ZeroMq IRC Log

Tuesday October 5, 2010

[Time] Name	Message
[00:12] plowette	hi
[00:12] plowette	anyone have much experience with the Java binding?
[06:29] sustrik	mikko: i am not an expert on build systems
[06:29] sustrik	however, if you rpath patch works
[06:29] sustrik	send it to the mailing list so that mato can apply it
[07:21] mato	mikko: i'm here now
[07:45] mikko	mato: rpath issues with tests
[07:46] mato	mikko: they can't find the built libzmq.so?
[07:46] mikko	mato: look at the build output here: http://valokuva.org:8080/job/ZeroMQ2_master/20/console
[07:46] mikko	yes
[07:46] mikko	https://gist.github.com/c7580288014ab2b9b717
[07:46] mikko	that fixes it
[07:47] mato	hmm
[07:47] mikko	but should the tests link statically ?
[07:47] mato	no
[07:47] mato	but the thing is...
[07:47] mikko	because you want to be sure that you test against version that you've just built
[07:47] mato	before i started maintaining the autoconf stuff, palo malosek added the bit at the end of configure.in
[07:47] mato	which strips rpath
[07:47] mato	now, i'm not sure why that got added
[07:47] mikko	if it loads via LD without rpath you might load from ldpath?
[07:48] mato	i'm not sure
[07:48] mikko	and actually test the already installed version (assuming you run make check before make install)
[07:48] mato	this is one of those issues people seem to be in two minds about
[07:48] mato	personally i'd like to know why we're stripping rpath on linux
[07:48] mikko	distributions i assume
[07:49] mikko	they usually remove rpaths
[07:49] mato	yes, but, ... that should be done by the distro packaging scripts IMO
[07:49] mato	not in our autoconf
[07:49] mato	or, we should provide an option to not hardcode rpath
[07:50] mikko	but that wouldn't help tests
[07:50] mato	yes it would
[07:50] mikko	if you run make check without having 0MQ installed already
[07:50] mikko	where would the link-time rpath point to?
[07:50] mikko	something like /usr/local/lib?
[07:50] mato	no idea, but you shouldn't be doing that
[07:50] mato	besides, make check will build zmq, won't it?
[07:51] mikko	yes
[07:51] mikko	hmm, in my opinion people _should_ run tests before installing
[07:51] mato	then as long as we get rid of / make into an option the rpath stripping behaviour we should be fine, no?
[07:52] mikko	i think that should be fine
[07:52] mikko	although i don't see it helping the tests (without forcing the user to install first and updating ldd cache)
[07:52] mato	?
[07:53] mikko	runtime linkers cache
[07:53] mato	but, but... if the tests are linked with rpath then that shouldn't matter...
[07:53] mato	anyway, let me try it
[07:53] mato	that's the best way to find out
[07:57] mikko	i changed the hudson build to poll SCM every 15 minutes so you should see the results from there as well
[07:57] mikko	i need to run to a client meeting
[07:57] mikko	bbl
[12:13] mato	mikko: when you get back; i tested the rpath change and emailed to the ML
[12:13] mato	mikko: removing the hack we have will make the tests work as I described
[12:47] mikko	mato: i dont have the mail yet (i think)
[12:54] mato	Subject: [zeromq-dev] [PATCH] Don't change libtool RPATH behaviour on Linux
[12:57] mikko	ah
[12:57] mikko	9:15 am
[12:57] mikko	that was ~100 emails abck :)
[12:57] mikko	back*
[12:57] mikko	the lines you are removing looks like a hack
[12:57] mikko	indeed
[12:58] mato	mikko: yeah, i think it was added in way back in 0mq 1.x days, i just posted to the ml to check
[12:58] mato	if no one complains in a day or so i'll just take it out
[12:59] mato	AFAICS it won't break anything (maybe a few distros with old libtool, but they can just patch it themselves)
[13:00] mikko	i had a lot of problems finding documentation about rpath
[13:00] mikko	seems like it's grossly underdocumented feature
[13:01] mikko	it seems that opinion between LD_LIBRARY_PATH vs rpath is also divided
[13:02] mikko	mato: i read about builbot a bit
[13:02] mikko	mato: it seems that it requires one build master per project
[13:10] mato	mikko: that's entirely possible, i've not used it, just saw that the python people were using it
[13:10] mato	mikko: and Python is nicer than Java
[13:10] mato	mikko: but do what works best for you ...
[13:13] mikko	this far adding the 14 builds has been fairly easy
[13:13] mikko	i guess the complications arise from distributed builds
[13:19] sustrik	mato: iirc is had to do with fedora distro rules
[13:19] mato	sustrik: what did?
[13:19] sustrik	rpath
[13:19] sustrik	not 100% sure
[13:19] mato	sustrik: well, tough, it's a hack
[13:19] mato	sustrik: if no one complains i'll put it back in
[13:20] sustrik	ack
[13:20] mato	I suspect it will cause problems for people building with old libtool from Git releases
[13:20] mato	but again, tough
[13:26] mikko	--disable-rpath or something might work for distros?
[13:27] mato	yes, but i won't actually do that until someone really asks for it :-)
[13:27] mato	and the problem is (I believe) that libtool doesn't support that anyway
[15:59] ptrb	it seems that for at least 1KB messages, I can publish to a ZMQ_PUB about twice as fast as I can receive on the ZMQ_SUB... does that seem right?
[16:03] cremes	ptrb: yes, that seems right
[16:03] cremes	publishing is "fire and forget" whereas a subscriber has to do some byte matching on each message
[16:04] ptrb	even if you do a zmq_setsockopt(.., ZMQ_SUBSCRIBE, NULL, 0); ?
[16:06] cremes	ptrb: i don't know... you should check the code and see if it short-circuits that check for a NULL subscription
[16:06] cremes	also, recall that the subscriber has to do a lot more mallocs that your publisher
[16:06] ptrb	ok. just wanted to sanity-check my benchmark, anyhow. thanks!
[16:35] dermoth	I have a question regarding the zdevices json configuration... why not http://www.hyperrealm.com/libconfig/libconfig_manual.html#Configuration-Files ?
[16:50] dermoth	I remember reading somewhere why JSON was chosen, libconfig seems just as good based on what I remember (iirc ne of the point was that other unix configuration formats didn't allow multiple levels like JSON)
[19:13] FSX	Hi, I was wondering if zeromq could be used to make an IRC bot? :)
[19:15] omarkj	You could make it play a part in an irc bot, sure.
[19:18] FSX	I mean for the networking. I currently use Asyncore. I thought I could try something else.
[19:21] Guthur	FSX, I would assume that because the IRC server is probably not using zeromq that you would have to use a more raw solution of special IRC lib at some point
[19:22] Guthur	There usually is a IRC lib available for most language
[19:22] Guthur	languages*
[19:24] FSX	Ok, but I just need something to do async networking. It's just something for fun. ^^
[19:28] ptrb	if your question is "will zmq do async networking" then the answer is "yes" and IRC doesn't enter into it
[19:29] FSX	Ok, thanks.
[20:00] Guthur	is there a channel log
[21:50] novas0x2a	is it possible to query a socket to see how deep the backlog is?
[21:53] cremes	novas0x2a: no
[21:53] dermoth	novas0x2a, I don'T think so... fwiw for a push/pull architure i'm planning i'll be sending message in the system and expect a worker to reply with a udp backet on a specific port, so I can chekc the latencyÂ If latency goes up, then the backlog is.
[22:00] dermoth	novas0x2a, since code speaks better than words, http://zeromq.pastebin.com/3z3VQ1vR
[22:00] dermoth	this is an early draft of the nagios monitoring plugin for it
[22:02] dermoth	oh there's an extra excho in there, that I used to "reply" manually to it as I didn't have to worker running
[22:41] novas0x2a	dang :(
[22:43] novas0x2a	any idea if a patch adding that as a sockopt would be accepted?
[22:44] novas0x2a	or is this an intentional thing?