Community
Participate
Working Groups
Currently the profile lock is re-entrant. Pretty cool right... I'm beginning to think it really isn't so cool like when John begins to show me things like bug 266205. There are some nasty cases where the inner transaction could update the profile and later on the outer transaction would merrily overwrite those changes. I think the profile lock should probably be a regular old style non-reentrant fail-fast mutex until we know what we're getting into.
As we learned in bug 266881, re-entrant engine calls can cause all sorts of bad things to happen, as we are often making the assumption that there is only one active engine session. We should make this change soon to protect against more failures like that one.
Fixed in HEAD