Depends On D110941
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
I am not sure why this is desirable?
I expect the ternary state to be useful in the future: None should be the catchall case that makes copies of everything it needs to.
My reasoning was that ops that we don't know about do not have an aliasingOpOperand. Therefore, nullptr is a valid return value. (Judging just from the function name, without reading the comment.) We would call hasKnownBufferizationAliasBehavior first when the op may be an unknown one. (It is not always the case.) This would make getAliasingOpOperand symmetrical to getAliasingOpResult, which also does no longer return Optional<>.
I am still on the fence about this one.
Note that in the other bigger CL, I point thatwe should prob. move to returning a list now.
Can we make that API change (which may also impact getAliasingOpResult) and see where we stand once it is in?
No need to support any new case, we would just bail on any list of size != 1 for now.