IRC Logs for #crux-devel Wednesday, 2019-05-01

*** stenur has quit IRC00:12
*** xor29ah has quit IRC00:44
*** xor29ah has joined #crux-devel00:48
*** dlcusa has joined #crux-devel05:44
TimB_pedja: iirc, you had a suggestion about lapack? by any chance about using openblas?09:59
pedjaTimB_, I've bben using openblas for a while now. numpy seems to be happy with it :)10:00
TimB_I just packed that up and changed my lapack port to depend on it10:00
TimB_and I forgot to open a bug about consolekit2 yesterday ~.~ I'll do it now10:03
pedjait's no rush :)10:03
TimB_ah well, best to put it out there10:03
pedjathere are several lapack implementations, perfect use case for alias use :)10:04
TimB_jue is using consolekit2 as well according to his repo :)10:05
TimB_http://jue.li/crux/ports/consolekit/Pkgfile10:05
pedjaoh, nice. so it's been battle tested by several people :)10:05
TimB_about lapack: there is just romster that has it in his repo, so.. since I have it in my lo repo as well (which I could also merge to contrib?), maybe he wants to drop the port if I'll maintain one with openblas in contrib?10:06
pedjaor just push openblas as openblas to contrib, with the README explaining that it's optimised lapack? tbh, idk10:08
TimB_well, openblas will conflict with the blas libs lapack provides10:09
TimB_so I needed to trim those from the lapack package10:09
pedjaah, yes. good point10:09
pedjawell, ping him here or at flyspray and ask, I guess10:10
TimB_pedja: https://crux.nu/bugs/index.php?do=details&task_id=173210:10
pedjaafaik, openblas is the most commonly used lapack, so it would make sense to use that10:11
pedjaTimB_, assign it to 'crux developers'10:11
TimB_done10:12
pedjanice job :)10:12
TimB_openblas seems to do the job for me10:21
pedjaif you are interested, I can paste my port, for reference10:31
TimB_sure10:31
TimB_I also have the feeling you should do the same with pulseaudio10:32
pedjaOK. as soon as the build finishes and numpy.test() approves of the update :)10:33
pedjaTimB_, pulseaudio?10:33
pedjamy pa port is a bit different from the one from opt10:34
TimB_well, that's why :P10:34
pedjaah :)10:34
TimB_I am still not all too happy with pa10:34
TimB_pulseeffects is doing it's part, maybe I need to up niceness.. not sure X_X10:35
TimB_if pa and effects are set to 100% I get cracking audio when just lowering the volume on a yt video or whatever - pretty sucky10:35
TimB_if I lower the pa output to 60% it works good10:36
pedjapa port http://dpaste.com/3XVVKXK10:36
TimB_but I have to redo that after every god damn reboot10:36
TimB_is that patch in the offical port?10:36
TimB_lol, the last sed line, really? great work pa :D10:37
pedjayeah, the upstream has some weird defaults :)10:38
pedjaI am not sure if that patch is needed anymore, I picked it up from opensuse iirc10:39
TimB_I'd test it if it makes a difference10:39
pedjait's for echo-cancelation module10:40
TimB_well then, maybe it's not that related10:41
TimB_:D10:41
pedjayeah :)10:41
pedjaTimB_, openblas http://dpaste.com/3BJR65S10:53
TimB_pedja: thx10:55
pedjanp10:55
TimB_looks prettier than my first draft, that's for sure :P10:56
pedjaheh. it had a few iterations until I got it right10:57
TimB_mhmh, your port provides more links10:59
TimB_I had just blas in mind10:59
pedjaopenblas is full drop-in replacement for reference lapack10:59
TimB_I see10:59
pedjaat least according to people way smarter then me :)10:59
TimB_so I could drop lapack completly?11:00
pedjaapplication depending on it would require testing, but yes11:00
pedjaso far I had no issues11:01
TimB_interesting, I'm gonna install your port11:01
pedjatakes a while to build, thou11:02
TimB_perfect, just how I like it x_x11:03
pedjaI did try a specific cpu build but weird bugs occurred11:03
TimB_ok11:03
pedjaapparently, some intel's penryn arch cpu's have sse_4 and some don't, but there is no way to tell just by the name :)11:04
pedjathat one was fun to track down11:05
TimB_sounds like it :D11:05
pedjaso I just build ALL THE THINGS and let the applications dynamically choose11:05
TimB_sounds like a safe bet for contrib :)11:06
pedjapeople are free to try to 'optimise' for their cpu only, but, in my experience, there lies madness :)11:08
TimB_I just see that brianlfs has openblas in his repo as well11:08
pedjaiirc, we talked about it few months back. it is a req for some 3d slicing software, I think11:10
TimB_yeah, when I was working on the libreoffice repo :)11:12
TimB_openblas just finished, let's see11:13
pedjaalready? ah, yes, not everyone has a potatoputer like me :)11:15
TimB_he :)11:16
TimB_I mean, this thing 5 years old as well.. or 611:18
pedjatry 10 :)11:19
TimB_you win :P11:19
pedjaheh11:20
TimB_R seems to work, lo calc as well11:20
pedjanice11:20
TimB_pedja: would you mind if I adopt your pkgfile for openblas?12:01
pedjanope :)12:02
TimB_I'll try to build libreoffice with it, but if that's fine I think I'll take that12:02
TimB_great12:02
TimB_ccache doesn't seem to pick up a difference as soon as coin-or-coinutils is rebuild against openblas, nice12:12
*** stenur has joined #crux-devel12:35
juestenur: it's not necessary to change ssd, in cases like dnsmasq we have to specify another match rule12:36
juestenur: see here -> https://84752944ab800a1b.paste.se/12:36
juepushed the fix to 3.512:40
stenurjue: sorry, posted on #crux!  Me and irc, my first steps.12:45
stenurjue: i see; --name i did not see at all, but still needs adjustments everywhere.12:46
pedjacan we just switch to systemd already instead of hacking s-s-d :)?12:46
pedjaACTION hides12:46
jueyeah, adjustments for some ports are necessary, did it already for mariadb, stunnel etc.12:47
stenurjue: and isn't it broken?  Since the check is about a user id, it should need a user id!12:47
jueno, works as expected for me12:47
jueyou can test it with dnsmasq now12:48
stenurpedja: oh, sad world!12:48
TimB_release the hounds D:12:48
juestenur: for most daemons it works without modification and for the rest the fix is easy, so don't worry ;)12:52
stenurjue: did not see it before, i use so few things.12:53
pedjaas an experiment, I made a systemd port a while back. it's huge, because *everything* is enabled by default, which is...silly12:54
stenurthe only thing i know that i really agree with is the state notification via socket.12:56
stenurbecause of all the priviledge separation and daemon that fork and fork ..12:56
stenurfinal process notifies manager, e.g. s-s-d.12:57
stenurbut every daemon needs to be patched for that.12:58
stenurand OpenSSH for example (yet) refused to integrate that patch upstream.12:58
stenurYeah, DBUS can or could do many good things...  Somehow the stuff needs to be done.12:59
stenurHere on my box only Firefox uses it, for whatever.  I mean i now start the thing via rc for the first time, for looking what happens...13:00
stenurActually nothing says the monitor :)13:00
stenurOther than that i would be fine with the opposite scheme, like runit uses.13:01
stenurI.e., do not fork at all (or keep "some" parent with connection hanging), to keep it in the table of the manager.  Like that it sees when daemon changes state.13:03
stenurYou have some useless intermediate processes, but true supervising.  At least still today, daemon support that it seems.  Void Linux still offers runit, but i do not really use it.13:05
stenurOther than that i for one dislike anything else of systemd.  I mean yeah, i understand many decisions from a logical point of view, but i do not want to have it.13:06
stenurI really want a normal cron, syslog, ntp client, even if it means more space and some more CPU consumption.13:08
stenurI was happy to come back from ArchLinux to first Alpine, now even more so CRUX!13:09
stenurI finally knew again what happens on the system, more or less!13:09
stenurThat is value by itself.13:10
stenurReally, really much more satisfied now.13:10
jueglad you like it :)13:11
stenurI do!13:12
stenurjue: you are right, not broken; it verifies the process name matches after getting the pid13:18
stenurpedja: yes it is hacking, i never really had a look.  jue wanted to import the entire dpkg repo into core, i just made it compile on CRUX, and that was it.13:22
pedjastenur, in case it wasn't obvious, I am kidding :)13:22
stenurOh. :)13:23
pedjawhen I want to play with systemd, I spin up a opensuse vm13:24
pedjacan see the value in it, but very much dislike the kitchen-sink approach of it13:26
pedjatoo many moving parts for my taste13:27
stenurMy words to Gaetan Bisson of ArchLinux!13:28
stenurYou have to keep up with all the changes and development in order to know what is going on on your box.13:28
stenurOr simply trust and let it roll on.13:29
pedjaafaict, some distributions do a better job beating it into submission then others :)13:30
stenurThinking about the s-s-d thing, maybe that CRUX adjustment should be dropped at a later time.  Adding --name isn't that much of a burden.13:31
pedjais s-s-d still developed upstream?13:32
pedjadebian, right?13:33
stenurpedja: yes.  Well, a few things each year since i look at it.  Mostly cosmetical.13:39
*** xor29ah has quit IRC14:13
*** xor29ah has joined #crux-devel14:13
*** stenur has quit IRC16:06
*** BurnZeZ has joined #crux-devel17:39
*** onodera has joined #crux-devel17:50
*** BurnZeZ has quit IRC18:05
*** BurnZeZ has joined #crux-devel18:21
*** BurnZeZ has quit IRC18:34
*** BurnZeZ has joined #crux-devel19:20
*** onodera has quit IRC21:55
*** stenur has joined #crux-devel22:14
stenurHello. Hu, shadow is a thing.  But PAM i never ever got right.22:20
stenurI found out shadow having been converted to PAM on CRUX just recently.22:22
jaegerYes; CRUX 3.5 uses PAM, so shadow, su, sudo, openssh, and others need to be changed/recompiled to support it22:27
jaegerwell, 'su' is part of shadow but the point stands22:27
stenurAmazing work, i would never have dared that!22:28
stenurI was in the other window to search for Samuel Beckett "Losigkeiten", but Wikipedia is a mess.  It is "Lessness" in the english version. Well.22:29
stenurWell :), stupid me.22:29
stenurThe thing is, logname(1) does not work, and after having a hard hour i came to this commit.22:29
stenurhaha!22:29
stenurI have added pam_loginuid.so thanks to Google, second last in pam.d/login.22:30
jaegerhrmm... perhaps coreutils needs PAM support as well? though it doesn't seem to work in 3.4 either22:30
stenurlogname?22:32
jaegeryes22:33
jaegersays "logname: no login name"22:33
stenurPrograms use the getpw series it seems.22:35
stenurI had been slammed by a "programs do not need DNS for local hostname", they just use gethostname() today (but forgotten where),22:37
stenurthen walked along that line and came over to logname for the login..22:37
stenurand only saw 32-bit max in /proc/self/loginuid which i did not know before,22:38
stenurbut strace gave that for logname.  So, then.22:38
stenurNever special treated gethostname before looking up something via DNS myself.  Just looked.22:42
jaegerI'm not sure I'm parsing your comments correctly. What actually uses logname and is failing?22:42
stenurNothing i know but logname itself.22:44
stenurNothing is failing. No, a mail flew by today, don't know where (new S/MIME 4.0 standards came out today), with a "moral pointing finger", said programs have to use gethostname etc.22:46
stenurSo that led me along looking around.  And i came to logname in the end, and it did not work.22:47
jaegerah22:48
stenurWhich surprised me a lot, so i used strace, came to that loginuid file which i had not seen consciously yet, and it was 32-bit "-122:48
stenurWhich made me even more ??22:48
stenurYeah, fuzziness..22:49
stenurSo that finally made me realize logname belongs to shadow, which has been based upon PAM for 3.5..22:50
stenurAnd that there is a PAM module which does nothing but filling in that ID field so that logname(1) works in children of that session.22:51
stenurSo yay! I now can have a new login(1) session and get a meaningful entry that logname uses.22:52
stenurI use id(1) in my profile.  That always works regardless.22:52
ryuohm. this is a very nice use for variadic macros.23:07
ryuoa viable alternative to basic variadic functions.23:08
jaegerI wonder what's actually supposed to create/own /proc/self/loginuid23:46
jaegerit doesn't exist at all on my systems23:46
ryuojaeger: it's on the proc FS. it's probably managed by the kernel.23:49
ryuoBINGO it is23:51
ryuoit exists only if...23:51
ryuoCONFIG_AUDITSYSCALL is enabled on the kernel.23:51
jaegerAh, ok23:54

Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!