Re: Re* using tree as attribute source is slow, was Re: Help troubleshoot performance regression cloning with depth: git 2.44 vs git 2.42

From: Jeff King <peff@peff.net>
Date: May 3, 2024, 17:46 UTC
Message-ID: <20240503174653.GD3631237@coredump.intra.peff.net>
In-Reply-To: <xmqqzft6aozg.fsf_-_@gitster.g>

On Fri, May 03, 2024 at 08:34:27AM -0700, Junio C Hamano wrote:

> And for folks who had been happy with the pre 2.42 behaviour,
> we could do something like the attached as the first step to a real fix.

It looks like lots of discussion happened with out me, and everybody already posted all of the responses I was going to. Good. :)

In particular...

Show 7 quoted lines

> ----- >8 --------- >8 --------- >8 --------- >8 -----
> Subject: [PATCH] stop using HEAD for attributes in bare repository by default
> [...]
> The right fix for this is to optimize the code paths that allow
> accesses to attributes in tree objects, but that is a much more
> involved change and is left as a longer-term project, outside the
> scope of this "first step" fix.

...this was the exact first step I was going to suggest. And your patch looks correct to me. I assume you'd target this for 'maint'. The regression goes back to v2.43.0, so it's not exactly new, but given the severity in some cases it seems like it's worth getting it into a release sooner rather than later.

I am mildly surprised nobody noticed the issue until now. I wonder if t/perf would notice it and nobody is running it, or if this is a gap in our coverage there. If the latter, it might be worth adding such a script, which should be able to show off that your change here takes us back to the v2.42 state.

Running the perf suite against linux.git between 2.42 and 2.43 would answer the "is this a gap" question, but I haven't had a chance to do so, and it takes a while.

-Peff