Re: [PATCH v2 00/13] Optimization batch 11: avoid repeatedly detecting same renames

On Mon, May 3, 2021 at 7:12 PM Elijah Newren via GitGitGadget
[off-list ref] wrote:

This series avoids repeatedly detecting the same renames in a sequence of
merges such as a rebase or cherry-pick of several commits. It's
unfortunately become a bit lengthy, but much of the length (the first five
patches) is owed to special testcases and documentation.

Since this obviously hasn't inspired much review, let me note that you
can cut out 90% of the review size by skipping patches 2 & 5.

Patch 2 is essentially written as something approaching a formal
proof, so yes it's dense and lengthy, but it's not at all required;
there's no code there.  Think of it as insurance for if someone wants
to introduce some new tricky optimizations or radically different
features to the merge machinery, because the remember-renames
optimization by its nature tends to interact with other optimizations.
I figured because of that interaction that documenting why and how the
remember renames optimization works at a much deeper level than is
typical for optimizations or code in git in general that it might help
with future maintenance...and it happened to help me catch two minor
bugs.

Patch 5 is very much related to patch 2; it's testcases inspired by
that document.  Most of those tests were just "what could possibly go
wrong in a new from-scratch implementation of this optimization?"
based on what's written in this proof-like document.  Most of the
tests didn't turn up anything, but a couple found some small issues in
my implementation.  I decided to just include all of them; it's nice
to be thorough.

You can get 95% of the whole idea behind this optimization skipping
those patches and reading Junio's great two-paragraph summary at
https://lore.kernel.org/git/xmqqzgyrukic.fsf@gitster.g/ (local), and then just
read the other patches in this series.