Saturday February 19, 2011

[Time] NameMessage
[00:20] Kinbote would it be possible to use zermomq as a kind of "router" for udp traffic, specfically SIP+RTP?
[01:04] stodge Anyone know if there are ubuntu packages for openpgm/zeromq/pyzmq?
[01:05] stodge Oops - accidentally left the channel. Did my question make it?
[01:06] stodge Are there Ubuntu packages for openpgm/zeromq/pyzmq?
[01:31] stodge How do I build pyzmq with openpgm support? I built openpgm, built zeromq --with-pgm and then built pyzmq. But zmq.EPGM isn't available
[16:09] Steve-o sustrik: have a new alternative to types.h, following MSVC's CRT definitions
[17:36] sustrik Steve-o: would that fix the 0mq build?
[17:37] sustrik i've undefined bool
[17:37] sustrik which fixed sunstudio build
[17:37] sustrik however, the msvc build is still broken
[17:37] sustrik no idea of what's going to
[17:37] sustrik on*
[17:37] Steve-o msvc has never complained here
[17:38] Steve-o haven't tested 0mq & sunpro for a long while
[17:38] Steve-o got an output log?
[17:44] sustrik wait a sec
[17:45] sustrik Steve-o:
[17:45] sustrik that's 0mq master
[17:45] sustrik it looks like there's still problem with bool
[17:45] sustrik note that the "missing" function has bool argument
[17:46] Steve-o I see
[17:46] sustrik it's pgm_sender and pgm_receiver
[17:46] sustrik which include pgm_socket
[17:46] sustrik which in turn include pgm.h
[17:47] Steve-o so what changed, how come 2.1.0 builds fine?
[17:48] sustrik so what happened is that i've added a new argument to one internal 0mq function
[17:48] sustrik it was a bool argument
[17:49] Steve-o and conflicts trickled through, ok
[17:49] sustrik right
[17:49] sustrik then i've added something like this:
[17:49] sustrik #include <pgm.h>
[17:49] sustrik #ifdef bool
[17:49] sustrik #undef bool
[17:49] sustrik #endif
[17:49] Steve-o as per your mail
[17:49] sustrik yes
[17:50] sustrik that fixed the sunstudio build
[17:50] sustrik but caused different errors in msvc build
[17:50] sustrik the ones you see now
[17:51] Steve-o so the header and the code are using different definitions of bool still
[17:52] sustrik yes, looks like that
[17:52] guido_g howdy
[17:52] sustrik no idea what exactly is wrong
[17:52] sustrik hi
[17:52] Steve-o well pgm.h is only included twice
[17:52] sustrik yes
[17:52] sustrik i've undefed bool at both places
[17:52] guido_g any ideas why (e)pgm stopped working on the master?
[17:53] sustrik no, but it looks like 0mq problem
[17:53] guido_g ahh ok
[17:53] sustrik presumably one of the recent patches have broken it
[17:53] guido_g just asking because there was no feedback to the mail
[17:54] sustrik are you experiencing the problem?
[17:54] sustrik my problem is i have no test env
[17:54] sustrik so it's hard for me to find out what's the problem
[17:54] guido_g yes, i experienced the problem and reported it
[17:54] sustrik i may try pgm over loopback though...
[17:54] guido_g huh?
[17:55] sustrik i have just a single box :)
[17:55] guido_g lo doesn't do multicast
[17:55] guido_g too bad
[17:55] Steve-o my blade chassis is switched off and 50,000 miles away, I only have a windows laptop currently
[17:55] Steve-o :D
[17:55] sustrik still in NY?
[17:56] Steve-o back in HK next week
[17:56] Steve-o just working through a lot of contractor jobs
[17:56] sustrik right
[17:56] sustrik guido_g: there's simulated multicast over lo
[17:56] sustrik i may try that
[17:57] guido_g you simply use the external ip, mc isn't routed anyway
[17:57] guido_g btw, this was the setup where i first discovered that something went wrong
[17:58] guido_g normaly mc traffic is also routed back to the originating host
[17:58] sustrik if i do that i won't see whether the message goes through or not
[17:58] guido_g huh?
[17:58] guido_g sorry, can't follow you
[17:59] sustrik when i use a multicast group that nobody listens to
[17:59] sustrik "you simply use the external ip, mc isn't routed anyway"
[18:00] guido_g ment to mean that the mc traffic is not gouing to leave your local network when using the external ip of you box
[18:00] guido_g and you don't have to play tricks w/ the network stack to make lo mc aware
[18:01] guido_g or did you mean multicast loopback?
[18:01] sustrik ok
[18:01] sustrik yep
[18:01] guido_g ahhh...
[18:01] guido_g sorry
[18:01] guido_g my fault then
[18:01] sustrik never mind
[18:01] guido_g i assumed loopback device
[18:02] sustrik i'll try that am i'll see whether i can discover the bug that way
[18:02] guido_g ok
[18:02] guido_g thanks
[18:02] guido_g if you think you have it, i can try it w/ two machines
[18:13] Steve-o sustrik: ugggggh, so I really have to use a pgm_bool_t? Is MSVC picking up BOOL from inside pgm.h and breaking some C++ templates expecting bool?
[18:23] Steve-o the xtree template warning with bool/BOOL conflict no longer occurs with OpenPGM trunk
[18:24] Steve-o i.e. the new types.h:
[18:33] pieterh sustrik, do we have a place in the git for tools like web api generation?
[18:33] sustrik Steve-o: well, i'm not sure what's exactly going on
[18:33] sustrik pieterh_: hm, not yet
[18:33] pieterh I guess this applies also to mikko's tools
[18:34] sustrik the build system is scattered across the directiories
[18:34] sustrik which is how autotools work
[18:34] sustrik what about putting the scripts to doc subdir?
[18:34] pieterh hmm, ok what I'll do is make a separate tool that does this work
[18:34] pieterh in its own git
[18:34] pieterh since we have other projects with the same needs, e.g. zfl
[18:35] sustrik as you wish
[18:35] sustrik however, there are documentation generation scripts in doc subdir
[18:35] sustrik so placing one more there won't be completely ridiculous
[18:36] sustrik Steve-o: i though that undefining bool should solve the issue...
[18:36] pieterh sustrik, what's there now is the straight man->txt/xml stuff
[18:36] Steve-o not defining in the first place is better :D
[18:36] pieterh let me think this over for a couple of hours
[18:37] sustrik pieterh_: txt->man/http
[18:37] sustrik html*
[18:37] sustrik Steve-o: yes
[18:37] pieterh yes, txt to man/html/xml
[18:37] pieterh I'm using the xml as input for this process
[18:37] sustrik but as far as I understand you use your "bool" in PGM API, right?
[18:38] sustrik pieterh: what xml?
[18:38] pieterh try 'make zmq.xml' in the doc directory
[18:38] sustrik it's some intermediate asciidoc step?
[18:38] sustrik a-ha
[18:38] pieterh sure
[18:38] sustrik i see
[18:38] pieterh docbook xml format
[18:39] sustrik Steve-o: maybe the bool in the API should be simply changed to int
[18:40] sustrik pieterh_: i would say generating the wikidot source could be part of the build process
[18:40] sustrik it introduces dependency on perl though
[18:40] pieterh sustrik, indeed
[18:41] Steve-o try the update first, although it might indeed be that BOOL is too strongly typed
[18:41] pieterh let's aim for stand-alone tool which mikko can then integrate into his build processes
[18:41] sustrik pieterh_: ok
[18:41] sustrik Steve-o: ok
[18:41] pieterh :-)
[18:42] pieterh if there are other things we can profitably mechanically upload to the wiki, let me know
[18:43] sustrik ultimately, the docs for stable version
[18:43] sustrik on the other hand, raw html is much faster than wikidot
[18:43] sustrik it's a trade-off, speed vs. nice design
[18:47] pieterh up to you, really
[18:47] pieterh it's faster with pure HTML but we can do more interesting things with Wikidot, e.g. Table of Contents, navigation, etc.
[18:48] pieterh our limiting factor is URLs, really
[18:48] pieterh would work
[18:48] pieterh for each stable version released
[18:49] pieterh there are still people using 2.0.8 in production...
[18:49] pieterh sorry, my bad, wikidot doesn't allow subdirectories...
[18:50] pieterh
[18:52] sustrik no preference
[18:52] sustrik should we ask on the ml?
[19:08] pieterh sustrik: ack
[21:28] mikko g'evening
[22:03] pieterh mikko, top of the evening to you!
[22:43] sustrik pieterh_: still there?
[22:57] Steve-o wow, it's surprisingly tedious to detect single core machines with hyper-threading
[22:58] mikko Steve-o: runtime?
[22:58] Steve-o yup
[22:59] Steve-o Windows XP SP3 has a backported API to walk the CPU topology to count packages, cores, logical processors
[22:59] Steve-o CPUID seems to be a bit broken on new i3, i5, i7 processors
[22:59] Steve-o every Unix has it's own custom API or /proc system
[23:00] Steve-o I'm trying to see if I can at runtime catch the case of 1 core with HT enabled to stop spinlocks spinning
[23:01] Steve-o it's probably futile as on Linux at least the system clock is pretty much only low resolution in such configurations as HT causes it to jump all over the place
[23:02] mikko i got a vague memory about reading about this or something related
[23:03] Steve-o I have something working in Windows but it's probably not worth committing
[23:04] Steve-o I already use hyper-threading CPU pauses in spinlocks so that multi-core HT systems run fine
[23:06] mikko btw, i don't know whether you saw this:
[23:06] mikko
[23:07] mikko seems to be functional now
[23:10] Steve-o ok great
[23:10] mikko intel seems to have articles on detecting HT
[23:10] mikko but when i try to follow the link to the article it 404s
[23:11] Steve-o just have the recent fixes from sustrik for C++ and should be good to release soon
[23:11] Steve-o detecting HT is easy with CPUID, counting the physical cores is a bit difficult
[23:11] Steve-o or rather the number of logical processors CPUID returns no longer appears valid
[23:12] Steve-o there's a question on stackoverflow with i7 cores
[23:12] Steve-o anyway, must dash, ciao
[23:13] mikko see you