♪ Any singleton will do #20273
♪ Any singleton will do #20273
Conversation
adamemerson
force-pushed
the
wip-any-singleton-will-do
branch
from
68e1d49
to
a43d197
Compare
src/include/any.h
Outdated
| */ | ||
|
|
||
| #ifndef INCLUDE_STATIC_ANY | ||
|
|
There was a problem hiding this comment.
i think you intended to put #define INCLUDE_STATIC_ANY here.
adamemerson
force-pushed
the
wip-any-singleton-will-do
branch
from
a43d197
to
e5cdeb2
Compare
|
Oops. Well, thank you. |
src/librbd/Journal.cc
Outdated
| m_work_queue = new ContextWQ("librbd::journal::work_queue", | ||
| cct->_conf->get_val<int64_t>("rbd_op_thread_timeout"), | ||
| thread_pool_singleton); | ||
| cct->_conf->get_val<int64_t>("rbd_op_thread_timeout"), |
There was a problem hiding this comment.
please drop these formatting only changes. otherwise we will have them whenever developer touches a file with different editor settings.
There was a problem hiding this comment.
I'll pull these out and drop this in a moment.
src/common/ceph_context.h
Outdated
| }; | ||
|
|
||
| // I would use std::type_index, but the version of the library on | ||
| // gitbuilder doesn't support it for whatever reason. |
There was a problem hiding this comment.
weird ... we should have std::type_index. could you point me to the build failure? also, we are not using gitbuilder anymore. we are using jenkins for building/testing the PRs.
There was a problem hiding this comment.
It was the first push I did for this branch. I'm not sure if they're kept. I could stick type-index back in and repush to show you.
std:any is nice and all, but they don't support uncopyable objects. This is annoying to us, since we have a perfect use case for std::any in the associated object map, but almost everything we want to put into it is uncopyable due to having a mutex or something. Enter ceph::immobile_any<>. A zero overhead type erasure that cannot be moved or copied (you stick it in a container and it stays there) of fixed size that does no allocation. Also unique_any and shared_any, because occasionally move and copy semantics are useful but you still want to be able to store an arbitrary type. Signed-off-by: Adam C. Emerson <aemerson@redhat.com>
This cleans up the interface, makes things more robust by using both name and type, and does fewer allocations. Signed-off-by: Adam C. Emerson <aemerson@redhat.com>
adamemerson
force-pushed
the
wip-any-singleton-will-do
branch
from
e5cdeb2
to
143c59c
Compare
|
@tchaikov All right! I have made all the changes. |
| std::type_index type = typeid(T); | ||
|
|
||
| auto i = associated_objs.find(std::make_pair(name, type)); | ||
| if (i == associated_objs.cend()) { |
There was a problem hiding this comment.
Ok, trying to add a duplicate is /not/ a failure.
Remove the Singleton class and dynamic allocation with ceph::immobile_any. This is similar to std::any except that it allows immovable/uncompyable objects and does no allocations.
Also clean up the interface a bit.
Also actually /get rid/ of the various singletons we're holding on shutdown rather than just letting them dangle.