simonbyrne]

No worries, sorry for lumping that on you. Add whatever you've done, and I'll have a look over the weekend to see if I can sort it out.

[ararslan]

I've pushed my interpretation of the plan you outlined. Will add some tests as well for the new functionality.

[ararslan]

Alright, well those changes broke Pkg, Documenter, or both...

[MichaelHatherly]

Seems like it failed on Pkg.update, not in Documenter itself. As far as I recall I've not been using anything from Base.LibGit2 aside from GITHUB_REGEX.

[tkelman]

does this usually get passed as a vector of one or the other concretely-typed hash objects?

[ararslan]

Before this change, it's always been a vector of concretely-typed hash objects because there was only one hash type. I didn't think letting this be either would cause a problem, since the hashes are passed to get individually, and the separate get methods return the same type.

[tkelman]

well now it might be a Vector{GitHash} or Vector{GitShortHash} which are not subtypes of Vector{AbstractGitHash}

[ararslan]

Oh I see what you mean. Hm.

[ararslan]

Good to know, thanks @MichaelHatherly.

[ararslan]

For now I'm just going to remove the commits that extend functionality and return this PR to its initial scope of simply renaming Oid. I'll work on the other changes some more and post it as a separate PR once I've got things figured out.

[ararslan]

Fascinating. My initial renaming commit had originally passed its tests but now they're failing. Hmm.

[tkelman]

are there new occurrences now on master from anything that got merged in the interim?

[ararslan]

You're a wizard, Tony. That seems to have been the problem.

[ararslan]

FYI: I've pushed the stuff with GitShortHash that I had been working on here to a separate branch in my fork, which I will continue to update: https://github.com/ararslan/julia/tree/aa/gitshorthash

[tkelman]

should be repo now

[ararslan]

Hey, thanks so much for the extra commits, Simon! It must be that "allow edits from maintainers" feature that allows you to do that. Much appreciated! 👍

Is this good to go now?

[simonbyrne]

Thanks!

[bjarthur]

NEWS.md incorrectly says GitOid (instead of Oid) was renamed to GitHash.

also, it would've been nice to add this to Compat, though at this late stage maybe not worth the effort.