IRC log of #maemo-ssu for Saturday, 2012-12-29

*** ruskie has joined #maemo-ssu00:09
*** int_ua has quit IRC00:11
*** jade has joined #maemo-ssu00:24
*** jade has quit IRC00:24
*** jade has joined #maemo-ssu00:24
*** Jaded has quit IRC00:27
*** ekze has quit IRC01:07
*** ekze has joined #maemo-ssu01:13
*** kolp has quit IRC01:53
*** MrPingu has joined #maemo-ssu02:03
*** luf has quit IRC02:20
*** arcean has quit IRC02:21
*** arcean has joined #maemo-ssu02:28
*** MrPingu has quit IRC02:38
*** joshgillies has joined #maemo-ssu03:11
*** NIN101 has quit IRC03:25
*** arcean has quit IRC03:28
*** joshgillies has quit IRC04:01
*** joshgillies has joined #maemo-ssu04:05
*** jade has quit IRC04:12
*** Jade has joined #maemo-ssu04:12
*** Jade has joined #maemo-ssu04:12
*** Martix has quit IRC04:46
*** jon_y_ has joined #maemo-ssu04:47
*** jon_y has quit IRC04:47
*** jon_y_ is now known as jon_y04:49
*** joshgillies has quit IRC04:59
*** M4rtinK has quit IRC05:01
*** AndrewX192 has quit IRC05:29
*** AndrewX192 has joined #maemo-ssu05:29
*** AndrewX192 has joined #maemo-ssu05:29
*** amiconn_ has joined #maemo-ssu05:54
*** amiconn has quit IRC05:54
*** amiconn_ is now known as amiconn05:54
*** DocScrutinizer05 has quit IRC06:04
*** DocScrutinizer06 has joined #maemo-ssu06:04
*** DocScrutinizer06 is now known as DocScrutinizer0506:08
*** jon_y has quit IRC07:14
*** jon_y has joined #maemo-ssu07:14
jon_yfreemangordon: is the formatting correct? http://paste.debian.net/219994/07:47
jon_ythis should really be done with Perl reporting :)07:48
*** nox- has quit IRC08:01
*** zogg__ has quit IRC08:51
*** zogg__ has joined #maemo-ssu08:52
*** kolp has joined #maemo-ssu09:31
*** joshgillies has joined #maemo-ssu10:10
*** luf has joined #maemo-ssu10:24
jon_yanybody has the source for kp52?10:24
merlin1991jon_y: Pali has ;)10:36
merlin1991jon_y: or you go here: https://garage.maemo.org/plugins/ggit/browse.php/?p=kernel-power10:37
jon_ymerlin1991: I just got it via git10:37
jon_yand now I'm wondering how to build it10:37
merlin1991so I'm too late :D10:37
merlin1991dpkg-buildpackage -rfakeroot -us -uc (inside scratchbox) ?10:37
jon_yI'm using a plain cross compiler10:38
merlin1991uh oh10:38
jon_ypali says he did not use scratchbox10:38
jon_yso I went and put up a VM for that :)10:38
merlin1991hm well you can probably run the makescript10:39
jon_yno make script, I think it uses dpkg10:39
jon_ywell, I wonder if dpkg-cross can come in and help10:39
merlin1991oh it has a makescript10:40
merlin1991just look it up in debian/rules ;)10:40
merlin1991bah this git clone takes forever10:40
jon_ymake -f debian/rules??10:40
merlin1991nah10:40
jon_yah wait, I'll just look in it10:40
merlin1991yep, and find the build call in there :)10:41
jon_yok, found it, now to guess where the working directory should be10:41
jon_ymerlin1991: where do you run dpkg-buildpackage at?10:43
merlin1991that is run at the level where the debian folder ist10:44
merlin1991s/ist/is inside/10:44
infobotmerlin1991 meant: that is run at the level where the debian folder is inside10:44
jon_yok10:44
merlin1991but the rules file builds in kernel-power-2.6.28/debian/build/kernel-power/10:47
jon_y?10:48
merlin1991well the debian/rules dls original maemo kernel (with wget), extracts it into said folder10:49
merlin1991applies patches and then runs make in there10:49
jon_ylooks like I really do need scratchbox10:50
merlin1991nah you can do pretty much the same thing10:50
merlin1991just need to throw your cross compiler in instead of host gcc10:50
jon_yok, I'll try that10:50
merlin1991you're not going to get fancy .deb ofc :D10:51
merlin1991just a proper zImage10:51
*** dhbiker has joined #maemo-ssu10:51
merlin1991(and modules)10:51
jon_yyes, I just need the zImage and modules10:51
jon_yI thought it was as simple as make menuconfig && make10:52
jon_ynow to figure out how to apply the patches10:52
merlin1991quilt10:53
jon_yok, installing that too10:53
merlin1991it's that line from debian/rules: cd $(KSRC) && ( QUILT_PATCHES=$(CURDIR)/debian/patches quilt push -a -q || test $$? = 2 )10:54
jon_yok, will do that10:54
jon_yI was under the impression the quilt was deprecated10:56
merlin1991look at the version of debhelper maemo is using (5)10:57
merlin1991there's loads of old cruft in various maemo packages .D10:57
merlin1991current debian-testing is using debhelper 9 :D10:58
jon_yso stable it is stagnant10:59
lufmerlin1991: is the libxml2 dsc file in cssu devel?10:59
lufmerlin1991: or where can I find it?10:59
merlin1991libxml2 dsc file for which version10:59
luflatest cssu (from gitorious)10:59
jon_ywell, at work, my company uses kernel 2.6.12 <- that is the most bleeding edge :)10:59
merlin1991I'm soon going to see what kernel I'll have to base my userspace stuff I'll be doing at work11:00
merlin1991I bet it's in the 2.6.* range aswell11:00
merlin1991btw luf why do you need the .dsc?11:00
lufmerlin1991: I need to patch libxml2 and I want to rebuild the package.11:01
merlin1991luf: it's here: http://repository.maemo.org/community-testing/pool/fremantle/free/libx/libxml2/11:02
lufthx11:02
merlin1991but why don't you simply clone the git repo?11:02
jon_yeek, where kernel config file11:02
merlin1991rx51-something11:02
merlin1991look it up in rules :D11:02
merlin1991also sorry luf, there's only the debs11:03
lufmerlin1991: I want dsc file not the .deb11:03
merlin1991silly maemo repo is non standard layout11:03
lufI cloned the git.11:03
merlin1991luf: you want http://repository.maemo.org/community-testing/pool/fremantle/free/source/libx/libxml2/11:03
lufI think I need the dsc file to build the package (am I wrong?)11:03
merlin1991nope11:04
merlin1991you need the content of the debian/ folder11:04
merlin1991more exactly you need at least debian/control debian/changelog and debian/rules11:04
jon_yreadme mentions rx51power-defconfig11:04
jon_ycan't find it :|11:05
lufmerlin1991: I see. Hmmm next lesson for me :D11:05
merlin1991jon_y: rules file has DEFCONFIG := rx51_defconfig11:05
merlin1991and it's using that later on for the make calls11:05
merlin1991or rather the make call (which prepares the source)11:06
jon_yok11:06
lufmerlin1991: I need this patch: http://git.gnome.org/browse/libxml2/commit/?id=a7e79f28689c574e0bbef17f4cb3da00249181ff11:07
merlin1991luf: just as a general note, the dsc file has definite information on a specific version of a package, it says that packages build from $source with $version had a source consisting of $tar.gz and possible $diff.tar.gz with certain checksums11:08
jon_ywhat is this rx71 board?11:09
merlin1991luf: you can either pull that branch into your repo and use git cherry-pick $commithash or you just redo what it does and reference the original11:09
lufmerlin1991: I know but I thought that it's prerequisite (and I see it isn't).11:09
lufmerlin1991: come on it's just 4 lines ... It's faster to write it :)11:10
merlin1991hence the "or" section of my line ;)11:10
merlin1991make commit msg of something like backported upstream commit a7e79f28689c574e0bbef17f4cb3da00249181ff11:10
merlin1991maybe include that url aswell11:11
merlin1991and as always test if shit explodes with the patch ;)11:11
lufmerlin1991: sure the test are the first thing ...11:12
jon_yI wonder if selinux and friends will be useful on the n900 :)11:16
*** joshgillies has quit IRC11:18
lufjon_y: I see no good reason for it.11:19
lufjon_y: in case of phone usage ;)11:19
lufGood zlib 1.2.7 is now working on my phone ... next step is to patch it with neon optimizations.11:21
lufmerlin1991: how to create merge request on gitorious from my local git repo?11:24
merlin1991you need to "clone" the repo on the webui, push your changes into the cloned repo and then use the webui again11:25
luf:(11:25
lufI suppose it's the preferred way for pushing changes into libxml2 ...11:26
jon_yhmm git checkout lists 2.6.28.10-power51r111:31
lufmerlin1991: ok, I creted the merge request11:40
lufBTW I tested libxml2 with stock zlib and also newer zlib.11:42
lufmerlin1991: I have dependency question :)11:45
luflibxml2 is depending on zlib and newer zlib crash older libxml2 ... I don't want to create cycle in dependency tree. Is there some way I don't see or have I to use also Depends: libxml2 (>= some_version) in zlib?11:46
kerioluf: no, Conflicts: libxml2 (<= some_version)11:49
kerioor, i suppose, (< some_version)11:49
keriobut to be fair, it should be in libxml211:49
kerioalso, check every package listed in apt-cache rdepends libxml211:50
keriogregoa: packaging help please - newer version of a package doesn't work properly with an older package it's depended upon - is it right to use Conflicts on the new package?11:54
freemangordonluf: did you find a wat to benchmark libpng?11:55
freemangordon*way11:55
merlin1991luf: could you elaborate on what does not work with what?11:59
merlin1991I'm not sure I got the whole newer older thing right in my head :D12:00
luffreemangordon: no I didn't :(12:00
lufmerlin1991: (from zlib homepage) If you are using libxml version 2.7.6 or earlier, you will need to update libxml to version 2.7.7 or later  before installing zlib version 1.2.4 or later.  libxml 2.7.6 and  earlier made unnecessary assumptions about the undocumented internal  structure of zlib that were changed in zlib 1.2.4 and result in libxml  crashing.  This was fixed in libxml 2.7.7.12:00
merlin1991which version of libxml do we ship?12:01
lufmerlin1991: and the N900 ends in never ending boot loop when I upgrade just zlib (no widgets displayed and it reboots in few seconds after boot)12:01
luflibxml2-2.6.32...12:01
merlin1991well then libxml needs a dependency on zlib <= something12:02
merlin1991or a conflict on zlib >= 1.2.412:02
merlin1991though I'd love to hear gregoa on what's the proper debian way12:02
lufmerlin1991: I need to upgrade zlib but it's impossible with current libxml2. For sure I have to change something in zlib package.12:03
merlin1991why do you need to upgrade zlib?12:03
freemangordonluf: yes, why?12:04
lufmemleaks, some fixes and I want to use neon patches (from meego againist 1.2.7).12:05
merlin1991the dependency system works the other way around, libxml needs zlib so any incompatabilities with zlib have to be modeled in libxmls control12:05
freemangordonluf: memleaks? link please12:05
freemangordonluf: BTW can't you just use harm zlib?12:05
lufwhat is harm zlib? do you mean from meego?12:06
freemangordon(I assume it already contains neon patches)12:06
freemangordonI mean from harmattan12:06
freemangordonharmattan != meego12:06
lufhttp://zlib.net/ChangeLog.txt: Changes in 1.2.4 (14 Mar 2010) - Fix memory leaks in gzclose_r() and gzclose_w(), file leak in gz_open()12:07
freemangordonharmattan is what runs on n9/50. meego is what runs nowhere12:08
lufwhere can I find harmattan zlib?12:08
freemangordonluf: are you sure those patches are not backported by nokia in fremantle zlib?12:08
lufI'm not sure. But what is the problem with libxml2 patch?12:09
freemangordonit is possible not only zlib depends on those internal functions12:10
merlin1991freemangordon: other way around12:11
freemangordonaah, sorry12:11
freemangordonI am still having my first coffee for the day :D12:11
merlin1991but we still could have $application crash suddenly because of this12:11
luflibxml older 2.7.7 depends on zlib internals what is wrong behaviour.12:12
keriodammit, adding a conflict in zlib sucks :s12:12
keriowe don't really know it's the only package affected by this12:12
freemangordonluf: correct. but how to check if there isn't another lib/app doing the same?12:12
keriofreemangordon: check every package in apt-cache rdepends zlib12:13
lufYes we can have application which failed because of change in obexd.12:13
keriofreemangordon: to be fair, it's a bug in libxml12:13
lufIt's zlib internal which shouldn't be used at all.12:13
kerioand in every package that did that12:13
freemangordonok, ok :D12:14
freemangordonluf: do we know those function names?12:15
luffreemangordon: I don't understand. Please rephase "those"12:15
merlin1991he's asking if we know what changed, so we can look for the structure/function calls in other sw12:16
freemangordonthe names of the internal function which libxml2 uses wrongly. we can grep for them in /usr/...12:16
merlin1991jon_y: any luck building the kernel?12:16
keriofreemangordon: if it's a struct access, how can you fix that?12:16
freemangordonkerio: fix?12:17
kerioi assume you wanted to add some patches to zlib so that the old stuff still worked12:17
freemangordonno12:17
freemangordonyou got me wrong12:17
freemangordonI want to check if there is other sw but libxml which is buggy12:18
freemangordonso we can either fix it (somehow) if it is FOSS or declare "we can't upgrade to newer zlib" if the sw in question is closed source12:18
lufhttp://pastebin.com/vAtHs08S12:19
merlin1991or find nothing and be superhappy :D12:19
keriofreemangordon: can we really limit ourselves to check the reverse dependencies of zlib, though?12:19
keriois there a way to check which libraries are referenced by binaries?12:20
merlin1991kerio: fremangordon wants to grep over ALL the files12:20
merlin1991not just stuff in dependencies12:20
keriooic12:20
lufmerlin1991, freemangordon: have you took a look into the patch?12:20
merlin1991luf: you move too fast for me :D12:21
freemangordonluf: yes, but I don;t see a patch12:21
freemangordonmerlin1991: you are slow :P12:21
merlin1991freemangordon: http://gitorious.org/community-ssu/libxml2/merge_requests/112:22
freemangordonluf: which is the buggy code, the part with gzread/gzunwind? or gzdirect?12:22
lufelse part is buggy.12:22
lufI think because of this: (z_stream *)context12:22
freemangordonok, i'll greap for those function in /usr/lib and usr/bin12:22
lufgzread and gzunwind is ok12:23
freemangordonluf: I see12:23
kerioheh, pointer conversions12:23
freemangordonso we can;t really verify that12:24
merlin1991which zlib version do we have?12:26
lufkerio: nearly everything is depending on zlib :D12:26
kerioyay! :D12:26
lufzlib-1.3.312:26
lufSorry.12:26
lufzlib-1.2.312:26
keriothat means that making it faster will make everything faster!12:26
freemangordonfor sure12:26
freemangordonyep, it is 1.2.312:26
lufChanges in 1.2.3 (18 July 2005)12:27
merlin1991according to this bugreport: https://bugzilla.redhat.com/show_bug.cgi?id=675207 1.2.3 is enough to apply the fix and be happy12:27
povbotBug 675207: was not found.12:27
freemangordonhmm, seems to me that only libxml is abusing those12:28
lufyeah some zlib developer without enough coffee wrote that  ...12:29
merlin1991hm so where's the actual problem now?12:30
merlin1991we apply that fix, then we are free to upgrade zlib to whatever we want12:30
merlin1991luf: or does it bootloop even with the fix applied?12:30
freemangordonmerlin1991: my concern was that there could be another piece of sw abusing zlib12:31
lufNo bootloop after installing fixed libxml2.12:31
freemangordonit torns out it is onli libxml to do ti12:31
freemangordon*turns12:31
merlin1991okay so let's fix libxml :)12:31
* freemangordon is still having his first coffee12:31
merlin1991so your not one of those people who down the whole first mug in a splitsecond?12:32
merlin1991*you're*12:32
freemangordonit is double esspresso, I can't just drink it up12:33
freemangordon:D12:33
luffreemangordon: close your eyes and drink! :D12:34
merlin1991well I don't drink coffee at all :P12:34
freemangordonmerlin1991: bad decision. except if your health stops you12:35
merlin1991hm why is it that bad?12:35
merlin1991my reason not to drink it is the taste, I don't like it12:36
* merlin1991 ducks12:36
freemangordonmerlin1991: it contains a lot of good stuff. And I like the taste, I have mine without sugar :D:D:D12:38
freemangordonok, lets try to find a way to benchmark libpng and zlib (when it is ready)12:38
luffreemangordon: drinking coffe is the same as drinking hot asphalt ...12:39
freemangordonnever tried hot asphalt :P12:39
freemangordonDocScrutinizer05: please, step in and explain about the coffee :D:D:D12:40
luffreemangordon: no neon opts in harmattan code (https://build.pub.meego.com/package/files?package=zlib&project=home%3ACaCO3) or I'm unable to find right sources/git12:40
kerio~coffee12:40
infobotit has been said that coffee is the reason the net exists, the drug of choice for a GNU generation, http://www.chez.com/emarsden/downloads/coffee.el, /usr/share/doc/HOWTO/en-html/mini/Coffee.html, geiseri's favorite beverage12:40
freemangordonkerio: thanks12:41
* freemangordon hands kerio a cup of double esspresso, black as moonless night12:41
lufAnother visit :( I have to leave now ...12:41
* luf afk12:41
merlin1991benchmarking zlib should be easy, stream fixed data through a really dumb programm and time that12:41
freemangordonmerlin1991: does zlib has a binary?12:42
freemangordon*have12:43
freemangordoni mean executable12:43
freemangordonbesides .so file12:43
lufgzip?12:43
freemangordonso we can pipe /dev/random | compress | uncompress | /dev/null12:43
lufor compress is better :) I always forget it.12:44
merlin1991there's zlib-bin with "sample programs" according to the package description12:44
luffreemangordon: if /dev/random is fast enough (usually urandom is better)12:44
merlin1991also you don't pipe random into a compression engine and the time that vs other random data12:45
freemangordonbtw I doubt neon in zlib will make it bette,12:45
freemangordon*better12:45
merlin1991you dd some random into a file and pipe that to both incarnations12:45
freemangordonpure ARM memcpy is faster than NEON optimized memcpy AFAIK12:45
freemangordonfor small chunks that is12:46
freemangordonhmm, there is pngtest in libpng12:52
freemangordonyep, pngtest will do the job12:56
freemangordonmerlin1991: do you still have stock device?12:56
freemangordonaah, scratch that, i'll use mine12:56
*** dhbiker has quit IRC12:56
*** dhbiker has joined #maemo-ssu13:03
jon_ymerlin1991: yeah, build complete13:06
jon_ytrying to figure out how to build the injection drivers13:06
jon_yor if I need it at all13:06
keriofreemangordon: have you ever played gyakuten saiban 3?13:06
kerio(phoenix wright: ace attorney - trials and tribulations)13:06
freemangordonNFC what that is13:07
keriothen this is going to be a bit less meaningful13:07
keriohttps://www.youtube.com/watch?v=HMnrl0tmd3k13:07
jon_ykerio: that finger pointing game? :)13:08
keriojon_y: >:C13:08
kerioone could say that i object to that classification13:08
jon_y:)13:08
jon_yI haven't actually played it, just seen screenshots of it13:08
kerioit's a really nice adventure game13:09
jon_yanybody know if the wireless inject modules are tied to a kernel version?13:10
keriothey're kernel modules, ofc they are13:11
jon_yI was wondering how to build them13:11
jon_ysince Pali's git checkout has them empty13:11
keriothe version string has to match, or you can try forcing them to load but it's not guaranteed to not make everything asplode13:11
jon_ykernel and modules are already built13:11
jon_yjust need the wireless injection drivers13:12
freemangordonluf: did you upload neon libpng .deb somewhere?13:14
jon_yany ideas where these files come from? compat-wireless-2.6.tar.bz2 compat.tar.bz2 wireless-testing.tar.bz213:19
merlin1991jon_y: well the rules file extracts compat-wireless-2.6.tar.bz2 compat.tar.bz2 and wireless-testing.tar.bz2 then touches $(COMPAT_WIRELESS_TREE)/compat_version runs cd $(COMPAT_WIRELESS_TREE) && GIT_TREE=$(WIRELESS_TESTING_TREE) GIT_COMPAT_TREE=$(COMPAT_TREE) GIT_COMPAT_WIRELESS_TREE=$(COMPAT_WIRELESS_TREE) ./scripts/admin-update.sh  and cd $(COMPAT_WIRELESS_TREE) && ./scripts/driver-select wl12xx13:19
merlin1991and  then builds with $(MAKE) -C $(COMPAT_WIRELESS_TREE) $(NJOBS) KLIB_BUILD=$(KSRC)13:20
jon_yhmm13:20
jon_yI can't find any script called admin-update.sh eitger13:20
jon_y*either13:20
merlin1991it's in one of those tar.bz2s13:20
jon_yso, where are those files from?13:21
merlin1991hm good question13:21
merlin1991wl1251-maemo-source is a package13:22
merlin1991prob contains them13:22
merlin1991it's listed in the build depends of kernel-power13:22
jon_ygetting it now13:22
*** M4rtinK has joined #maemo-ssu13:28
*** int_ua has joined #maemo-ssu13:29
jon_ymerlin1991: yup it's in there13:34
*** NIN101 has joined #maemo-ssu13:36
jon_yhuh, building the wireless modules will call modprobe13:50
jon_yI wonder what was it about13:51
*** FIQ has joined #maemo-ssu14:02
freemangordonluf: http://pastebin.com/jFQGBX5814:18
freemangordondoes not look good14:19
freemangordoneither my benchmark is flawed, or neon optimizes nothing14:19
freemangordonyes, it is my benchmark that sucks :(14:25
* ShadowJK wouldn't really be surprised if there was no performance difference14:25
*** bsdmaniak has joined #maemo-ssu14:26
* freemangordon too14:26
ShadowJKmost of png cpu time would be in the entropy coding things, so basically in the equivalent of gzip14:27
DocScrutinizer05I have a slight feeling of deja-vu14:42
DocScrutinizer05optimizing ong, heard that before14:43
DocScrutinizer05png that is14:43
DocScrutinizer05with same results14:43
DocScrutinizer05or I dream it up14:44
freemangordondamn, too much jitter :(14:47
luffreemangordon: I uploaded. _you_know_who/~luf/libpng14:50
freemangordonluf: BTW __ARM_NEON__ is nowhere defined ;)14:50
luffreemangordon: what's wrong with benchmark?14:51
ShadowJKjpeg would be interesting, the dct could benefit from a neon dct :)14:51
freemangordonI can't make a deffinitive conclusion14:51
jon_ywhen booting a new test kernel, it said something about modules.dep and then quickly rebooting14:51
jon_yany ideas?14:52
freemangordonbuild it in scratchbox/madde14:52
jon_yhmm ok14:53
luffreemangordon: why I'm able grep the function name in that case from .so?14:53
lufgrep png_read_filter_row_neon .libs/libpng12.so.0.49.014:53
lufBinary file .libs/libpng12.so.0.49.0 matches14:53
DocScrutinizer05jon_y: depmod ?14:55
jon_yDocScrutinizer05: ?14:57
DocScrutinizer05iirc there's a nice cmd called depmod14:57
DocScrutinizer05usually called like `depmod -a`14:57
DocScrutinizer05it might be related14:57
jon_ydepmod -a before booting the new kernel?14:58
DocScrutinizer05man depmod14:58
jon_yok, will try that14:59
luffreemangordon: I think the question is if the function png_read_filter_row is called in the benchmark.15:01
freemangordonluf: if it is not called for images in /usr/share/backgrounds, i'd say the patch is useless15:02
freemangordonbut i bet it is called15:02
*** arcean has joined #maemo-ssu15:02
freemangordonluf: I am doing another set of benchmarks, lets see15:07
DocScrutinizer05coffee? just having my first one today :-)15:10
jon_yDocScrutinizer05: looks like it booted15:11
jon_ybtw, does /lib/module/current need updating too?15:11
DocScrutinizer05good question15:12
DocScrutinizer05I think somebody nuked it15:12
jon_yit was complaining about symbol mismatches but loaded anyway15:12
DocScrutinizer05(the semantics of this symlink that is)15:12
jon_yI thought it might be related15:12
DocScrutinizer05well, did you build your modules for current kernel?15:13
jon_yyes, all the modules are rebuilt15:13
DocScrutinizer05since multiboot (real multiboot, not the friggin package)  /lib/module/current is kinda obsolete15:15
DocScrutinizer05aiui all kernels directly refer to their /lib/modules/<version>/15:15
jon_yis it related if I use u-boot?15:15
DocScrutinizer05ask pali15:15
jon_yok, will ask when he comes in15:16
*** Martix has joined #maemo-ssu15:19
freemangordonluf, ShadowJK: http://pastebin.com/yEep06UB15:23
freemangordondoes not look good :(15:23
freemangordonluf: i used "gcc pngtest.c `pkg-config --cflags --libs libpng` -DPNGTEST_TIMING -o pngtest" for pngtest binary15:24
freemangordonstill, that way of benchmarking could be totally wrong, but I can;t think of a better one15:25
freemangordonthough it clearly shows gcc4.7.2 is better :D15:26
lufI'm trying to get results from callgrind but I'm not successful15:26
*** bsdmaniak has quit IRC15:26
freemangordonluf: ofc it is up to you and merlin1991 to assess whether this neon patch makes any sense, having in mind it is not much tested.15:28
freemangordonTBH I was hoping for better results15:29
DocScrutinizer05did you google for results of similar efforts some years back?15:30
lufDocScrutinizer05: they told 20-30%15:31
DocScrutinizer05hmm, I have a feeling at best, of contrary results15:31
DocScrutinizer05for the life of mine can't recall where and when I heard of png optimization15:32
ShadowJKFor the specific function(s) or for decoding entire image? :)15:32
lufShadowJK: http://lists.linaro.org/pipermail/linaro-multimedia/2011-September/000074.html - it should be enough even for the specific function ;)15:33
DocScrutinizer05but I seem to have a record in a dark corner of my mind, that those efforts were basically futile15:33
lufDocScrutinizer05: it should be true.15:33
lufI'm just trying to find if it's called or not during benchmark.15:33
*** BCMM has joined #maemo-ssu15:34
ShadowJKah, Måns Rullgård :)15:34
DocScrutinizer05it also reminds me of the glamo disaster15:34
ShadowJKDid he write the neon patch for libpng too?15:35
lufShadowJK: ?15:35
DocScrutinizer05which in fact might be related, at least for my cloudy memory15:35
ShadowJKThe author of the email you linked15:35
luffreemangordon: can you try callgrind?15:35
freemangordonluf: what for?15:36
lufShadowJK: I know but I think there should be some message behind the name?15:36
luffreemangordon: to see if the _neon is called.15:36
freemangordonluf: just put g_warning before calling that function, along with static counter15:36
freemangordonand you'll be fine ;)15:36
lufAs you cen see in the last link I posted not all images is using the png_read_filter_row15:36
luffreemangordon: ok, I'll do it the hard way. Strange that callgrind is failing on my N900 ...15:37
ShadowJKluf; well I'm just seeing that he's saying "in principle" and making an assumption that the function could be sped up 4X?15:37
lufShadowJK: It's not proove. It's just 26.12%  pngbench  pngbench           [.] png_read_filter_row15:37
DocScrutinizer05anyway, in all your evaluations don't forget about the overhead for setup you usually introduce. Often once you count that in, you don't need benchmarks to see it's useless since overheat outweighs the speed boost of the core15:37
DocScrutinizer05overhead*15:38
DocScrutinizer05not heat15:38
lufShadowJK: and even more for the third image (however nearly nothing for the second image)15:38
freemangordonwhere is that pngbench?15:38
luffreemangordon: no idea. http://lists.linaro.org/pipermail/linaro-multimedia/2011-September/000074.html15:39
freemangordon:(15:39
ShadowJKluf; he estimated 4X for that function, if someone were to try optimize it. What happened later, did someone try to write optimized filter function, did they achieve mru's 4X guess?15:40
freemangordonooh, is that my friend mru?15:40
ShadowJKmru is Måns, yes15:40
lufShadowJK: I'm trying neon optimized patch for png_read_filter_row15:41
freemangordonaah, I see. well, i'll take everything he says with a grain of salt ;)15:41
lufShadowJK: I didn't say it has to work ... just I'm trying.15:41
freemangordonluf: pngtest counts filter functions call counts15:41
freemangordonFilter 1 was used 102 times15:42
freemangordonFilter 2 was used 422 times15:42
freemangordonFilter 3 was used 206 times15:42
freemangordonFilter 4 was used 14 times15:42
ShadowJKI'd say that if he sys 4X is possible, it's only possible for a superhuman coder to reach 4X :)15:42
freemangordonShadowJK: mru was the one ta say "no matter what you say about stable thumb2 on n900, I say it is impossible, period" :D15:43
ShadowJKlol15:43
freemangordonyeah, DocScrutinizer05 was on the same channel by that time, he can confirm15:43
freemangordonhowever15:43
ShadowJKHe also doesn't actually use or program any CPU that more than 500 people have access to ;)15:44
freemangordon:D:D:D15:44
DocScrutinizer05well, iirc he said "don't use it. period"15:45
ShadowJKAnd his thought process is proobably "why even bother with that stupid early buggy omap3 cpu, this omap5 is better in every way"15:45
ShadowJK:)15:45
freemangordonexactly15:45
freemangordonI was adviced o change the CPU :D:D:D15:46
freemangordon*to15:46
DocScrutinizer05ShadowJK: +115:46
*** ShadowX has quit IRC15:46
* freemangordon is out for cigarettes15:47
ShadowJKHe's actually discovered alot of the known hw bugs in omap hw too15:47
DocScrutinizer05and maybe he doesn't like to see his "babies" getting smashed ;-D15:48
ShadowJKMight've actually been why nobody used omap4 much, he discovered the ram controller is so slow and bad, that two Cortex-A9 cores perform worse than a single Cortex-A8 core of omap315:49
jon_yDocScrutinizer05: new kernel killed the camera :(15:50
DocScrutinizer05btw about OMAP bugs, did I tell about that non-public silicon error in omap4 WONTFIX, that causes IRQ misses?15:50
jon_yfcam won't load15:50
jon_yah well, will try __BUG() some more15:50
DocScrutinizer05jon_y: aiui it is like it ever been: new kernel renders fcam broken15:51
DocScrutinizer05we've seen that with every single KP rev15:51
jon_yso, normally fcam comes with a new build per kp release?15:52
DocScrutinizer05this IRQ detection flaw is a valid reason to not use OMAP415:52
kerioi think that kp ships the modules15:52
jon_yhmm15:52
kerioDocScrutinizer05: what do you mean by IRQ misses?15:52
jon_ykerio: really?15:52
kerio~pkg15:53
infoboti heard pkg is http://maemo.org/packages/15:53
freemangordonDocScrutinizer05: what? IRQ misses?15:53
jon_ykerio: I didn't see anything about fcam in the kernel build rules15:53
DocScrutinizer05when e.g. OMAP4 is playing video with hw accel, it may miss hw-IRQ that's meant to wake up HSI interface to modem. ->result: you miss inbound calls15:53
keriohm, apparently not, fcam just has to build a version for each KP release15:53
freemangordongreat :D15:54
DocScrutinizer05comment TI: will get fixed in OMAP515:54
kerioDocScrutinizer05: what the shit15:54
jon_yfcam is made of binary blobs?15:54
keriojon_y: no, it's open15:54
jon_yok, I'm trying to look for the modules15:54
jon_y*source15:55
DocScrutinizer05"work-around" the peripherals have to send IRQ over and over again, hoping for it to eventually maybe take effect X-P15:55
kerioDocScrutinizer05: i seem to repeat myself, but what the shit15:56
DocScrutinizer05a big modem chip manuf had at least 2 customers who blamed the modem chip to not go active when it should, it turned out it always been OMAP4 to not go active when modem asked for it15:57
freemangordonluf: you said you backported linaro libpng neon optimisation as well, ain't?15:59
DocScrutinizer05^^^ the "workaround" (c) TI16:01
kerioDocScrutinizer05: TI is just ahead of the game16:02
DocScrutinizer05and of course you don't find THAT in any public SiErr-list16:02
lufSo the neon filter is used ... but Filter cound doesn't correspond to the count ...16:02
kerionobody really wants to answer calls anyway16:02
kerioit's a way to increase battery life and reduce annoyances!16:02
freemangordonluf: well, it seems this optimisation does not really optimize16:03
freemangordonluf: do you have linaro code backported?16:03
lufwhat is linaro code? from libpng upstraem?16:03
freemangordoni guess16:03
freemangordonhttps://launchpad.net/linaro-multimedia-project/+milestone/2011.1016:03
lufWhat libpng are you using for non-neon?16:04
freemangordonluf: the same16:04
lufHow you get it?16:04
freemangordonrecompile16:04
lufSo you used neon enabled both? Or how you disable the neon opts?16:05
freemangordonmaster branch on gitorious does not have __ARM_NEON__ defined16:05
lufRight. Ok.16:05
freemangordonso when I want neon, i add -D__ARM_NEON__ in debian/rules CFLAGS16:05
lufIt's not needed.16:06
luf__ARM_NEON__ is defined somewhere else.16:06
freemangordonaccording to grep it is not16:06
lufStrange it is here ...16:06
freemangordonwhere?16:06
kerio"#ifdef __ARM_NEON__" "1 / 0;" "#endif" somewhere?16:07
kerioand then try to compile16:07
freemangordonkerio: been there,...16:07
kerio(is there a better way to cause a compilation failure?)16:07
freemangordonno16:07
lufI'm using SDK from nokia without thumb enabled and I don't need to define __ARM_NEON__16:07
lufAnd regarding grep I have neon optimized code.16:07
freemangordonluf: you have neon function compiled, but it is not called16:08
lufIt's called for sure.16:08
freemangordonluf: hmm, it could be this is compiler define16:08
freemangordonhowever, i used libpng from repos to test ARM only16:09
freemangordonand libpng from thumb repo to test -thumb16:09
lufARM only from cssu-devel and neon opts from master.16:10
lufOk. That's correct for sure.16:10
freemangordonARM only - stock libpng, the one in nokia repo16:10
freemangordonstill correct16:10
freemangordonluf: so, do you have upstream neon patches backported?16:12
lufcan you test the one from cssu-devel?16:12
freemangordonok16:12
luffreemangordon: No I don't have.16:12
luffreemangordon: BTW pngtest read and write the png so maybe the benchmark isn't very good for testing the read part ...16:14
freemangordonbut it test read times, we can ignore write times16:14
freemangordon*tests16:14
gregoakerio, merlin: in debian I would add "Breaks: libxml2 (<= 2.7.7)" to a newer zlib; but I don't know if the ancient packaging toolchain on maemo understands Breaks. Conflicts should do the same, more or less.16:19
keriowhat's the difference between Breaks and Conflicts?16:20
gregoakerio: http://www.debian.org/doc/debian-policy/ch-relationships.html 7.3 and 7.416:21
freemangordoniirc Breaks is supported16:21
*** arcean_ has joined #maemo-ssu16:22
kerioah yes, Breaks should be the case here16:23
gregoaok, Breaks was added (in debian) to dpkg in 1.14.6 (2007) and we have 1.14.25maemo3+0m516:23
keriothe old libxml and the new zlib can coexist while the new libxml is installed16:23
freemangordonhmm, no16:24
keriofreemangordon: during the upgrade process16:24
freemangordonshould be ok, yes16:25
freemangordonluf: CPU time used = 80.980 seconds (decoding 2.590,16:25
freemangordonencoding 78.260 , other 0.130 seconds)16:25
*** arcean has quit IRC16:25
luffreemangordon: so it's quite same for decoding?16:28
freemangordonyep16:28
DocScrutinizer05hey gregoa, long time no see! :-D16:28
freemangordonluf: going to try to backport linaro patch16:29
gregoaDocScrutinizer05: long time no packaging question :) besides that I'm always here16:30
DocScrutinizer05:-D16:30
DocScrutinizer05gregoa: you ever had a discerning look at the optional/alternative "specs" for CSSU? http://maemo.merlin1991.at/cssu/meetings/2012-05-14.txt16:33
jon_yugh, fcam should really load by uname -a16:34
DocScrutinizer05we could use any comments and help available on that16:34
gregoaDocScrutinizer05: not really. I vaguely remember the meeting but haven't thought about it further (and forget the details in the meantime)16:34
gregoaDocScrutinizer05: ok, tab open in my browser for reading later16:35
DocScrutinizer05many thanks, much appreciated16:35
*** int_ua has quit IRC16:59
*** arcean_ is now known as arcean17:00
luffreemangordon: ping17:08
lufHow to set the exact cpu speed for the CPU?17:09
*** BCMM has quit IRC17:12
lufI get it.17:14
freemangordonkernel-config lock 50017:15
luffreemangordon: I have better results with comparing ARM versus NEON libpng ...17:26
freemangordonluf: link?17:27
lufARM is in all test faster then the fastest NEON :D17:27
freemangordon:D:D:D17:27
lufI re-run it ten times for both.17:27
lufSo ARM is around 5-20% faster againist NEON :D :D :D17:29
lufI don't want to backport the linaro part ...17:30
lufI don't think it'll be faster.17:30
lufOMG why I think that developers are testing their code before thay say it's faster ...17:31
freemangordonluf: I am packporting it17:31
freemangordonbut it seems broken :(17:31
luf:D :D17:31
freemangordonactually I backported it17:32
freemangordonbut pngtest fails :D:D:D17:32
lufDo you have some benchmark for zlib? I want to test zlib-1.2.3 versus zlib-1.2.717:32
freemangordonno17:32
ShadowJKmru said 6X before anyone had written any code, let alone tested it ;)17:33
DocScrutinizer05coding is voodoo17:35
DocScrutinizer05well, estimations on coding are, a lot17:35
ShadowJKvectorizing stuff is even more voodoo17:36
DocScrutinizer05you never know what the final code will look like, or perform. Until it's finished and tested17:36
ShadowJKyou take the mathematical algorithm, turn it inside out in 6 Dimensional space, then write the neon code for it17:36
DocScrutinizer05and more often than not coders implement a 'smart' algo that looks great but does nasty things (e.g. re-instantiating objects in a loop)17:37
DocScrutinizer05and generally overhead for setup of any runtime environment is neglected17:38
DocScrutinizer05see glamo disaster. This critter *could* do pretty fast gfx operations, but only on max 512*512, and you need to pump all the data from CPU thru a terribly slow hw if to the glamo chip, and possibly results back via same way17:43
DocScrutinizer05the hw results of genuine coder thinking17:44
freemangordonwhat worries me is that upstreamed neon code does not wokr :(17:44
freemangordon*work17:44
luffreemangordon: I can write even faster libpng with failed tests (just function which returns error) :D17:44
DocScrutinizer05for NEON et al be prepared to face similar minor inconveniences17:44
freemangordonI know17:44
*** dhbiker has quit IRC17:45
freemangordonDocScrutinizer05: not sure, that is suposed to result in the same bits17:45
DocScrutinizer05I know on several processors it's not worth using the math coprocessor since setting up stuff there is slower than doing the same thing in software on the main processor, even in 16bit17:47
freemangordonaah, I see now, it is my fault17:47
freemangordonI need the memory aligned on 16byte boundary17:47
lufYes .There is a #define for it ...17:48
freemangordonyep, i saw it17:48
freemangordonlets see if i can backport that17:48
lufIt's quite easy.17:49
lufJust it reduced the buffer from 64 to 48 bit (as I understand)17:49
DocScrutinizer05how is alignment reducing buffer sizes?17:51
lufDocScrutinizer05: alignment is also done by another thing. But in that part code is also used smaller buffer (in newer libpng).17:52
freemangordonluf: i don;t think it is that easy17:52
freemangordonif I read the code correct, we have 64 byte aligment in 1.217:53
jon_ywelp, fcam drivers can kill a system boot17:53
jon_yswitched back to stock kp for now17:53
jon_yapparently loaded at boot17:53
jon_yI'll need to also find out if uboot loads the previous kernel for act-dead17:55
DocScrutinizer05jon_y: sure, camera-ui gets preloaded17:55
DocScrutinizer05by dsme17:55
jon_ydsme?17:55
DocScrutinizer05device state management entity, or whatever17:55
DocScrutinizer05dsmetool --help17:55
DocScrutinizer05ps|grep camera17:56
jon_ysomehow fbcon showed up even after selecting stock kp52 for boot17:56
jon_ycurse you incompatible symbols17:56
DocScrutinizer05the joy of multiboot systems: each kernel needs its own modules17:57
jon_ythe dumb thing about fcam is that it is not using uname -a17:57
DocScrutinizer05#1 bootloop reason, by far17:57
jon_yit would be all cool if it did that17:57
jon_yit only had "vanilla" and "power"17:58
jon_ydoesn't even say which kp version, but it is really for kp52 as now it is working again17:58
jon_yDocScrutinizer05: you see what I'm getting at?17:59
*** dhbiker has joined #maemo-ssu17:59
keriojon_y: so it just forcefully loads them? :s18:00
jon_yDocScrutinizer05: also, is the formatting OK? http://paste.debian.net/220113/18:00
kerioalso, uboot doesn't have any special coding for ACT_DEAD i think18:00
jon_ykerio: it would be easier to keep multiple kernels if it used uname18:00
jon_yright now it only has "vanilla" and "power"18:01
kerioyeah but the module for the "power" kernel can't be cleanly loaded on kp52 if it was compiled for kp5118:01
keriothe version string is different18:01
kerioso it's loaded forcefully18:01
jon_yyes, this is the stupid part18:01
jon_yjust use uname to keep them separate18:02
kerioyou'd have to ship a shitton of modules18:02
jon_ykerio: just have the modules ship as part of the kernel or something18:02
jon_ylike what kp does with wireless inject drivers18:03
keriothey were included with the fifty-first release of kernel-power18:03
kerionot before18:03
jon_ywhich is a good step18:03
jon_yon uboot, somehow the previous kernel was still loaded after selecting the boot entry18:03
jon_yfbcon!18:04
keriowat18:04
jon_yno idea what was going on18:04
keriohm, i wonder if it actually *does* have a way to handle ACT_DEAD mode in a sensible way18:04
jon_ykp52 does not have fbcon loaded by default, but it was showing all the console boot up messages18:04
jon_ymy previous selection had fbcon builtin18:05
jon_yso no module loading there either18:05
keriojon_y: didn't you add fbcon to /etc/modules?18:05
jon_yno, I removed them18:05
kerioare you suuure?18:05
jon_yyes, I couldn't get fbcon to show up reliably anyway18:06
jon_ywhen it does show, it is a mystery18:06
*** nox- has joined #maemo-ssu18:12
DocScrutinizer05jon_y: some formatting oopsies in that pastebin bq27k-detail remake18:24
DocScrutinizer05e.g for temperature in °C18:24
DocScrutinizer05has two decimal separators18:24
jon_yyeah, I haven't had the time to test it on the n900 yet18:25
jon_yDocScrutinizer05: ok, fixed the dot18:26
DocScrutinizer05if you wanna make sure it's identical, just use a i2cdump >file and use that file as input, then do a diff on output from bash bq27k-detail and your version18:26
jon_yok18:26
luffreemangordon: I'm using zpipe for the zlib benchmark and using /dev/(u)random isn't very good as it generate the numbers so it use quiet a lot of CPU.18:43
freemangordonluf: use a lile18:44
freemangordon*file18:44
lufSure I'm using ...18:44
freemangordonluf: backporting aligned memory in libpng is not trivial18:44
lufIt's just and idea I see so I want to share ...18:44
freemangordonit uses a new png_struct_def member ;)18:45
luffreemangordon: ok I don't think it helps so much ... I'll see with zlib. But I'm affraid it will be another loose of my time.18:45
lufAnd also it seems it's not good to do compress/uncompress at once as compress use around 90% of cpu and uncompress just below 5% ...18:47
jon_yDocScrutinizer05: did the diff, fixed most of them18:49
jon_ynow differences are mostly whitespace18:49
jon_yDocScrutinizer05: perl version http://paste.debian.net/220129/18:56
lufzlibs 1.2.3 and 1.2.7 seems to have same behaviour (times).19:01
freemangordonluf: upstream results for libpng, gcc 4.2.1:19:13
freemangordonCPU time used = 52.020 seconds (decoding 1.990,19:13
freemangordonencoding 49.910 , other 0.120 seconds)19:13
freemangordonthe same benchmark as for the previous runs19:14
freemangordonabout 1 second down, i'd say this one is ok19:14
lufThis is faster?19:23
lufAre you just count deconding?19:24
freemangordonstock is 2.840 this one 1.990. decoding only19:24
lufHmmm but the question is if neon helps or just the align or whatever changes you did.19:25
freemangordonaligment on 16 bytes should not make change for ARM19:26
lufDid you checked my previous work that it has no impact while this have so big impact?19:26
freemangordonhmm, no19:26
lufMaybe I made some mistake in Makefile ...19:27
luffreemangordon: can you check it?19:35
freemangordonI will19:35
lufAgain with zlib I have slower result with neon asm code.19:36
luffreemangordon: is your code somewhere accessible so I can check it too?19:43
freemangordonno. I am using an ugli hack for backport, so will try to fix "your" patch to achieve the same speed19:45
freemangordon*ugly19:45
luffreemangordon: https://gitorious.org/0xdroid/external_zlib/commit/5d42d250693f443ba3ad0f663d269177b2d02300/diffs - this slowdown the zlib reading from aprox. 0.57 to 0.61; I have to do something wrong.19:47
freemangordonluf: ARM memcpy is faster than NEON memcpy19:48
freemangordonluf: I think i see whay "your" libpng patch is slow, it uses 32bits, not 12819:49
freemangordon*why19:49
luffreemangordon: is it general or just for N900?19:50
DocScrutinizer05ha, native datatypes are kinda important, yes19:51
freemangordonluf: the 4bpp code path19:51
freemangordonupstream is using 128 bit SIMD, "your" loops 4x32bits for the same data19:52
luffreemangordon: :D19:52
freemangordonno matter it is using NEON in that loop19:52
DocScrutinizer05a 686 is slower than a 386 when running in 16bit mode (made up example)19:52
freemangordonDocScrutinizer05: well, not the same, but still19:53
luffreemangordon: as I wrote I don't know asm ... it's too low for me :)19:53
lufI understand 32-bit versus 128-bit ...19:53
freemangordonluf: I know, I am explaining what is going on in higher-level language19:53
freemangordonso, instead 4 pixels per instruction we are doing 119:54
freemangordonluf: but this is in 3/4 bpp paths19:54
freemangordonI expext for >4 bpp we'll have better results19:54
freemangordonI need .png to test with19:54
freemangordonanyone?19:55
luffreemangordon: don't you have one as you tested it with upstream backport?19:55
freemangordonit falls in 4bpp path19:56
freemangordonwhere bpp is BYTES per pixel, don;t ask me why19:56
freemangordonhmm, you know what. I think I can make that much better ;)19:58
freemangordonI can call C function until the remaning bytes align to 1620:00
freemangordonthen use neon untill there are < 16 bytes to process20:00
freemangordonthen again C function20:01
luf:D20:02
lufYou can also ask user for some help (with some dialog) :D20:02
DocScrutinizer05freemangordon: well, that's how any simple silly memcpy and similar stuff works20:04
freemangordonDocScrutinizer05: I know20:04
DocScrutinizer05so it's probably a proven concept20:04
freemangordonluf: sorry to tell you, but "your" patch is useless :(. just compare path code with upstream. NEON has paeth insctruction ;)20:06
DocScrutinizer05just amazing "upstream" didn't know of those standard design patterns20:06
freemangordonDocScrutinizer05: they did aligned memory20:06
DocScrutinizer05aah20:06
freemangordonwhich is better20:06
freemangordonbut I cannot backport it cleanly, as there is ABI change for that20:07
freemangordonadditional member in one of the structures20:07
DocScrutinizer05:-/20:07
DocScrutinizer05understood20:08
freemangordoni did some hackish tricks in memory allocation routines, just to check if is makes sense to use NEON20:09
freemangordonbut I wont push that in CSSU :P20:09
* DocScrutinizer05 meanwhile thinks freemangordon has changed his mind towards a more conservative notion regarding CSSU changes, and appreciates that change very much20:13
freemangordonDocScrutinizer05: wrong :P20:14
freemangordonIt's been always the case20:14
luffinally adler is quiet faster ... aprox 0.56 -> 0.54 (so 5% faster)20:17
freemangordonadler?20:17
lufhttps://build.pub.meego.com/package/view_file?file=zlib-1.2.7-adler32_vec_kaffeemonster.patch&package=zlib&project=home%3Astskeeps%3Acheck-zlib&rev=c0999fc2e13c75f801849437bed4f9ea20:17
freemangordonluf: aah, you find a patch that actually works?20:18
luffreemangordon: yes :D20:19
luffreemangordon: so if I understand well no patch for libpng (even it's 30% faster)20:24
freemangordonluf: there will be20:24
freemangordonbut I need time20:24
lufok.20:25
*** NIN101 has quit IRC20:27
*** NIN102 has joined #maemo-ssu20:28
*** luf has quit IRC20:56
*** NIN102 is now known as NIN10120:59
merlin1991freemangordon: is your sb x86 or x64?21:09
*** bsdmaniak has joined #maemo-ssu22:02
gregoaDocScrutinizer05: I had a look at the log from the May meeting about the optional stuff. technically speaking, I think merlin1991's sketched proposal should work, and I agree that some extra field+HAM-specific changes are a bad idea.22:20
gregoaDocScrutinizer05: in general, I'm not so sure I'm excited about the idea. it seems a bit over-engineered and complicated and adds another layer of complexity22:20
gregoaDocScrutinizer05: my impression nowadays is that maemo has more repositories than developers (exaggerating, but 3x extra + 3x devel + defunct nokia repos), and I'd rather see them merged / reduced to a lower number22:21
gregoaDocScrutinizer05: I'm also not so sure we need to so "shy" of shipping new versions that will overwrite existing packages. after all, that's what all linux distros do; a maintainer decides and users get the updated packages without any decision about having 3 alternatives22:23
keriogregoa: i think the issue is that cssu wants to avoid the "all or nothing" situation as much as possible22:24
gregoakerio: right, that's my impression too. and I have some doubts about that strategy. both from the concept (although I accept it) and from reality (current + future manageability)22:25
kerioDocScrutinizer05: you heard the man - upgrade *all* the things! _ò/22:26
ShadowJKOriginally Mer (Maemo reconstructed) went the route of reconstructing maemo into more normal linux22:29
ShadowJKin the process it broke everything and eventually when it's today working again, it pretty much needs 4X the hardware resources of N900 :/22:29
kerioShadowJK: does it run fine on four n900s?22:32
ShadowJKOne for calls, one for sms, one for music, one for homescreen22:32
ShadowJKyeah maybe22:32
keriobrilliant22:33
kerioisn't that what those weird modular proof-of-concept phones do?22:33
DocScrutinizer05gregoa: I seem to find a magic threefold of repositories in almost every arbitrary distro around22:41
DocScrutinizer05gregoa: 3 * devel?22:42
gregoaDocScrutinizer05: sorry, s/devel/cssu/ (stable, testing, devel)22:42
DocScrutinizer05cssu-devel is no official repo by any means22:42
gregoaDocScrutinizer05: right, the tree-step is fine, but we have at least 2x 3 repos -- the length of the sources.list on my n900 is impressive :)22:43
DocScrutinizer05cssu-stable and cssu-testing are complementary22:43
keriogregoa: meh, i have 7 repos22:43
DocScrutinizer05gregoa: then you're doing sth wrong, since you're not supposed to have extras, extras-testinf, and extras-devel acive all 3 concurrently22:44
kerioyou're not supposed to have any one of those active at all, you're supposed to use extras-devel-light!22:44
DocScrutinizer05and while CSSU-T/S is kinda like the update-repo usually found on every distro, the extras packages are more like packman22:45
kerioDocScrutinizer05: nokia ssu&cssu is one repo divided in two22:45
kerionokia apps&maemo extras(-whatevs) is one repo22:45
DocScrutinizer05and btw cssu optianla/alternative packages won't even introduce a new repo22:46
kerioand those could be merged, anyway22:46
gregoaDocScrutinizer05: right; I have 2 of the active, and 1 would be enough. maybe I was confused by the sheer amount :) but still those exist, and I'm just wondering if this huge infrastructure is really the best way to go22:46
keriogregoa: debian has stable/testing/unstable/experimental22:46
DocScrutinizer05kerio: nokia ssu repo is a hoax22:46
kerioDocScrutinizer05: meh, it has its uses i think22:46
keriofor dependencies, at least22:46
kerionokia apps is definetely used for dependencies22:47
gregoaDocScrutinizer05: I understand that the optional idea wouldn't introduce new repos; it's just that looking at the log reminded me of these thoughts I had earlier22:47
gregoakerio: oh really? :) but in debian that's one "tree" not like 3 as in maemo (nokia / extras / cssu)22:48
kerio...i think, at least22:48
*** bsdmaniak has quit IRC22:48
keriogregoa: yeah, that situation is a bit stupid22:48
keriolet's actually change situation22:48
keriolinux mint22:48
DocScrutinizer05the point of CSSU is: we can't ship that stuff via nokia repos, though they would belong there22:49
gregoabtw, the split between extras-* and cssu-* has the negative side effect that I can't build/upload a package for/to extras* that (build)depends on something in cssu (new qt e.g.)22:49
keriomint, ubuntu, ubuntu-updates, ubuntu-security, ubuntu partner22:50
DocScrutinizer05otoh we don't want to get completely rid of nokia repos since that would introduce *real* unmanageable overhead in maintenance22:50
kerioDocScrutinizer05: meh, a mirror wouldn't cause that22:50
kerioi'm not sure it would be legal, though22:50
gregoaDocScrutinizer05: right. maybe this whole changes now can help to consolidate the old nokia repos and the cssu repos. would kill one "tree". and then the question is if we should "force" all users to cssu ...22:50
merlin1991gregoa: nah22:50
merlin1991cssu in no way is as "functional" as a stock image22:50
keriogregoa: <DocScrutinizer05, on TMO> CSSU stable is now recommended for all N900 users22:51
merlin1991we introduce regressions with every step22:51
DocScrutinizer05again: cssu is not the alternative repo for extras22:51
merlin1991DocScrutinizer05: cssu technically should live together with extras in one repo22:51
gregoamerlin1991: easy solution: don't introduce regressions :)22:51
kerioDocScrutinizer05: normal distros only have one repo, is the point that gregoa is making, aiui22:51
merlin1991gregoa: easier said than done with all those fancy closed source foobar applications (c) nokia that don't work like everything else has to22:52
gregoamerlin1991: I mean, that's what all linux distros do: upload new stuff to unstable or rawhide or whatever. yes, things can break there and have regressions. bugs get fixed, and packages migrate to testing22:52
DocScrutinizer05kerio: ORLY?22:52
DocScrutinizer05shall I send you a list of repos I got enabled on a pretty default OpenSuse?22:52
merlin1991debian has 1 repo22:52
keriothat could mean that opensuse is a bad distro22:52
* kerio flees22:52
merlin1991+ the backports repo22:53
kerio+ the security repo22:53
merlin1991kerio: I don't really count the main repo as independend of the scurity repo because they basically ship the same :D22:53
merlin1991only + security patches22:54
tadzik+ multimedia repo22:54
gregoanah, that's _not_ debian22:54
merlin1991which multimedia repo?22:54
keriotadzik: why?22:54
merlin1991o_O22:54
tadzikI have an deb  http://www.deb-multimedia.org/  testing main non-free22:54
keriohe means medibuntu, probably, but that's been discontinued or at least not encouraged anymore22:54
tadziktbh, I'm not sure why it's there ;)22:54
tadziknah, that's plain debian testing22:54
merlin1991wtf is that repo22:55
gregoathat's a private repo by christian marillat. not more not less.22:55
gregoa(and partly useful)22:55
tadzikmust be that I needed something from it22:55
tadzikbut yeah, that's not likely upstream debian repo22:55
DocScrutinizer05FWIW http://wstaw.org/m/2012/12/29/yast2-000.png22:57
merlin1991madness22:58
DocScrutinizer05I don't think many finegrained repos are sth negative, au contraire22:58
kerioDocScrutinizer05: that's not fair, a ton of those would be merged in the same repo in apt22:58
kerioin different components22:59
DocScrutinizer05mad gigantic one-for-everything repos like extras-devel *are* something negative22:59
keriodebian has free non-free contrib, ubuntu has main restricted universe multiverse22:59
merlin1991dafaq source has it's own repo22:59
merlin1991kerio: don't forget source22:59
kerioif you count those as separate repos, the count becomes huge22:59
kerionobody actually uses source repos23:00
kerio:P23:00
merlin1991fu ;-)23:00
keriosource is separate in debian repos though23:00
*** arcean has quit IRC23:00
kerioa deb-src entry instead of a deb entry in sources.list23:00
merlin1991from the repository side it has it's own index files (one per component)23:01
merlin1991funny to abstract the "source" away at the same level as different architectures23:02
kerioit makes sense, really23:02
merlin1991only if you can build each and every arch from the same source package23:03
keriowhy shouldn't you be able to?23:03
keriofile a bug!23:03
merlin1991special patches needed here and there for foobar things in various cpus?23:03
keriothe source tarball should not be architecture-specific23:04
merlin1991I hate it when a package does all kinds of crazy for different architectures in order to build properly23:04
merlin1991you never find the part that builds what you want :D23:04
kerioheh, have you ever looked at the gmp source?23:04
DocScrutinizer05merlin1991: well, that's the linux way however23:04
merlin1991well there's a difference between calling ./configure with arch flags and doing all kinds of tricker at a higher level23:05
kerio"higher level"?23:05
merlin1991above the plain building23:06
DocScrutinizer05even higher than .configure?23:06
merlin1991yep23:06
kerioautoconf level?23:06
merlin1991I've seen things ;)23:06
DocScrutinizer05couldn't figure any further abstraction level23:06
keriowhy didn't you file them as bugs?23:06
merlin1991still got bad dreams23:06
keriomerlin1991: ever compiled nethack without the autoconf patch?23:06
merlin1991kerio: because sometimes you just want to get something running and never think about it again23:07
DocScrutinizer05next higher abstraction level would be actual separate repos for each arch23:07
*** dhbiker has quit IRC23:07
DocScrutinizer05linux main branch per definition should build on all arch23:08
DocScrutinizer05aiui23:09
merlin1991DocScrutinizer05: I'm talking about abstraction in the build process, you usually have debian/rules - debhelper / cdbs / custom - autotools / cmake / whatever23:09
merlin1991i hate it when the arch hacks live @ debian/rules level23:09
keriogmp does specific multiarch in the right way23:09
DocScrutinizer05whatever, maemo isn't multi-arch right now23:10
DocScrutinizer05and probably won't ever be23:10
DocScrutinizer05also it's not a child of main23:10
merlin1991it's hard enough to follow those with all the implzit things that happen (imo cdbs is worse than debhelper because it does even more without being asked specifically)23:10
merlin1991or rather asked visibly, because yeah you import some make file and it does stuff for you, but you don't see from the import what will happen23:11
gregoamerlin1991: be happy if you never saw a package using yada as its build system :) (I hope they are all gone from debian by now)23:11
merlin1991gregoa: do you also find it funny when you find a package that does most things in the rules file itself23:12
gregoamerlin1991: well, it can cause headaches, but sometimes it might be necessary23:12
merlin1991there are a few packages in maemo extras written like that23:13
merlin1991it feels so weird to look at their buildscripts after messing with debhelper all the time23:13
gregoamerlin1991: or do you mean doing things manually instead of using dh_*? yes, that's a spleen of some people but annyoing for alle others23:13
merlin1991yep I mean that23:13
gregoawell, a problem in maeno in general is from my POV that most people are more experienced in programming than in packaging, they just have to press their code into a .deb somehow, and that's how the packages look like23:14
gregoaincluding packages that don't have source code in the "source package" but compiled stuff that they "cp" in d/rules23:15
merlin1991but hey I'm one of them thanks to aegis in harm I wrote this hack: https://github.com/kelvan/gotoVienna/blob/master/bdist_hdeb.py#L4323:15
keriogregoa: isn't that the reason why maintainers are usually separate people from the developers?23:15
gregoaand cargo-culted d/rules files that look like from 200523:15
gregoakerio: ack, at least that's how linux distros work23:16
gregoaI mean, having debhelper 5 (and a separate debhelper7) version and dpkg from 2009 etc. also doesn't help23:16
keriogregoa: and no debconf!23:16
gregoaii  debconf                 1.4.70.osso3+0m5        Debian configuration management system23:17
kerios/no/no interfaces for/23:17
infobotkerio meant: gregoa: and no interfaces for debconf!23:17
kerioso no package uses it23:17
gregoaoh yes. - I guess I have some package installed that needs/uses debconf :)23:17
kerionot installed here23:18
keriofakedebconf is installed23:18
keriooh god why23:18
merlin1991probably to get debconf stuff run through ham?23:19
kerioit doesn't actually run through, though23:19
gregoahm, not sure why I have it. maybe I tried to port something from debian at some point.23:19
merlin1991kerio: yeah well fake it through so dpkg is happy23:20
merlin1991oh yes the user said DO it to everything, trust me!23:20
kerioyou might think he's happy, but he's crying on the inside23:20
keriolonging for an end of suffering that will never come23:20
merlin1991but hey we have a ham specific ask dialog23:21
merlin1991that is very helpful when you install something via ssh and the device is somewhere else23:21
kerioindeed23:21
gregoaoh, yes23:21
kerioit's especially annoying when the phone is in your room and you're in the bathroom, pooping23:21
merlin1991kerio: actually in that case I'd rather take the phone with me instead of ie a laptop :D23:22
gregoait's also annoying when it's on your desk, and you just don't look at it :)23:22
merlin1991after flashing I usually install extra decoders support23:22
keriogregoa: most recent preinst/postinst scripts that do that usually print "Confirm the information on the n900 screen" on the terminal23:22
keriomerlin1991: hold on, then why would i have a laptop?23:23
merlin1991and the fsckd flac/ogg package has a "this might take a while" dialog23:23
gregoakerio: yup, that's a bit less insane23:23
keriomerlin1991: i wonder if you can ^c maemo-confirm-text23:23
merlin1991nah you'd ^c dpkg / apt in that case23:23
merlin1991because there are layers at work and you only have control over the top layer :/23:23
keriook, killall maemo-confirm-text from a separate shell23:23
keriowhat's the return code of a sigkilled program?23:23
merlin1991well that would give it a non0 exit status23:24
merlin1991thus making dpkg abor23:24
merlin1991t23:24
gregoain a loop! hey, we need a new dbus interface and hal could care for it !!11123:24
gregoawith settings stored in a looong gconf key23:24
keriogregoa: plz add debconf support to the whole of maemo23:24
gregoa:)23:25
gregoakerio: works for me. I can run "dpkg-reconfigure debconf" and answer it's debconf questions. and in /var/lib/dpkg/info I see some maintainer scripts using debconf23:26
keriogregoa: only upstream packages though23:27
gregoakerio: you mean "properly done packages"? :)23:29
*** arcean has joined #maemo-ssu23:29
keriogregoa: i can't run dpkg-reconfigure, weird23:29
gregoakerio: https://paste.debian.net/220183/23:29
gregoakerio: well: which dpkg-reconfigure ==> /usr/local/sbin/dpkg-reconfigure *cough*23:30
kerioand what's that?23:30
gregoa/usr/local/sbin is managed by the local admin23:30
kerioi know23:31
gregoai.e. I've copied it there from a debian machine manually, I guess23:31
kerioyou disgust me23:31
gregoas/you/maemo/, I suppose :)23:32
kerioyep :(23:32
keriohm, do you also get error message about debconf eventually falling back to the Teletype frontend?23:32
kerioor maybe you have a proper perl install23:32
kerioconsidering your fetish for badly-designed languages that are indistinguishable from line noise23:33
gregoakerio: that might be libterm-readline-perl-perl (usually I prefer libterm-readline-gnu-perl, not sure why I have the other one on the n900)23:34
keriogregoa: it would still try and fail to use Dialog23:35
keriowhich is a suggested package but it's not in the repos23:35
gregoaif you dpkg-reconfigure debconf you can chosse between dialog, whatever, readline23:35
gregoa*choose23:35
kerioi can't dpkg-reconfigure :(23:36
gregoawhat happens? have you tried "dpkg-reconfigure -plow debconf"?23:40
kerioi don't have a dpkg-reconfigure23:40
kerioit's nowhere in the repos23:40
kerio...what the hell is -plow anyway23:40
gregoaoh, I thought you copied it in the meantime :)23:41
gregoapriority low, as in "show me all questions"23:41
keriooh23:41
keriohttp://static.tvfanatic.com/images/gallery/mr-plow-picture.jpg23:41
kerioto be fair i do have a sheevaplug with debian23:41
gregoakerio: http://paste.debian.net/220187/ (the version I'm using on the n900, a bit older than what I have on unstable)23:42

Generated by irclog2html.py 2.15.1 by Marius Gedminas - find it at mg.pov.lt!