Re:Regression in Linux 4.14 – using bcache can destroy the f...
by bonzini on Wednesday November 22, 2017 @11:21AM
(#15757982)
> If you don't override it, it will copy everything properly.
Sure, so does memcpy or bare struct assignment in C. The problem is that trivial copy constructors are relatively rare.
> Here even if you could not just have derived Clone (struct bio looks non-trivial) your hand-rolled implementation would have complained that you had not properly initialised the member.
Right, but this is more similar to an assignment operator. DerefMut in Rust I think wouldn't have caught the bug.
Re:Regression in Linux 4.14 – using bcache can destroy the f...
by masklinn on Wednesday November 22, 2017 @11:53AM
(#15758273)
> Right, but this is more similar to an assignment operator.
Only if you don't zoom out. If you do, this is alloc/init/fill which in C++ or Rust would use a regular copy and RVO.
If you don't override it, it will copy everything properly.
> Rust might have put a 0 in the uninitialized member
Rust does not do that. If a member is not initialised, it's uninitialised and in this case the structure will be rejected. Here even if you could not just have derived Clone (struct bio looks non-trivial) your hand-rolled implementation would have complained that you had not properly initialised the member.