merge commits in the SoT repo obliterate actual git history

[petergaultney]

we have an internal repo that we open source parts of.

the internal repo has the correct git blame on every line.

the external repo has the correct current code - but completely wrong git blame. Every line is a merge commit that had nothing to do with the file (indeed, nothing to do with anything anywhere near the file!).

the relevant bit of our config is

core.workflow(
    name="open-source",
    mode="ITERATIVE",
    origin=git.origin(
        url="file://INTERNAL_REPO_PATH_PLACEHOLDER",
        # this looks goofy, but I have been unable to find any alternative that will work
        # with dynamic input paths.
        # see https://github.com/google/copybara/issues/142#issuecomment-1782115935
        ref="HEAD",
        first_parent=False,  # try to get the underlying branch commits instead of just the merge commits.
        partial_fetch=True,
    ),
    destination=git.destination(
        url="file://EXTERNAL_REPO_PATH_PLACEHOLDER",
        fetch="main",
        push="open-source",
    ),
)

there must be something i'm doing wrong with our config, but I can't figure out what it is...

[petergaultney]

the public repo in question, where merges are messing with the git history. https://github.com/TrilliantHealth/trilliant-data-science/blob/main/libs/mops/docs/magic.adoc

we changed nothing in this merged PR that touches anything in this library at all.

[hsudhof]

I suspect this might be due to using HEAD here, with different fetch and push refs, but have to think about this.

[petergaultney]

I can try using the same ref (main) on Monday. this is only set up this way so that somebody manually verifying a new set of files to open source can run copybara on a branch locally. but the final run config can look different if that helps.

though I think in practice it generally ends up running on main in CI anyway.