[RFC][ORC] Revisiting Bug 23991

Hi all,

I would like to start another discussion on bug 23991 (https://bugs.llvm.org/show_bug.cgi?id=23991), regarding the missing reregistration of EH frames in ORCv2.

Compared to the time where the bugzilla thread was created, I think the structure of ORCv2 is fairly stable now. So I try to create a small patch trying to fix that: https://reviews.llvm.org/D71683

My concern is that whether this is the proper place to put the deregistration logic. And whether we should put a note to remind future MemoryManager implementations to take care of the deregistration since it’s not suitable to put deregistration in MemoryManager’s dtor as explained in the code review description.


Discussion’s probably fine to take place on the review - but I think Lang’s out of the office at the moment, possibly until the new year some time, so responses might be a bit delayed for review/design discussion/etc.