On Tue, Sep 25, 2018 at 1:57 AM Casey Bodley <cbodley@xxxxxxxxxx> wrote: > > > > On 09/24/2018 01:39 PM, Sage Weil wrote: > > On Mon, 24 Sep 2018, Patrick Donnelly wrote: > >>> - The only code that has been converted to ceph::mutex so far is Finisher. > >>> You can see what that transition looks like here: > >>> > >>> https://github.com/ceph/ceph/pull/24133/commits/8535965ba4298537c489897ccb191a0f458d60d7 > >>> > >>> - lock("Foo::lock") -> lock(ceph::make_mutex("Foo::lock")) in ctor > >> It's odd seeing what looks like a move/copy constructed mutex. If I'm > >> understanding the `make_mutex` code correctly, this just selectively > >> discards the arguments to the lock ctor if not debug mode? > > Right. I'm fuzzy on the underlying mechanics of the factory function, but > > it seems to work! > > > > sage > > While this looks like it's trying to use mutex's deleted move > constructor, the factory function relies on return value optimization > (now guaranteed in c++17) in order to construct the mutex in place. yeah, it's now known as "copy elision", so namely not considered an optimization anymore in C++17 =). so we are safe. to avoid repeated work, i am switching src/auth to debug::mutex, see https://github.com/ceph/ceph/pull/24263 . -- Regards Kefu Chai
