IRC Logs for #crux-devel Friday, 2007-08-10

*** pedja has quit IRC02:02
tilmanRomster: profiling results? :D02:21
tilmanprologic: noted. i need to talk to jaeger and sip about charlie's non-existing responsiveness02:22
Romsterah bosu with everyhting i'll get them this night.02:55
*** treach has joined #crux-devel03:23
tilmanthe longest port name that we have currently is 36 characters long04:25
tilmanthe average is 10 chars04:26
tilman</fyi> ;p04:26
treachmmh, what's that?04:27
treach36 chars for a portname.. that's half an essay. :x04:27
tilmanwanna guess in which repo it is?04:27
tilmani'm thinking of making 47/48 the hard limit04:28
treachallow 64 bytes, and watch anton go ballistic :D04:29
tilmananyone who uses 64 characters for a port name is a) insane and b) stupid04:29
tilmani think it's safe to not allow that04:30
treachI know, I was just kidding. :>04:31
tilmanbtw, the memory overhead for that can be fixed with 3 lines of code04:31
tilmanwithout adding the extra malloc call overhead :)04:31
tilmanit's a neat trick. see pkg_package_entry.c:3604:32
treachmalloc is good, it's proper memory management, remember? ;)04:32
tilmani've erased it from my memory banks04:32
tilmanblah, i suck04:44
tilmanthe current limit *is* 63 chars04:44
*** treach has quit IRC05:15
Romsterthought jack-audio-connection-kit would of been quite long, figures that xorg names beat it :P05:19
*** treach has joined #crux-devel05:33
tilmanRomster: figures what?05:36
Romsterthat your xorg repo had the longist named port05:42
tilmanthat xorg- prefix has some disadvantage05:43
Romsterlike long dependency line05:44
Romsterconsidering xorg-* ports are in the xorg/ repo do they need the xorg- prefix?removing them wouldn't off a computer wouldn't be any harder than a for loop over the xorg/* and pkgrm.05:46
Romstereh i word tings a bit funny..05:47
Romsterchanged my thought while typing..05:47
tilmanwell, the package database contain contain port information05:48
Romsterguess it would make more complications or something i'm overlooking.05:48
RomsterFATAL: Module oprofile not found.06:00
RomsterKernel doesn't support oprofile06:00
Romstercrud... so i haev to enable a module.. but which..06:01
tilmanit's near the end in menuconfig06:01
tilmaninstrumentation support -> profiling support06:01
tilmanand oprofile system profiling as a module06:02
Romsterright wasn't too hard to find thanks for that.06:04
Romsterand yeah i'm making it as a module.06:05
Romstermaybe that should be in the oprofile README but i guess it's self explanitory when ya run it for the first time.06:06
Romsterwhat the FATAL: Error inserting oprofile (/lib/modules/ Unknown symbol in module, or unknown parameter (see dmesg)06:24
RomsterKernel doesn't support oprofile06:24
Romsterhmmz maybe this kernel is too old..06:24
tilmandid you reboot with the new kernel?06:24
Romsterobviously not06:24
tilmanjust loading the module won't help06:24
Romster 21:27:03 up 28 days,  9:21,  1 user,  load average: 0.37, 0.75, 0.8006:24
tilmanto me it isn't obvious :p06:25
Romstercan't just modprobe :/06:25
Romstererll i never disconnected did i now :P but then again ya never know if i had a bnc or not.06:25
Romsterthey haven't made the kernel panless yet...06:26
*** cptn has joined #crux-devel06:27
*** ChanServ sets mode: +o cptn06:27
Romsterhi cptn06:27
tilmanhi cptn06:27
cptnI've actually started a small sqlite benchmark script this morning :-)06:29
tilmanwrt /var/lib/pkg?06:29
tilmangot any numbers yet?06:30
cptnwell, none that support flat file so far ;-)06:33
cptn$ sh bench106:35
cptnsqlite: 40 s06:35
cptnpkginfo: 109 s06:35
cptnscript is here:
cptnmmh wait, there's a bug in there06:36
cptn is better06:36
cptnis kinda hard to come up with a fair benchmark06:37
cptnsince in a sqlite db, you'd probably have separate tables for files and ports06:37
cptnwhich makes checking if a port is installed a lot faster than with the current db06:37
cptnhowever we could easily have a separate file for the port names as well06:38
cptnand when I say "we" I mean "you" ;-)06:38
cptnupdated script took 50s / 113s (sqlite/pkginfo)06:39
tilmanis this bench1 or bench2?06:41
cptnusing grep in bench2 seemed close enough :-)06:44
tilmanyour db is quite large, eh?06:45
tilmanjust timed pkginfo like you did in bench2, and it takes 10s06:46
cptnI have 558 packages installed06:46
tilman252 here :D06:46
cptnso in my previous benchmark, that means even though sqlite saved over 50s, that's still in the region of tenth of seconds difference per package06:47
tilmancptn: before declaring that we need to move away from the flat file, i'd like to experiment with optimizing our code06:47
tilmanlike using mmap on the db06:47
cptnwell, yeah06:48
tilmanofc that means i need to have a proper db to test with of course06:48
tilmans/ofc / ;)06:48
tilmancptn: i really like that we can link pkgutils statically atm06:49
tilmanif we do this with sqlite3, the binary will probably be huge06:49
tilmanmmh, maybe not06:49
tilman388kb for
cptnI'd even say that the savings in time are so minimal compared to the build time that speed really isn't the reason here06:50
cptnif the API allows you to develop faster, that would be a good argument06:50
cptnbut saving a couple of tenth of seconds... not sure06:50
treachIt should also decrease the amount of code you have to maintain, right?06:51
cptneven if someone would use a system with only binaries, the download time of the package would still be larger than the time spent in DB code06:51
tilmanbut probably *that* much06:51
tilmanfirst we should make sure that we understand what the problem is06:52
tilmanwhat operations exactly are slow?06:52
cptndoesn't use the DB06:52
tilmanexactly my point06:52
cptnbut the footprints06:52
treachcptn: I know.06:53
tilmanfootprint cache ftw06:53
treachBut is there any reason not to stuff the footprint in the db?06:53
treachif you're using sqlite, I mena06:53
treachplease tell me to shut up if you think I'm fstupid..06:54
tilmancould do that, sure06:54
*** sepen has quit IRC06:55
Romster753 packages installed here.06:55
treachwe should get jaeger to run those tests. :>06:56
Romsterand i haev to reboot to do oprofile tests..06:56
tilmancptn: did you see ott's tre patch in flyspray?06:57
Romsterlol with gnome and whatever else he has, where's yahfri when ya need soemone with alot of installed packages...06:57
tilmantre is a drop-in replacement for regcomp etc06:57
cptntilman: yeah, I saw that one06:59
cptntilman: seems harmless :-)06:59
Romsteranother anoyance is installed package then the Pkgfile goes byebyes and ya left with a built package but no Pkgfile to make a new version and host it yourself. so storing the Pkgfile would be a nice feature.06:59
tilmancptn: i hope Romster can test it eventually.06:59
Romsteri gotta reboot and loose 28 days uptime :/07:00
cptnbut the default is not to use regex07:00
cptnso using tre wouldn't have an impact on fsearch07:00
tilmanand btw07:01
tilmani think it's a super feature to be able to repair the pkg db with vim07:02
tilman(without a hex editor)07:02
Romsterso a cache for fsearch alone maybe enougth..07:03
Romsterwhat does prt-get cache, cache exactly?07:03
treach"cat /var/lib/pkg/prt-get.cache"07:07
Romstereh looks kinda pointless cache to me..07:11
treachwell, it's not07:12
cptnRomster: with all due respect, but you should really think about what you say07:14
cptnthat cache has been benchmarked, and shown huge improvements07:14
cptnyet you go along and claim it's looks pointless07:14
cptnat the same time you want pkgutils to use sqlite07:15
cptnwithout understanding what the impact would be07:15
Romsterwell is gotta do some caching why not footprints too?07:16
Romsterargh oh i hate firefox bugs...07:16
Romsterhmm i've had 4 hours sleep after being up 24 hours, sorry i shouldn't think more before typing.07:18
*** jjpk has joined #crux-devel07:19
Romsteryou want it easy to repair yet i've yet to have a broken database, may be someone has.. but the chance of that is how many percent and there can be redundent backups.07:20
cptncare to try this:07:20
cptn(run fsearch --generate; fsearch <term>)07:23
Romstermaking the cache takes some time heh.07:29
cptnyou probably have lots of ports in your /usr/ports?07:33
Romsterhmm not a bad idea..07:33
cptnunlike prt-get, the script will also take all ports07:33
cptnso if you have inactive repos in /usr/ports, the script will index them too07:33
cptnbut then again, it's only a 24 lines shell script07:33
cptnand IMHO it's fast enough07:34
Romsteryeah i see a -all option to show all and no option t show the installed files, so your saying to leave the prt-get fsearch as is and just use a shell script for a replacment search.07:35
cptnno, it was rather meant as a workaround for prt-get not caching files07:36
Romsterthe ideal way would be for pkgadd to store a file of installed files. there there is no need todo a cache, prt-get can fsearch off the database of files, for installed programs, just gets a issue for not installed programs :/ hmmz i think i should leave this for another day..07:38
cptnpkgadd does that07:38
cptnprt-get fsearch _only_ searches the ports tree07:38
cptnif you want to search installed files, use 'pkginfo -o <file>'07:38
cptnwhich only searches the DB07:39
cptnso there is a cache for files which are installed07:39
*** deus_ex has joined #crux-devel07:39
Romsterhmm yeah your quick bash script is the method i was looking for to do a fast look for a file and see what package it belongs too.07:40
Romsterknew i could type something like that but to have it in prt-get.07:40
cptnit's different functionality though07:40
cptnthere's no guarantee a file will be in a port forever07:41
cptnso if you want to find the owner of a file in your filesystem, fsearch isn't safe07:41
cptnbecause the port might have removed that file07:41
Romsteralong that line is there a program now that shows what files that don't belong to any port? being ghost files.07:42
tilmanpkgorphan i think\07:43
Romsterbut thats a diferent thing altogether and not related to this current discussion.07:43
Romsterhmm sounds familer but i don't see it..07:44
cptnthere's prtorphan, but it does something different07:44
Romsterah no thats a diferent thing again..07:45
cptnit's not too hard to implement though07:45
cptnbut probably fairly slow07:45
tilmanyeah, that's what i just thought07:45
cptnsince you'll have to look at every file on your disk07:45
cptnbut basically you do:07:46
cptncreate a list of files on disk07:46
cptncreate a list of files in the pkgdb07:46
cptnthen, merge those lists and pipe them through |sort|uniq -u07:47
Romsterhrmm its kida like locate but for packages.07:47
tilmangotta run07:48
cptnoh noes07:48
Romsterlater tilman07:48
tilmancptn: it's good to see you tossing around ideas etc :)07:48
cptnsee you07:49
Romsteri know i maybe a pain around here, tell me to bug off if i'm geting under anyones skin.07:49
Romsteryeah your been on more oftern cptn have you got more freetime now?07:49
cptnI just missed tilman :-)07:50
cptnno, it's more that we've talked about pkgutils a bit yesterday07:50
Romsterhe is nice person :)07:50
cptnand I kept thinking about it, so I dropped by today again to continue07:50
Romsteri see.07:51
cptnmmmh, the script find false positives if you're looking for a string which is in the package name07:51
Romstermy views are diferent i wanna see a sqlite database of installed packages, but now thining of it i'd haev to see the good and bd points, to if its worth it.07:52
cptnI think if there are convincing arguments for sqlite they'll seriously considering it07:53
Romsteri'll have a good think and seeing a good API from pkgtools for all other utilites to use.07:54
cptnone of the things we've talked about previously was to have an abstraction layer for the package database07:55
cptnallowing to change the backend07:55
cptnthis would mean that tilman could continue with the rest of pkgutils, while those interested could port the backend to sqlite07:55
Romsterfsearch is preaty much a locate for packages, but its able to get stale fast too, unless pkgrm and pkgadd is involved in keeping the cache uptdo date for fsearch07:55
cptnerrm, no07:56
cptnfsearch searches the ports tree07:56
cptnpkgrm and pkgadd have no effect there07:56
cptnyou just have to regenerate it whenever you change something in /usr/ports, i.e. by running ports -u or manually07:57
Romsteryeah that would be a good idea. having the backend seperate.07:57
Romsterah sorry your correct...07:57
Romsterthink i'm confusing two things together that arn't related.07:57
*** jaeger has joined #crux-devel07:58
Romsterpkgmk and ports -u, would be what alters it. now thats a point footprint missmatches, after installing a package then doign a ports -u reverts the .footprint thats a messy situration unless you also use installed files for installed packages and the rest from the ports tree for not installed packages, all this is gonna get mighty complex and where after KISS.08:00
Romsterhi jaeger08:00
treachthe more I think about it, the more I ask myself if you couldn't remove the ports tree entirely as such, and wether that would still be KISS..08:01
treachoh well, there's no way I could code that anyway. :P08:01
jaegermorning, all08:02
cptnhi jaeger08:03
Romsteri'll have a think of it and see what i can type out in a text file on a possable method.08:03
cptntreach: it's a nice idea, though :-)08:03
treachyeah, put it all in a sql db.08:03
treachit's just data anyway, no real need for it to be in discrete files, if you think about it.08:04
cptnthe main problem would be the sync methods08:04
cptnsince they're currently file based08:04
treachaye, I can see some pretty extensive rewrites needed for all tools involved.08:04
cptnunless you could mount the db...08:05
Romsteronly to pkgadd and pkgrm if i'm thinking corectly.08:05
cptnno, for ports -u08:05
Romsterand ports -u08:06
Romsteri'm thinking of instlaled packages .footprints arn't always the same as what ports -u changes them back too, what pkgmk modifyed them for the current system..08:06
cptnyes, that's why you use pkginfo -o...08:07
Romsterdue to a optional program being present or not on the system.08:07
Romsteri hadn't had used that option08:07
Romsterso only 'ports -u' would have to generate the fsearch cache, now thats looking viable.08:11
Romsteronly prt-get edit and manually editing and pkgmk'ing a package wouldn't be put in the cache becasue it would only be in ports -u.08:12
Romstercan't find no pkgorphan either..08:14
treachthree cheers for cp/paste in screen. :)08:18
Romsterah no one made it into pkgorphan...08:21
*** deus_ex is now known as pedja08:34
*** cptn has quit IRC09:11
*** pedja has quit IRC09:59
*** sepen has joined #crux-devel10:51
*** sepen has quit IRC13:09
aonwhat about wxgtk?13:14
* aon is too lazy to read backlog and the hilights weren't coherent enough :)13:14
jaegerI just wondered if there was a specific reason for 2.6 being in ports instead of 2.813:22
jaegerit's not a problem, I was just curious13:22
aon41beta could work, though13:27
tilmanhello aon \o/13:28
* aon looks at the amount of ck4up diff13:29
* aon shrieks13:29
jaegerok, just wondered13:34
jaegermight play with it and try to patch13:35
tilman[ Some people don't think performance issues are "real bugs", and I think13:35
tilman> such people shouldn't be allowed to program. ]13:35
tilman -- torvalds on the git ml13:35
j^2hey aon13:35
jjpkI would consider performance to be a macro bug.13:37
jjpkIt will never be fast enough, but it is worth working on. :D13:38
*** deus_ex has joined #crux-devel13:44
*** deus_ex is now known as pedja13:44
aon  hey jjpk13:51
aonand j^2 too :)13:51
jjpkaon: o/13:51
*** Viper_ has joined #crux-devel13:55
*** jjpk has quit IRC14:04
Romsteromg aon is here :P14:08
Romsterhi dude14:08
Romsteryeah i found that out wxgtk 2.8 also breaks xmule too.14:09
aonyep, and amule14:11
aonthere's a patch for amule, though14:12
tilmanxmule, haha14:14
aonarnuld is on the ratpoison ml :DD14:33
treachfor a moment I missed the "ml" part. :/14:34
tilmanlol :D15:09
tilmanaon: see his latest few blog postings?15:09
tilmansome pretty fucked up stuff15:09
aoni'll have a look15:10
aoni feel like i have hangover :)15:10
aonalthough it's been a month since i had any alcohol :)15:10
aonomg wtf15:12
aon1.) Most people just live their entire lives without understanding "why I am doing this job/business ?" or "why I chose this profession?"15:12
aonthat doesn't seem too accurate :)15:12
treachmh, s /understanding/caring/ :/15:18
treachI read today about a young guy, 20 something, who was "treated" two days for O.D. What was actually ailing him was a stroke. :/15:20
treachresulting in life in wheelchair apparently15:23
aonthat really sucks15:26
*** jaeger has quit IRC15:31
*** Viper_ has quit IRC15:42
tilman" C / C++ / AJAX software engineer"15:45
tilmannow that's a nice combination15:46
aonnot impossible, tho15:52
*** jaeger has joined #crux-devel16:01
clbUpdate from opt: 10 Aug 21:09 - [security] xpdf: added a patch for CVE-2007-3387 || 10 Aug 20:42 - wine: updated to 0.9.43 || 10 Aug 19:20 - splint: updated md5sum16:11
aonwonder if i missed something16:12
treachwho needs xpdf when there's rakt? :p16:15
tilmanthat reminds me...16:19
tilmantreach: your poppler pkgfile looks stupid (tab messup) :P16:22
tilmanuse cat or less to see what i mean16:23
treachsorry, can't fix it right now.16:23
tilmani suspect your tab settings in vimrc might be silly16:23
jaegerset ts=silly16:23
treachheh, I don't have any. :p16:23
treachit's just defaults.16:24
tilmanset sw=416:24
tilmanset sts=016:24
tilmanset ts=416:24
tilmanset expandtab16:24
tilmanand live happily ever after16:24
treachYes Sir!16:24
aon4 :-(16:24
tilmanyou can change 4 in *all* places to 8 if you prefer16:25
tilmanbut note that any other values are totally fucking insane16:25
tilmanodd numbers are the devil here, obviously16:26
tilman2 is sick16:26
tilman6 is just strange16:26
tilmanQED :)16:26
treach4 sound fine to me. 8 is just nuts.16:26
aon8 is great16:27
treachno it isn't..16:27
treachI try to stay within 72 columns..16:27
aon#include <linus_quote_about_functions_being_too_large_if_they_became_too_wide_with_8-space_tabs.h>16:28
aonactually i think that's in some kernel programming style guid16:28
treachthat's obviously bull, but if you nest deeper than three levels you need to rethink.16:28
treachAt least that's my motto.16:28
tilmana coworker of mine uses 2 spaces16:29
tilmanand he has a nesting depth of about 10 (or so it seems)16:30
aona friend of mine didn't indent at all some years ago :)16:30
treachTake him out through the backdoor. And shoot him in the head - repeatedly.16:30
aonnow he does, though16:30
aonwith 8 i think :)16:30
aonbut he has insane nesting depths, too16:31
treachthat was meant for tilman, but I guess you could do the same. :P16:32
aonnah, i don't fire guns in my free time :)16:35
treachtilman: I just cat'ed the file, looks fine that way to me, but if I open it with vim with the new settings I see your point. :p16:40
tilmanah, i saw it when i diffed it16:40
tilmancat and less really don't show it16:41
treachI'll fix it when I get my desktop back on line, all my settings for stuff like that is stuck there atm. :/16:41
treachpassword to the ftp, sitecopy conf etc, and I'm too lazy to redo it now. :)16:42
treachspeaking of lazy.. :p17:07
treachmobilix:~% vim -p $(for i in $(find . -iname Pkgfile); do17:07
treachecho "$i "17:07
clbUpdate from xorg: 10 Aug 22:28 - xorg-xprop: updated to
tilmantreach: did you hear about the mobilix company being sued by the guys who have the copyright on asterix etc? ;)17:35
tilmanback in 2001 or so17:35
tilmanstupid typical lawyer crap17:36
tilmanneat, the new wine has lots of bug fixes17:37
treachhow utterly unexpected. :)17:38
*** prologic has quit IRC18:04
treachyay, talk about a shocker.18:08
treachthe judge finally decided to give sco a firm kick in the groin it appears. :)18:08
*** prologic has joined #crux-devel18:46
*** treach has quit IRC20:51

Generated by 2.11.0 by Marius Gedminas - find it at!