Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> writes: > Don't pass the object ID we pass into reflog_expire() back to the > caller, but rather our locked OID. > > As the assert shows these two were the same thing in practice as we'd It is unclear which assert you refer to, but a call to verify_lock() near the end of lock_ref_oid_basic() ensures this, I presume? > exit earlier in this function if we couldn't lock the desired OID, but > as part of removing the passing of the OID to other functions further > on I'm splitting up these concerns. > > As we'll see in a subsequent commit we don't actually want to assert > that we locked a given OID, we want this API to do the locking and > tell us what the OID is, but for now let's just setup the scaffolding > for that. OK. That sounds like a sensible direction to go in. > diff --git a/refs/files-backend.c b/refs/files-backend.c > index af332fa8fe4..ec9c70d79cc 100644 > --- a/refs/files-backend.c > +++ b/refs/files-backend.c > @@ -3098,7 +3098,7 @@ static int files_reflog_expire(struct ref_store *ref_store, > } > } > > - (*prepare_fn)(refname, oid, cb.policy_cb); > + (*prepare_fn)(refname, &lock->old_oid, cb.policy_cb); > refs_for_each_reflog_ent(ref_store, refname, expire_reflog_ent, &cb); > (*cleanup_fn)(cb.policy_cb);
