[PATCH] D101980: [RFC] [Coroutines] Put the contents of byval argument to the frame

Wed May 12 02:40:12 PDT 2021

ChuanqiXu added a comment.

In D101980#2753246 <https://reviews.llvm.org/D101980#2753246>, @rjmccall wrote:

> (I meant to type "That's *not* true", but I think you got the point.)

Yeah, I get your point is that there are cases `memcpy` can't work for movable classes even if they doesn't have non-trivial destructor. And you gives an example. I haven't looked into the example yet.

I had sent a mail to llvm-dev. Here I made a mistake that I send this to cfe-dev and llvm-dev which may disturb some people. Let's see if there is any opinions.

In D101980#2753331 <https://reviews.llvm.org/D101980#2753331>, @ychen wrote:

> I was thinking if pursuing D100415 <https://reviews.llvm.org/D100415> could solve this problem by marking the `init function` optnone until the frame is constructed.

I think we may still met the problem even if we split the coroutine. The key problem here is the coroutine didn't copy the argument right.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101980/new/

https://reviews.llvm.org/D101980