Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Split poller_t into two layers #225

Merged
merged 9 commits into from
May 11, 2018
Merged

Conversation

sigiesec
Copy link
Member

New poller_t (in zmq.hpp) is a direct mapping of the libzmq zmq_poller_* API.
New active_poller_t (in zmq_addon.hpp) is the previous poller_t, which is implemented based on the new poller_t, and provides a higher-level abstraction that actively manages callback handlers.

The PR contains some fixes to make zmq_addon.hpp compile again. It was broken before.

Solution: extract base_poller_t from poller_t, which provides a direct mapping of zmq_poller_* to C++ only
Solution: remove redundant assertions (tested elsewhere), and add assertion behaviour of calling wait on moved-from poller
Solution: convert to googletest based test, and fix syntax error
Solution: move to zmq_addon.hpp, rename to active_poller_t, and rename base_poller_t to poller_t
Solution: change to local variables
Solution: use non-deprecated operator!= instead
Solution: add tests
Solution: removed redundant assertion
@sigiesec sigiesec changed the title Split poller layer Split poller_t into two layers May 11, 2018
@coveralls
Copy link

coveralls commented May 11, 2018

Pull Request Test Coverage Report for Build 81

  • 0 of 0 changed or added relevant lines in 0 files are covered.
  • 2 unchanged lines in 1 file lost coverage.
  • Overall coverage increased (+0.7%) to 56.923%

Files with Coverage Reduction New Missed Lines %
zmq.hpp 2 64.38%
Totals Coverage Status
Change from base Build 80: 0.7%
Covered Lines: 37
Relevant Lines: 65

💛 - Coveralls

@coveralls
Copy link

Pull Request Test Coverage Report for Build 81

  • 0 of 0 (NaN%) changed or added relevant lines in 0 files are covered.
  • 2 unchanged lines in 1 file lost coverage.
  • Overall coverage increased (+0.7%) to 56.923%

Files with Coverage Reduction New Missed Lines %
zmq.hpp 2 64.38%
Totals Coverage Status
Change from base Build 80: 0.7%
Covered Lines: 37
Relevant Lines: 65

💛 - Coveralls

@bluca bluca merged commit 0f840ce into zeromq:master May 11, 2018
@sigiesec sigiesec deleted the split-poller-layer branch May 11, 2018 12:57
Copy link
Contributor

@kurdybacha kurdybacha left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good stuff.

@@ -8,117 +7,98 @@

TEST(poller, create_destroy)
{
zmq::poller_t poller;
ASSERT_TRUE(poller.empty ());
zmq::poller_t<> poller;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No asserts? :)

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I somehow waited for this question ;)

Well, poller_t has no state of its own, that could or would need to be tested. zmq_poller_* do not allow to query anything at the moment. Maybe that would be a good idea, and then an assertion using that could be added here.

@@ -1019,75 +1018,42 @@ namespace zmq
};

#if defined(ZMQ_BUILD_DRAFT_API) && defined(ZMQ_CPP11) && defined(ZMQ_HAVE_POLLER)
template <typename T = void>
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a need for that? Everything is casted to void* anyway. Besides no tests exercising different types.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It makes it typesafe when adding. active_poller_t uses its handler_t as the template argument. There is at least one test using int as an argument. I think it makes sense.

Copy link
Contributor

@kurdybacha kurdybacha May 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

but poller class is template, not add. There is a use of int in add but with poller<void*> as I can see. What is a point of having poller template as you need to cast it to void* in order to pass to libzmq. It would be ok to have it as template if at least you have a compilation error when trying to pass different type to add than poller template type in use. Right now it does not and your test with int is a proof :)

#endif // ZMQ_HAS_RVALUE_REFS

#if defined(ZMQ_BUILD_DRAFT_API) && defined(ZMQ_CPP11) && defined(ZMQ_HAVE_POLLER)
class active_poller_t
Copy link
Contributor

@kurdybacha kurdybacha May 11, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Much cleaner now after moving this to separate class. Can we change a name though, please? Something more zmq'ish and shorter? Maybe loop_t, or service_t? I have some pending additions - timers - and it would make sense to put it here. Maybe we can split this actually into separate file? zmq_monitor.hpp, zmq_service.hpp?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am not completely satisfied with my name as well. I think "poller" should be part of the name, but "active" is not so specific. loop_t would be strange IMO, as there is no loop within wait. A "service" is also something quite different for me, which has its own thread etc. These concepts could be built on top of active_poller_t again.

About splitting up the headers... I am somewhat undecided about that. It is a feature of cppzmq that it consists only of two header files. But on the other hand, when you can copy two header files, you can also copy a handful. If it is done, zmq_addon.hpp should include all the extra headers then.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yeah, naming is the most difficult part in software development. maybe zpoller_t (z - zuper) :)
We can stay with two header for now. If have more stuff we can split then and I like an idea of zmq_addon.hpp to have all , probably with a rename to zmq_addons.hpp

int wait (std::chrono::milliseconds timeout)
{
if (need_rebuild) {
poller_events.resize (handlers.size ());
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is wrong, sorry. resize does not clear items already in the vector so you might pass items from previous run to wait. Depending on this kind of stuff is unreliable and asking for trouble.

Copy link
Member Author

@sigiesec sigiesec May 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's why I changed this. It is not necessary to clear the items. zmq_poller_wait_all even clears all unused items. But this is unnecessary too. zmq_poller_wait_all returns the number of valid items, and we only access those.

If this were different, you would need to clear all items on every call to wait, regardless of the value of need_rebuild. But this would be quite inefficient if you had a large pollset.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I get a point about performance but my issue was that we rely on internal implementation of libzmq to clear them (set all members) for us. How about if libzmq decides to set only polleritem partially depends on the case e.g. ZMQ_POLLERR etc.. That was my original reason behind clear().

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the previous version did not clear the members of the event struct either.

We shouldn't rely on implementation details of course, but I don't think we do. We should rely on the specification of the API, which should specify that the indicated number of entries are properly set to be well usable IMO. However, I fear the zmq_poller_* functions are not properly documented yet. I will put this forward on the libzmq side.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sounds good.

void add (zmq::socket_t &socket, short events, handler_t handler)
{
auto it = decltype (handlers)::iterator {};
auto inserted = bool {};
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

depending on default initialization here makes it only harder to read.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm. I don't think so. Initializing with a value explicitly for which there is no code path where it could ever be read would be confusing IMO.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There is not difference in initializing it explicitly or default. But there is readability and maintainability gain. Explicit initialization makes it harder to break things in the future IMO. Your call here.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can understand your point of view regarding future changes. I disagree about readability though. If I read some variable is initialized with a specific value I assume this value will be used in some case, and search for a long time if there is no such case.

Anyway, there is no way without some disadvantage. What is more important is to do this consistently throughout cppzmq IMO (I am not sure if there is any other comparable place at the moment). We should prescribe either way in a brief style guide.

Copy link
Contributor

@kurdybacha kurdybacha May 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok. Agree.
We would not have this discussion if we switch active poller to C++17. What do you think about that? Most of the compilers support it now and that would give some more stuff for cleaner and safer code.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think this would be a good idea at the moment. I read that people can't use fbzmq e.g. because it requires C++14 support. For our own uses we will stick to VS2015 for several years to come e.g., which has only very limited support for C++14 or even C++17 features. I think there are also many other environments that at least have no recent gcc/clang by default. We already restrict the platforms quite substantially compared to libzmq itself which does not require C++11.

try
{
base_poller.add (socket, events, inserted && *(it->second) ? it->second.get() : nullptr);
need_rebuild |= inserted;
Copy link
Contributor

@kurdybacha kurdybacha May 11, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why not simply need_rebuild = true ?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In order to avoid an unnecessary rebuild?
IIRC it was like this before, just something like

if (inserted)
{
  need_rebuild = true;
  ...
}

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You not avoiding here anything. If you are after base_poller.add you have to rebuild regardless of inserted.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, I think add will throw necessarily if inserted is false, so the need_rebuild line will only be reached if inserted is true anyway. So... it could be changed...

need_rebuild = false;
}
const int count = base_poller.wait_all (poller_events, timeout);
if (count != 0) {
Copy link
Contributor

@kurdybacha kurdybacha May 11, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

count > 0 , wait_all should return unsigned or size_t.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I agree, that would be better.

Copy link
Member Author

@sigiesec sigiesec May 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, the condition is completely redundant then. The for_each below just iterates over an empty range if count == 0.

ASSERT_TRUE(message_received);
}

/// \todo this contains multiple test cases that should be split up
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we remove this comment?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you think it no longer applies?

Copy link
Contributor

@kurdybacha kurdybacha May 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it has never applied. All separate cases are already covered? This is a test that exercises more complex scenario - splinting it up does not make sense because you loose the intend of it.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, then remove it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants