Patrick Steinhardt says he was "appalled" by git-rerere's custom subcommand parsing while reviewing other rerere patches. His 12-patch series modernizes it so rerere behaves like other Git commands that have subcommands. It is based on the 9th batch of topics, with ps/rerere-remove-global-state merged in.

The early patches rename repo_rerere(), convert flags into an enum, drop the_repository from builtin/rerere and stop passing state around. Further patches refactor each subcommand (gc, diff, remaining, status, clear) to use parse-options as preparation. The final switch to OPT_SUBCOMMAND() removes the custom parsing, and another patch adds proper usage strings for each subcommand, with a documentation update.

The switch is backwards-incompatible. git-rerere no longer accepts "--[no-]rerere-autoupdate" for every subcommand. Steinhardt argues the old behavior was broken and misleading, since the flag was ignored by every subcommand but one, and says the series "cheats a bit" by calling the earlier acceptance a bug.

Junio C Hamano commented on the explanation of RERERE_NOAUTOUPDATE. He said the comment in the newer version of that patch was not wrong but added nothing over the earlier one. He said both versions fail to explain why RERERE_NOAUTOUPDATE is needed at all, meaning how it differs from the absence of RERERE_AUTOUPDATE, and called that the more important thing to clarify.

The excerpts show no other substantive objections, including to the compatibility change. Participants in the thread also include Karthik Nayak and Mark C. Chu-Carroll, but their comments are not in the excerpts.