IRC Logs for #crux-devel Wednesday, 2018-12-12

*** xor29ah has joined #crux-devel01:50
*** xor29ah has quit IRC02:07
*** xor29ah has joined #crux-devel02:13
*** dlcusa has joined #crux-devel04:07
*** dlcusa has quit IRC06:00
*** dlcusa has joined #crux-devel06:22
*** dlcusa has quit IRC08:25
*** dlcusa has joined #crux-devel08:35
jueteK_: FYI, just pushed a fix to llvm, see the commit message for the reason09:21
jueteK_: stumpled about that by testing new mesa3d, here I saw the error messages from llvm-config09:23
jueteK_: with the fixed llvm we no longer need the 'export LDFLAGS...' line in the Pkgfile of mesa09:25
juerelated, to avoid an error from configure I've added bash as the CONFIG_SHELL09:27
pedjaany thoughts on enabling rtti in llvm/clang? I built a shitload of stuff with them, nothing seems to be broken :)09:27
juefrinnst: guess you are running mesa 18.3, any problems?09:28
juepedja: why?09:32
pedjajue, some applications require that some of their dependencies be built with rtti-enabled llvm/clang, otherwise they fail at linking stage09:33
pedjaspecial snowflakes, I know :)09:34
jueok, have you an example, which one?09:34
pedjaappleseed renderer, for instance, requires that osl that it depends on, be built that way09:35
pedjaI know it's a special case/rarely used, so I'll just fork llvm/clang ports. I was just curious09:37
jueI have no objections to add rtti if nothing breaks, sure09:41
pedjajaeger did a full iso rebuild with them and, afaik, in the time he had to test it, didn't come across any issues, and neither did I, fwiv09:44
pedjait's nbd :)09:45
juewell, probably best to add a feature request for teK_ in our FS09:47
juemy commit to llvm from today was an exception, because it fixed a serious bug09:48
pedjaah, the errors that were shown? yeah, that was weird09:49
juenot only the errors, the main problem was that you cannot link against llvm libs without tuned LDFLAGS09:55
jueif llvm-config is used by the calling program09:55
frinnstjue: just installed it yesterday. havent had time to test10:17
frinnstbut just got a new rx590 gpu so that will be a good test i think :-)10:18
pedjanice10:18
pedjadid they finally fixed it in Linux?10:18
frinnstanybody know what else uses nss/nspr? just pidgin and chromium=10:19
frinnst?10:19
pedjapolkit, I think10:20
pedjafrinnst, give a heads up if you plan to remove them so I can move the ports to my repo :)10:21
frinnstyeah no worries yet. just checking around :-)10:22
pedjaheh10:22
pedjabtw, which rx590 did you get?10:22
frinnstthey are very low effort to maintain so no worries yet10:22
frinnstxfx fatboy or something10:23
pedjaah, the famous one10:23
pedjaGN was not impressed with the build quality, iirc10:23
frinnstlooks solid enough. the fake carbon fiber is lame tho10:24
pedjahaven't seen much of rx590's. I think saphire also has one, and that's it10:25
frinnstchose this one because it has 3 dp ports. also it was in stock :)10:26
pedjaafter those leaks, everyone is waiting for Navi :)10:26
*** xor29ah has quit IRC10:26
frinnstthere's always an upcoming "big thing"10:27
pedjayeah10:27
*** xor29ah has joined #crux-devel10:30
juefrinnst: firefox-bin10:34
pedjaone of these days I'll have to update gcc and gcc-fortran. loads of fun, those two10:34
*** xor29ah has quit IRC10:35
juedbus-glib as well10:35
juethat's what I see with finddeps10:35
*** xor29ah has joined #crux-devel10:38
juefrinnst: though, not checked if the others deps are provided by gtk310:38
frinnstfirefox-bin works without it. I dont have those packages installed at home but run less software there than at work10:39
juehmm, you don't see them if you run 'finddeps firefox-bin'10:40
jue?10:40
frinnstI do, yeah10:41
pedjaeasy enough to test. remove them, see if it breaks10:42
frinnstit doesnt, im pretty sure10:42
frinnstthey are shipped in /usr/lib/firefox also10:43
frinnstanyways, i'll give it a good testing10:43
juebut what is linked against nss?10:43
pedjaso the system libs take precedence?10:44
juefrinnst: you are right, libnss etc. are in the firefox directory and used by firefox10:58
jueat all it's a problem of finddeps10:59
juebbl11:03
Romsterhttps://crux.ster.zone/packages/3.4/firefox%2364.0-1.pkg.tar.xz compiled my own that wasn't that hard11:49
juefrinnst: dbus/dbus-glib is a real dep of firefox-bin11:49
jueI'm happy with firefox-bin, was able to remove a couple of progs installed only for firefox :-)11:57
ryuofrinnst: kmod supported gzip/xz compressed modules. what would it take to get that turned on for future builds of the package?16:44
ryuosupports*16:44
frinnstwhat kernel modules do we provide in the repos? nvidia?16:45
frinnstup to the maintainer i guess16:45
frinnstor am I missing something?16:45
ryuoit's a core port.16:45
ryuoso, who updates those?16:46
ryuothe dependencies kmod needs are already in core.16:46
frinnstoh for kmod itself?16:46
ryuoyes. that's all you have to change.16:46
frinnsta patch I guess ;-)16:46
ryuook.16:46
ryuoi'm asking because it would be great to have that enabled for my future kernel packages.16:47
frinnstah, requires it to be linked? well that could be somewhat problematic if abi were to break or some stuff like that16:47
ryuoi could add a phase to compress the kernel modules so the kernel package takes up less space on disk for modules people won't need.16:47
ryuofrinnst: no, afaik, the kernel doesn't have support for compressed modules.16:48
ryuohow it works is kmod decompresses it in userspace and feeds the result to the kernel like it would an uncompressed module.16:48
ryuoall this effects is packaging.16:48
frinnstyes but if zlib were to break then kmod would not run, probably16:48
frinnstjaeger, jue?16:49
ryuohow likely is that? zlib is pretty stable ABI wise.16:49
ryuoi've never seen an ABI breakage with zlib.16:49
frinnstsure, but user fuckup or some such16:49
frinnstprt-get remove zlib16:49
frinnst"i wanted to save disk space"16:49
ryuoif they do that, their system is likely hosed anyway.16:50
frinnsttrue, bad example16:50
frinnstanyways, afk for a while16:50
ryuook.16:50
ryuoit's not a requirement, but it'd be nice to turn it on for kernel packaging.16:50
ryuoso it could become the default in future crux releases.16:50
pedjaCONFIG_MODULE_COMPRESS?16:50
ryuopedja: the kernel has that as a setting?16:51
pedjayes16:51
pedjagzip or xz16:51
ryuoi thought that only compressed the kernel blob itself.16:51
pedjainitrd, kernel and modules can be compressed16:51
ryuoindeed, but you can't use compressed kernel modules unless kmod is compiled with support for it.16:52
pedjaI like the 'kmod MAY support gzip and xz' line16:52
ryuowhere did someone say MAY?16:52
pedjakernel kconfig for that option16:53
ryuoHm. I see now.16:53
ryuoThis doesn't modify the kernel itself. It just effects module installation. Something you can do with the Pkgfile instead.16:54
ryuothe actual support for this is in kmod.16:54
ryuocomment from their Makefile16:55
ryuo# CONFIG_MODULE_COMPRESS, if defined, will cause module to be compressed16:55
ryuo# after they are installed in agreement with CONFIG_MODULE_COMPRESS_GZIP16:55
ryuo# or CONFIG_MODULE_COMPRESS_XZ.16:55
pedjaimho, dracut or any other initrd is more elegant solution16:55
ryuoNot really, pedja. initrds don't store all available modules.16:55
ryuothis effects all kernel modules on disk.16:55
pedjathey can, if you want, iirc dracut docs16:56
ryuoit can cut the footprint by hundreds of megabytes.16:56
ryuodo you know who wrote the original dracut support for compressed modules?16:56
pedjanope16:56
pedjanever used it16:56
ryuoMe, because I added support for this to frugalware.16:56
ryuoAnd submitted the patches to upstream.16:56
juewe had a request like that already -> https://crux.nu/bugs/index.php?do=details&task_id=82616:57
ryuoMade dracut be able to detect their presence.16:57
ryuoI see.16:57
ryuoAlready was shot down.16:58
ryuoafaik Frugalware never had any issues with the feature. It helped save some space with our kernels.16:58
ryuothough i did find it weird how Debian never enabled support for it.17:00
pedjaone of the coolest thing about CRUX is that it has no 'official' kernel, apart from the one shipped on a iso17:00
pedjableeding edge, lts, patched to 11. run what you like :)17:01
ryuoi consider that a bit of a weakness, as it complicates new deployments if you have to fiddle with the kernel configuration.17:01
pedjakitchen sink config to cover all the things can also be considered a weakness17:02
ryuoonly if space is that much of a premium.17:02
ryuoi've only used a handful of Linux systems where that is literally true.17:02
pedjait's a matter of taste for some people17:02
ryuoembedded routers namely.17:02
ryuoi had also considered switching Frugalware's man page compression to xz, but i scrapped it.17:04
ryuothe average manpage is too small for a different algorithm to be of much help.17:04
ryuoi tested it and it was almost the same as the gzip compression.17:05
pedjazstd is pretty impressive17:05
pedjafast af, too, even on my potatoputer17:06
ryuopedja: you got a vintage spudnic?17:07
pedjaah, 'spud'=potato, right?17:08
pedja'informal potato'.so, potato in jeans and t-shirt, basically17:09
ryuopedja: yea. mr. potato head went into space aboard spudnic. =p17:34
*** onodera has joined #crux-devel18:20
jaegerI don't have a preference regarding module compression support19:13
ryuoi don't see it being accepted unless its depends are moved to /lib, due to the issue raised by jue.19:33
ryuoHm.19:33
ryuoSeems zlib already installs its main library to /lib19:36
ryuoyet... a symlink for it is in /usr/lib.19:36
ryuoACTION boggles.19:37
ryuohttps://crux.nu/ports/crux-3.4/core/zlib/.footprint19:37
ryuoHm.19:37
ryuo... we could enable only zlib and call it a day. it appears it should be safe from the issue raised by jue.19:37
ryuo~132MB uncompressed kernel modules for ubuntu kernel...19:43
ryuolet's see how much gzip takes off19:43
ryuoto 94MB19:45
ryuoNo...19:49
ryuo69M vs 223M?19:50
ryuoACTION boggles.19:50
ryuohm.19:50
ryuoscrew it. i'll make a tar of them and see how much it changes.19:54
ryuonow testing xz.19:57
*** crash_ has joined #crux-devel19:58
ryuowow. xz... isn't much of an improvement.20:03
pedjatbh, I don't see the point of the whole exercise20:06
pedjamy kernel modules dir is 75Mb, which is really not that much :)20:08
*** onodera has quit IRC20:08
pedjaand jue's comment from fs#826 is still valid, no real gain in it, imho20:15
pedjabut at the end of the day, not my decision to make :)20:16
ryuoactually it's not valid for zlib, but still is for xz.20:16
ryuoi just tested it. the kmod pkg links to /lib/libz.so.1 or w/e20:16
ryuoaccording to ldd20:16
pedjaI was referring to 'compressing them [is not] not a real win'20:17
ryuoOh.20:17
ryuofrinnst: http://dpaste.com/0Y53BPA20:17
ryuothis is how to enable only zlib in kmod.20:17
ryuoit doesn't suffer from the /usr problem jue mentioned, as its library is in /lib. i just checked.20:18
ryuoand a compile test supports my assessment.20:18
*** xor29ah has quit IRC21:30
*** xor29ah has joined #crux-devel21:32
*** ryu0 has joined #crux-devel22:26
*** xor29ah has quit IRC23:47
*** xor29ah has joined #crux-devel23:49
*** xor29ah has quit IRC23:59

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