Patrick Steinhardt relayed a report from a user on IRC: running git log --stat --follow -- include/linux/rsc_table.h in linux.git takes about 33 seconds with Git 2.55 and almost 8 minutes with Git 2.56. He reproduced it and bisected it to 304812ed33 (log: improve --follow following renames for non-linear history). He forwarded the report to the list and copied the original author.

The thread has since moved to review of a fix. Steinhardt pointed out that an assumption in the patch does not hold for copies. A test identical to the one added in 304812ed33, but using a copy rather than a rename, behaves differently.

He also raised the case of unrelated histories. There the merge-base computation would return no bases at all, so the check would always return 0. Commit 304812ed33 explicitly mentioned sub-tree merges of unrelated histories as a use case, so he asked whether that is itself a regression.

Steinhardt also told Miklos Vajna that reviewers expect replies to review comments directly in the thread, not only in the next version's cover letter. He asked why the speedup patch is a separate commit when it introduces a regression that only appeared in the preceding commit. He asked for a range-diff rather than an interdiff next time.

The excerpts show no agreed fix yet.