Open elyscape opened 8 years ago
Unfortunately I think you're right, I had a feeling that might be the case :(
here just testing an emit()
or read()
call at the JS layer, not the actual socket library's recv.
it('should not emit messages after pause()', function(done){
var push = zmq.socket('push')
, pull = zmq.socket('pull');
var n = 0;
pull.on('message', function(msg){
if(n++ === 0) {
msg.toString().should.equal('foo');
}
else{
should.not.exist(msg);
}
});
https://github.com/JustinTulloss/zeromq.node/blob/master/test/socket.push-pull.js#L42-L73
also wtf is should.not.exist(msg);
? Undefined? Null? exist
is not a type, IMO nor a convention in JS
I think should.not.exist is part of the mocking framework...
Could we have unref()
/ ref()
APIs just like the intrinsic Net.Socket objects?
I think we could implement that by passing poll_handle_
of binding.cc to uv_unref()
and uv_ref()
functions.
Could we have unref() / ref() APIs just like the intrinsic Net.Socket objects?
sounds interesting, what's the use case?
There may be cases that we want to skip checking pending RX/TX data. Just like the vanilla node sockets, one would want terminate the main program without explicitly closing sockets. I think this is particularly useful for ZMQ since it provides "asynchronous" semantics by its design, such as PUB/SUB sockets where the "actual" OS-level socket connection does not matter.
For me, I'm implementing an event-loop waiter as a C++ addon to nodejs that repeatedly calls uv_run()
until there is no more pending events to execute and wait until arbitrary user code finishes, where I do not have any prior information on what callbacks the input code would generate.
(This is what exactly the main loop of nodejs itself does to decide if it should terminate or not, for the entire program. Now I want to use the same blocking mechanism for just a specific region of codes.)
To make this work, I need to temporarily "unref" all existing callbacks such as timers and sockets.
BTW, my feature request may not be aligned with this thread, though I have crashed here after trying pause()
for my purpose. So I've created a new issue: #502.
I've found another bug concerning pause/resume. Socket.pause() only pauses AFTER flushing messages queued in the binding. User however might call pause() within 'message' handler but it won't stop flushing messages until the internal binding's queue is empty.
I think this line: https://github.com/JustinTulloss/zeromq.node/blob/7e2d99b33864fb54acec01b19542805257ffafa1/lib/index.js#L640 should be replaced with:
return !this._paused;
Thanks to this another _flushRead()
will NOT be called from _flushReads()
until user calls resume();
When calling
pause()
on a pull socket, messages continue to be retrieved in the background. The only difference is that no'message'
events are fired untilresume()
is called on the socket, at which point events are emitted for all previously-retrieved messages. This is unexpected behavior and should probably either be documented or fixed to behave as expected.