Patrick Steinhardt says he is working towards making commit-graphs pluggable, so that backends can control where commit-graphs are written and read. The series removes global state from builtin/commit-graph.c, replaces a helper that allocated a secondary options array with a preprocessor macro, and replaces references to the_repository with the repository passed to cmd_commit_graph().
The patch he expected might raise eyebrows changes the handling of "--object-dir=". Today both "git commit-graph --object-dir= write" and "git commit-graph write --object-dir=" work. He calls the former very atypical in Git; it is undocumented and no test uses it. Accepting it requires global state to carry the parsed option. He asked for extra thoughts on that change.
Junio Hamano read the first four patches and found them sensible. He called patches 6 and 7 "Meh" but did not think they hurt. He noted that until such code is lifted out of builtin/, there is no strong reason to avoid global variables or the_repository. Mark Chu-Carroll replied that everything looked correct and unobjectionable.
The replies so far do not directly address the --object-dir change.