Re: [PATCH 01/14] oidtree: modernize the code a bit
- From
brian m. carlson <sandals@crustytoothpaste.net>
- Date
- Mar 20, 2026, 22:30 UTC
- Message-ID
- <ab3KjF_1WW7hQBaA@fruit.crustytoothpaste.net>
- In-Reply-To
- <abzryk0qjlbvy8OL@pks.im>
On 2026-03-20 at 06:40:10, Patrick Steinhardt wrote:
Show 12 quoted lines
> On Thu, Mar 19, 2026 at 09:08:44AM -0700, Junio C Hamano wrote: > > I know the original also used GIT_MAX_HEXSZ to clamp the length for > > sanity, but because we know what algorithm is in use, I wonder if we > > want to use the limit more specific to it. > > That assumes that the passed prefix OID actually has an algorithm > attached to it, and that may not be the case. We could initialize the > overall oidtree with a hash algorithm in `oidtree_init()`, and if so we > can then become a bit more thorough with our asserts. > > But I feel like that would go beyond the smallish cleanups that I'm > doing in this patch.
We should stop assuming that a zero `algo` field in `struct object_id` means `the_hash_algo` because that makes libification hard and our Rust code doesn't support it (because accessing mutable globals without a lock is unsafe)[0]. So in general, I would be fine with forcing callers to set an algorithm per OID, both here and elsewhere in our code.
However, I am also fine with doing that in a different series for the sake of minimalism in this one. I will probably get to that at some point if nobody else does.
[0] Rust also typically initializes all fields explicitly (and zero-initialization is also unsafe), so there's no urge to be lazy and do `memset(p, 0, sizeof(*p))`, which is the usual source of the zero `algo` fields in our codebase.
-- brian m. carlson (they/them) Toronto, Ontario, CA