Re: [PATCH 2/3] path-walk: fix setup of pending objects

From: Derrick Stolee <stolee@gmail.com>
Date: Aug 20, 2025, 19:42 UTC
Message-ID: <c6ab0985-2a74-406b-9685-cc65f8748515@gmail.com>
In-Reply-To: <xmqqh5y1ak9o.fsf@gitster.g>

On 8/20/2025 3:02 PM, Junio C Hamano wrote:

Show 15 quoted lines

> "Derrick Stolee via GitGitGadget" <gitgitgadget@gmail.com> writes:
> 
>> The core problem here is that the "maybe_interesting" member of 'struct
>> type_and_oid_list' is not initialized to '1'. This member was added in
>> 6333e7ae0b (path-walk: mark trees and blobs as UNINTERESTING,
>> 2024-12-20) in a way to help when creating packfiles for a small commit
>> range using the sparse path algorithm (enabled by pack.useSparse=true).
> 
> OK, in other words, the bug is fairly contained within the path-walk
> traversal.  We treat things as reachable not just from ref tips and
> reflogs (where path-walk code can use the tree object to compute on
> what pathname each blob comes from) and the main index array (that
> has paths, even though it needs separate way to compute than those
> for trees), but also from places like REUC and TREE extensions that
> make associations between pathnames and objects.  Are they also OK?

The key integration point is the "pending" list operating a bit different from walking directly from tags or commits. I was trying to reproduce the issue from all of those other sources before unlocking the "singleton" nature of the problem, and failed to do so.

The resolve-undo cache (REUC) is something that I had not tested previously. Adding "git rm --cached x/y" to the test in the previous case leads to the 'git fsck' call giving a "dangling blob" warning, so that could be an interesting way to strengthen the test. Thanks, -Stolee