Re: [PATCH] commit: fall back to full read when maybe_tree is NULL

[Derrick Stolee <stolee@gmail.com>]

On 5/19/2026 2:15 AM, Jeff King wrote:
> On Tue, May 19, 2026 at 02:56:51PM +0900, Junio C Hamano wrote:

>> Looks quite straight-forward.  Don't you need to pay attention to
>> r->hash_algo and call parse_oid_hex_algop() instead?
>>
>> Or are we pretty much sure that "r" is always "the_repository" here,
>> in which case parse_oid_hex() that uses "the_hash_algo" would be
>> sufficient?
> 
> No, I didn't even think about it, since the use of the_hash_algo is
> hidden behind the function. We definitely should use the hash algo from
> "r", since we have access to it. I'm not even sure if you can have repos
> of two different hashes loaded in the same process at this point, but
> certainly it is the correct long-term direction.
> 
> Here's a re-roll with the one-line fixup:
> 
>     diff --git a/commit.c b/commit.c
>     index cfc87ad185..499a9602ad 100644
>     --- a/commit.c
>     +++ b/commit.c
>     @@ -448,7 +448,7 @@ static void load_tree_from_commit_contents(struct repository *r, struct commit *
>      
>      	if (type == OBJ_COMMIT &&
>      	    skip_prefix(buf, "tree ", &p) &&
>     -	    !parse_oid_hex(p, &tree_oid, &p) &&
>     +	    !parse_oid_hex_algop(p, &tree_oid, &p, r->hash_algo) &&
>      	    *p == '\n')
>      		set_commit_tree(commit, lookup_tree(r, &tree_oid));
>      

I figured that this was already tested via the test variable that
runs the test with SHA256, but the multi-repo case is an interesting
one that I'm sure would catch us at some point in the future.

I'm happy with the re-roll here.

Thanks,
-Stolee