Removing merge commits that could have been fast-forward

[raphaelcoeffic]

Here is the setup:

• I have a git subset of a mono-repo that was previously done with the filter-branch (and very slow)
• I managed to produce a python script that works with git-filter-repo to help me obtaining the exact same result (important for me)
• I run multiple passes:
  • first some positive directory filters,
  • then a second pass with negative directory filters,
  • and then at last a sub-directory filter together with my commit-callback that irons out the last bit.

The script used for the last pass is listed here (the first 2 use regular git-filter-repo command line):

#!/usr/bin/env python3

import sys

try:
  import git_filter_repo as fr
except ImportError:
  raise SystemExit("Error: Couldn't find git_filter_repo.py.  Did you forget to make a symlink to git-filter-repo named git_filter_repo.py or did you forget to put the latter in your PYTHONPATH?")


def nuke_empty_merge_commits(commit,metadata):
    if len(commit.parents) > 1:
        agraph = metadata['ancestry_graph']
        if agraph.is_ancestor(commit.parents[0], commit.parents[1]):
            commit.parents = [commit.parents[1]]
            commit.file_changes = None
        elif agraph.is_ancestor(commit.parents[1], commit.parents[0]):
            commit.parents = [commit.parents[0]]
            commit.file_changes = None

args = fr.FilteringOptions.parse_args(sys.argv[1:])
if args.analyze:
  fr.RepoAnalyze.run(args)
else:
  filter = fr.RepoFilter(args, commit_callback=nuke_empty_merge_commits)
  filter.run()

The small issue here is that it seems I should not be using metadata, as it seems to be part of the private API, and is thus not available when passing the function directly in the command line with --commit-callback (argument is generated as _do_not_use_this_var in _handle_arg_callbacks() / make_callback()).

For reference here is that what I think would be better for me to use (which seems quite odd and risky for compatibility with future versions):

git filter-repo --preserve-commit-hashes --prune-degenerate always \
    --prune-empty always --subdirectory-filter src \
    --commit-callback '
    if len(commit.parents) > 1:
        agraph = _do_not_use_this_var["ancestry_graph"]
        if agraph.is_ancestor(commit.parents[0], commit.parents[1]):
            commit.parents = [commit.parents[1]]
            commit.file_changes = None
        elif agraph.is_ancestor(commit.parents[1], commit.parents[0]):
            commit.parents = [commit.parents[0]]
            commit.file_changes = None
'

So I guess the basic question is: how can I find out that one of the parents of a merge commit has a common ancestor with the other, and I shall thus evict this commit?

Edit: just found this (git blame is cool :-)): 0b70b72
It suggests I should continue to use my separate Python script rather than using vanilla command line, is that correct?