Is there a reason to to add the mapping to the per-commit metadata in the repository (i.e. the git notes for each commit), and read that? We could consider a one-time import of the CSV file to backfill existing data (but that could be overkill; I think it's pretty likely we ~never read these older commits, unless I'm forgetting something).

I think the naming here is getting confusing because for mozilla-central we have three different possible revisions, two of which are git. Maybe we should call it canonical_rev_git. Here the implementation would look like:

@property
def canonical_rev_git(self) -> str:
    if self.cinnabar:
        raise ValueError(f"Commit {self.sha1} doesn't have a canonical git SHA1")
    return self.sha1

But then in GeckoCommit you'd override it to:

@property
def canonical_rev_git(self) -> str:
    if self.cinnabar:
        if "gecko-commit-git" not in self.notes:
            self.notes["gecko-commit-git"] = hg2git(self.sha1)
        sha1 = self.notes["gecko-commit-git"]
        assert sha1 is not None
        return sha1
    return self.sha1

Ok, updated