Re: [PATCH] [RFC][GSoC][PATCH] attr: use local repository state in read_attr

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 10, 2026, 23:37 UTC
Message-ID: <xmqqqzqsw4og.fsf@gitster.g>
In-Reply-To: <96329bc6-0490-454b-a21b-babb85c98bc9@gmail.com>

Tian Yuchen <a3205153416@gmail.com> writes:

Show 7 quoted lines

> Junio C Hamano <gitster@pobox.com> writes:
>
>  >The codepath read_attr() is in is usually not that hot but it is not
>  >cheap.
>
> I'm a bit curious—under what circumstances would calling this method 
> result in significant performance regression?

Significant?  I dunno.

And quite honestly, I do not care about significance very much in a case like this. Doing things that do not make sense, like checking the same configuration variable again and again when you _know_ that you never switched to a different repository since you last checked, is simply wrong. It burdens the readers with unnecessary cognitive load by making them wonder why you do such a nonsensical thing.

The read_attr() is called during an attr stack construction, which traverses the directory hierarchy of a single repositry's working tree (we do not traverse across submodule boundaries), and the same istate (i.e., index contents) structure is passed around throughout the callchain. The repository instance at istate->repo may be a good place to store "am I bare?" bit that is computed just once and reused whenever we need to know, like in the funcion under discussion.