ZeroMq IRC Log

Sunday April 11, 2010

[Time] Name	Message
[05:59] CIA-5	zeromq2: 03Martin Sustrik 07master * r6fea422 10/ include/zmq.hpp : getsockopt added to c++ binding - http://bit.ly/aWN1od
[08:08] mikko	sustrik: pretty early commit, huh
[08:08] mikko	(assuming you are somewhat close to GMT)
[08:08] sustrik	GMT+1
[08:09] sustrik	what about you?
[08:09] mikko	GMT + 1 at the moment i think
[08:09] mikko	british summer time
[08:09] mikko	(london)
[08:10] sustrik	ah, GMT+2 here then (CET summer time)
[08:10] mikko	is imatix a belgian company?
[08:11] mikko	at least the contact details seem to point to belgium
[08:11] sustrik	imatix is belgain, but i am independent, just managing the project
[08:11] sustrik	i'm based in bratislava, slovakia
[08:12] sustrik	btw, i'll have to go to london next week or so, so we can have a beer
[08:12] mikko	sure, just let me know when and where
[08:12] sustrik	ack
[08:12] mikko	i visited bratislava in 2006
[08:13] sustrik	haven't changed much since :)
[08:13] mikko	after football world cup
[08:13] mikko	a lot of road works
[08:13] sustrik	still the same
[08:13] mikko	seems like the whole city was being rebuilt
[08:13] sustrik	that's how cities look in balkans
[08:14] sustrik	like they are rebuilt all the time
[08:14] sustrik	bratislava is a bit on edge of the area
[08:14] sustrik	but the impression is the same
[08:15] sustrik	btw, how does the php api look like?
[08:15] sustrik	i think i've noticed you are using 'zeromq' prefix
[08:15] mikko	i tagged 0.1.0 beta
[08:15] sustrik	yep, seen that
[08:16] sustrik	some time ago we've tried to write a guidelines for APIs so that all the language bindings look more or less the same
[08:16] mikko	yep, class name in php can't start with a number
[08:16] sustrik	have tou seen that?
[08:16] mikko	no, missed that
[08:16] sustrik	wait a sec
[08:16] mikko	it's an easy change if needed
[08:16] sustrik	http://www.zeromq.org/guidelines:bindings
[08:17] sustrik	it's entirely up to you
[08:17] sustrik	but following the guidelines is meant to ease the transition between different languages
[08:17] sustrik	as well as to make communication between developers in different languages easier
[08:18] mikko	Preferred name of the object/namespace/prefix is "zmq"
[08:18] sustrik	yes
[08:18] mikko	let me change that
[08:26] CIA-5	zeromq2: 03Martin Sustrik 07master * r00cf3ce 10/ (include/zmq.h src/socket_base.cpp src/socket_base.hpp): multi-part message functionality available via ZMQ_SNDMORE and ZMQ_RCVMORE - http://bit.ly/aKIAhA
[08:40] mikko	i renamed the extension from 'zeromq' to 'zmq' as well
[08:40] mikko	Search -> Replace All
[08:40] mikko	:)
[08:40] sustrik	mikko: great, thanks
[08:51] mikko	repo renamed, links updated, docs update
[08:52] mikko	i think it's almost done
[08:58] sustrik	mikko: you've already done the poll part, right?
[08:58] mikko	yep
[08:58] sustrik	as you've seen in the guidelines there's no standard for polling API at the moment
[08:58] sustrik	on the mailing list there's a thread called "polling API"
[08:59] sustrik	i would be gread if you can comment on what the API looks like in PHP
[08:59] sustrik	the ultimate goal is to get some common understanding how polling interface should look like to be acceptable for all languages
[09:00] mikko	i modelled the polling interface after existing "select" api in php
[09:01] mikko	so it would be somewhat familiar for the developer already
[09:01] mikko	i'll take a look
[09:01] sustrik	thx
[09:02] mikko	hm, i actually hid away the context from the developer
[09:03] mikko	due to the complexicity of persistent resources in php
[09:03] mikko	inside one process i create one context per combination of options which is persisted based on configuration setting which can be set by administrator
[09:03] mikko	that way in shared hosting scenario the admins can disable persistence altogether
[09:12] mikko	is there any reason why you didn't use existing event loop libraries for polling?
[09:12] mikko	such as libev
[09:15] sustrik	mikko: how would you do that?
[09:15] sustrik	0mq socket is not a file descriptor
[09:16] sustrik	as for the context, it's intent is to separate different applications running in the same process
[09:16] sustrik	if PHP doesn't allow that kind of thing, there's no point in having context object exposed
[09:17] mikko	i don't think for 99.9% of phps usecases context object is not needed
[09:17] mikko	as sharing context inside a php script should be advantage rather than harmful
[09:18] mikko	if there are other extensions that use zeromq they would have their own context
[09:18] sustrik	it was originally designed for C
[09:18] mikko	usually i tend to try to design with simplicity in mind for php :)
[09:18] sustrik	the use case was that if there are 2 libraries, each using 0MQ
[09:18] mikko	yeah, in C it makes a lot of sense
[09:18] sustrik	you can link them to the same process
[09:18] sustrik	without them misiteracting
[09:19] sustrik	exactly, if a misinteraction probelm cannot happen in PHP there's no point in having context
[09:20] mikko	hmm, how does the polling work in 0mq at the moment?
[09:20] mikko	i would've thought that it checks for io at some point
[09:20] sustrik	I/O is done in I/O threads
[09:21] sustrik	what you are checking in zmq_poll is whether there are messages queued by I/O threads
[09:21] sustrik	destined for application threads
[09:21] sustrik	it's done via a lock-free polling object
[09:21] sustrik	ypollset.hpp
[09:21] sustrik	iirc
[09:22] sustrik	but the real point is that there's no way to simulate an fd in user space
[09:22] sustrik	so you can either go kernel
[09:22] sustrik	thus loosing portability
[09:23] sustrik	or you sacrifice fd's
[09:24] sustrik	the latter means that system polling mechanism - such as select of poll - cannot be used with your implementation
[09:24] sustrik	:(
[09:24] mikko	now i get it
[09:24] mikko	i've only used event loops in very simple scenarios
[09:24] mikko	such as websocket daemon etc
[09:24] mikko	where there is a writable/readable fd
[09:24] sustrik	ack
[09:25] sustrik	when /me is rich and powerful, i'll loby for adding user-space fds to POSIX
[09:27] mikko	http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod#code_ev_prepare_code_and_code_ev_che
[09:27] mikko	"Their main purpose is to integrate other event mechanisms into libev and their use is somewhat advanced. They could be used, for example, to track variable changes, implement your own watchers, integrate net-snmp or a coroutine library and lots more."
[09:28] mikko	that might be fd based as well
[09:29] sustrik	not sure
[09:30] sustrik	maybe its just a way to to say "use zmq_poll instead of poll"
[09:30] sustrik	and 0mq sockets instead of fds
[09:31] sustrik	hm, looking at the example, it's probably fd-based
[09:32] mikko	"The drawback with this solution is that the main loop is now no longer controllable by EV. The Glib::EV module uses this approach, effectively embedding EV as a client into the horrible libglib event loop."
[09:33] mikko	the author does not seem to be the most polite person
[09:34] sustrik	:)
[09:34] sustrik	but that seems to imply you can actually merge it with 0mq
[09:35] sustrik	i am not sure i understand it correctly
[09:35] sustrik	but does it mean you can delegate the actuall polling to zmq_poll
[09:35] sustrik	?
[09:35] mikko	i think event_poll_func is a callback for Glib::EV
[09:35] mikko	yes, seems like that
[09:36] sustrik	ok, if someone appears with a need for libev integration, let's point him this way
[11:21] mikko	sustrik: is the maximum size for ZMQ_IDENTITY exposed somehow?
[11:24] sustrik	it's 255
[11:24] sustrik	you mean like a constant or so?
[11:25] mikko	i mean is that subject to change?
[11:25] sustrik	no, it's not
[11:25] sustrik	255 characters should be enough
[11:33] sustrik	mikko: do you expect longer identites?
[11:33] mikko	no, just that if i hardcore the value and it changes in the future
[11:33] sustrik	strictly speaking, there's no real technological reason to limit the size of identity
[11:34] sustrik	in theory it can be made unbound
[11:35] sustrik	having identities longer than 255 characters is not a good idea
[11:35] sustrik	but why limit those that want to do this kind of ugly stuff?
[11:36] mikko	retrieving is easier if its limited
[11:36] sustrik	true
[11:36] sustrik	a different question: should the identity even be retrievable?
[11:36] sustrik	what is it good for?
[11:37] mikko	well, in case of php you might instantiate the object and then use it a few requests later
[11:37] mikko	effectively the socket is persisted and then tied to a new php object later
[11:38] sustrik	right, but why would you want to retrieve the identity?
[11:38] mikko	you mean instead of resetting it each time?
[11:39] sustrik	the question is what's the scenario where you need getsockopt (ZMQ_IDENTITY)?
[11:40] mikko	let's a say you got a following php script: <?php $object = new ZMQ(ZMQ::SOCKET_REP, 'mysock1'); ?>
[11:40] mikko	the socket structure is stored internally over multiple executions of this script
[11:41] mikko	to get persistent connections
[11:41] mikko	now, let's say you have an identity
[11:41] mikko	the script becomes <?php $object = new ZMQ(ZMQ::SOCKET_REP, 'mysock1'); $object->setSockOpt(ZMQ::SOCKOPT_IDENTITY, 'my best socket'); ?>
[11:42] mikko	so on each execution of this script (page load) you would reset the identity
[11:42] mikko	i dont know if that makes any difference
[11:43] mikko	if you can retrieve it you could do <?php $object = new ZMQ(ZMQ::SOCKET_REP, 'mysock1'); if ($object->getSockOpt(ZMQ::SOCKOPT_IDENTITY) != 'my best socket') { $object->setSockOpt(ZMQ::SOCKOPT_IDENTITY, 'my best socket'); }
[11:43] sustrik	the identity is meant to stay the same
[11:43] sustrik	so you open socket
[11:43] sustrik	set the identity
[11:43] sustrik	and that's it
[11:43] sustrik	you don't need to touch it anymore
[11:44] mikko	indeed, but in phps case if you can't retrieve the identity you can not know if it's set already
[11:44] mikko	as the socket is persisted over multiple requests but the PHP object is recreated each time
[11:45] sustrik	true, but still, setting it each time doesn't really matter
[11:45] sustrik	except as performance problem
[11:45] sustrik	but i would suspect there are much more perf problems in PHP itself and setting a single string shouldn't really matter
[11:46] mikko	well, in C you might have a large array sockets and each one have identity set at different stages
[11:47] sustrik	yes, that's the only use case i can think of
[11:49] mikko	i would think it makes no harm to limit/allow retrieving
[11:49] mikko	if someone needs longer identities let's cross that bridge when we get there
[11:49] sustrik	ok, good
[11:49] sustrik	let it stay as it is now
[12:00] CIA-5	zeromq2: 03Martin Sustrik 07master * r6cf0765 10/ include/zmq.h : C-style comments in zmq.h - http://bit.ly/dw16x4
[12:09] mikko	mmm
[12:09] mikko	protein bars taste horrible
[12:14] mikko	hmm
[12:14] mikko	one thing i was wondering that might be useful
[12:14] mikko	have the libzmq version exposed as constant as well
[12:14] mikko	that would show which version people compiled against vs runtime version
[12:18] sustrik	yeah, it would be useful, but the version numbers are generated from configure.in
[12:18] sustrik	try asking about it on the mailing list
[12:18] sustrik	people who understand how the build system works may have an opinion
[12:20] CIA-5	zeromq2: 03Martin Sustrik 07master * rdff79d7 10/ builds/msvc/platform.hpp : version number bumped to 2.0.7 for MSVC build - http://bit.ly/b2cKvx
[13:45] sustrik	people who understand how the build system works may have an opinion
[14:36] CIA-5	zeromq2: 03Martin Sustrik 07master * rfba28c7 10/ (6 files in 2 dirs): issue 1 - Change zmq_term semantics - http://bit.ly/c7n0NQ
[14:55] mikko	reply-to "ØMQ development list" <zeromq-dev@lists.zeromq.org>
[14:55] mikko	odd
[14:59] sustrik	it've been discussed on the mailing list
[14:59] sustrik	i have no opinion on it
[14:59] sustrik	feel free to complain
[14:59] mikko	sorry?
[14:59] sustrik	the reply-to thing
[14:59] mikko	ah
[14:59] mikko	:)
[14:59] mikko	im just wondering why it goes like that
[15:00] mikko	noticed your post of zmq_term semantics and trying to formulate my idea into an email
[15:00] sustrik	because people are used to hit reply button instead of reply to all
[15:00] sustrik	thus the conversations tend to disappear from public view
[15:06] sustrik	ah, you meant the strange character?
[15:06] sustrik	"Ø
[15:06] sustrik	I've just changes that to ordinary 0
[15:06] sustrik	changed*
[15:08] mikko	yeah, i mean the strange character
[15:08] mikko	meant*
[15:08] sustrik	fixed!
[15:17] mikko	my first mail to list sent
[15:18] sustrik	:)
[15:18] sustrik	let me reply
[16:01] mikko	im gonna hit the gym
[16:01] mikko	bbl
[18:47] mikko	back