IRC log of #maemo-ssu for Monday, 2011-11-28

*** vegai has quit IRC00:05
Sicelojust finished updating to Stable 1 :)00:11
merlin1991:)00:16
ThreeMhmm, mailapp crashed....00:21
*** Sc0rpius has quit IRC00:21
ThreeMis there any log for that? i would like so see whats the reason for that00:22
freemangordonYeah, install -dbg packages for modest tinymai libglib, attach gdb to it and wait for crash00:24
ThreeMthx00:25
andre__so what was agreed on? have two products for Maemo CSSU (S vs T) in bugzilla, or just adding two versions (S vs T) to the one existing product?00:30
andre__I'd really prefer the latter I think00:30
andre__and sorry that I haven't done this already :-/00:30
*** m0use has quit IRC00:41
*** freemangordon has quit IRC00:41
*** povbot has joined #maemo-ssu00:45
merlin1991we didn't really decide but I'm for the latter00:47
*** Sc0rpius has joined #maemo-ssu00:49
freemangordondamn, with modest and tinymail build on my machine i am not able to reproduce the crashes00:50
freemangordonThreeM ping00:50
freemangordonmerlin1991 ^^^00:51
*** thingie has joined #maemo-ssu00:52
DocScrutinizerwhen S and T are versions, what's maemo1 aka 17 then?00:54
*** ruskie has joined #maemo-ssu00:55
merlin1991I talked to mag earlier, i'll try to convince him to do a rebuild when I get a hold of him again00:58
merlin1991freemangordon: ^^00:58
freemangordonno,don't. I've just upload .debs and will ask the user who reported the issue to try those. If the issue is fixed i prefer my build to be used.01:00
DocScrutinizerwell, it seems there's no proper versioning for any of the components, so even while version AIUI is meant to be the version of the component, it's probably irrelevant for what we actually abuse it.01:04
*** andre__ has quit IRC01:05
freemangordonIt is not the version as such, it is correct. Lets see if modest build by me works, will investigate what has happened after that.01:06
*** jonwil has joined #maemo-ssu01:20
*** freemangordon has left #maemo-ssu01:20
*** scoobertron has quit IRC01:58
*** mirandir has quit IRC02:43
*** NIN101 has quit IRC03:24
*** nox- has quit IRC03:33
*** M4rtinK has quit IRC04:08
* merlin1991 just checked twitter and googled a bit for cssu04:17
merlin1991so much fud out there, it's unbeliveable04:17
*** amiconn has quit IRC05:18
*** amiconn_ has joined #maemo-ssu05:18
*** amiconn_ is now known as amiconn05:18
*** jonwil has quit IRC05:28
DocScrutinizerHAH05:30
DocScrutinizeryeah, esp since everybody seems to like thinking of CSSU as an alternative OS to maemo05:31
DocScrutinizerno, guys! CSSU is even less of a "fork" than packman05:32
DocScrutinizer"will you bring duke nukem 4ever to CSSU?" PFFFF05:33
DocScrutinizerall the "I can't find $RANDOM_ACME_APP for maemo, may I suggest it for CSSU" blubber05:35
*** psycho_oreos has joined #maemo-ssu06:32
*** zogg_ has quit IRC06:47
*** zogg_ has joined #maemo-ssu07:00
*** DocScrutinizer has quit IRC07:22
*** DocScrutinizer has joined #maemo-ssu07:22
*** mirandir has joined #maemo-ssu08:47
*** mirandir has left #maemo-ssu08:57
*** mirandir has joined #maemo-ssu08:59
*** neal has quit IRC09:35
*** andre__ has joined #maemo-ssu09:54
*** andre__ has joined #maemo-ssu09:54
*** scoobertron has joined #maemo-ssu10:12
*** freemangordon_ has joined #maemo-ssu10:19
*** mece has joined #maemo-ssu10:21
mececongratulations on stable release!10:21
freemangordon_MohammadAG, ping10:26
freemangordon_X-Fade ping10:38
*** m0use has quit IRC10:39
*** amiconn has quit IRC11:06
*** amiconn_ has joined #maemo-ssu11:06
*** amiconn_ is now known as amiconn11:06
MohammadAGfreemangordon_: Pong11:28
freemangordon_MohammadAG: saw my yesterday's coomment re modest and tinymail?11:31
*** jonwil has joined #maemo-ssu11:37
* andre__ wants MohammadAG's last word on https://bugs.maemo.org/show_bug.cgi?id=1249711:46
povbotBug 12497: Maemo 5 Community SSU needs field to distinguish between stable and testing11:46
*** scoobertron has quit IRC12:19
MohammadAGfreemangordon_: merlin1991 told me, its a clean build12:34
MohammadAGI built tinymail, installed it then built modest against it12:34
freemangordon_Then why is those reports? And why I am unable to crash modest with my build?12:34
freemangordon_Did you make git clone and build from scratch?12:35
MohammadAGandre__: Latter it is12:36
MohammadAGfreemangordon_: No idea12:36
MohammadAGYea12:36
MohammadAGIll rebuild it when i get home then12:36
MohammadAGBut it doesn't crash for me12:37
freemangordon_Do you have gmail account set up on your n900?12:37
DocScrutinizerfreemangordon_: didn't you find misalignment?12:38
freemangordon_Yes12:38
freemangordon_I mean there is from the gdb backtrace12:38
DocScrutinizeris it possible that misalignment gets handled poorly by the CPU?12:39
freemangordon_Of course, I don't think ARM CPU is able to execute instruction not aligned on DWORD boundry12:39
DocScrutinizerand is it possible that allignemnt at alrge changed when you built on your machine?12:39
freemangordon_And even if it can, the function pointer which is passed is incorrect, it points to start of callback function + one byte12:40
freemangordon_which is nonsence12:40
DocScrutinizerdo you think there's a compiler option or source metacommand to fix allignment?12:41
freemangordon_for code section? don't think so. anyway, I just made git clone of tinymail, copied debian/control.maemo-fremantle to debian/control and debian/rules.maemo to debian/rules and dpkg-buildpackage12:43
freemangordon_installed produced -dev packages and rebuild modes12:44
freemangordon_that's it12:44
freemangordon_MohammadAG: did you make a build from scratch or just pull the commits from gitorious?12:45
MohammadAGfreemangordon_: Pull the commits, its one of the reasons we use git12:48
DocScrutinizerseems kernel can be made to either or not catch allignment errors in sourcecode12:49
DocScrutinizerif it doesn't catch, it segfaults12:49
freemangordon_sure, but it is always a best practice to build a release using fresh source code copy.12:49
freemangordon_doc, it is not segfault, it is sigill12:50
freemangordon_this is generated by CPU12:50
freemangordon_and seems reasonable when we try to execute in the middle of the instruction12:51
DocScrutinizerecho 5 >/proc/cpu/alignment12:51
freemangordon_why is that?12:51
DocScrutinizersure it's best fixed in source resp at build time (compiler option)12:52
freemangordon_no,no , it has nothing to do with compiler options, they are ok. i suspect some object code remnants from previous builds12:53
DocScrutinizerhttp://bugs.debian.org/cgi-bin/bugreport.cgi?bug=64704912:53
DocScrutinizerI suspect that former builds simply had more lucky alignment, due to a few bytes difference - like your local build12:54
freemangordon_I don't believe that a function ponter could be misaligned12:55
freemangordon_#3  0x4164f466 in modest_mail_operation_get_msg (self=0x3410b8, header=0x30bec0, progress_feedback=<value optimized out>, user_callback=0x4167a999 <view_msg_cb+1>,    user_data=0x33d930) at modest-mail-operation.c:284012:55
freemangordon_<view_msg_cb+1>12:56
freemangordon_which means view_msg_cb is pretty ok, it is the address passed as an argument which is incorrect12:56
freemangordon_I just refuse to blame the compiler here12:57
DocScrutinizerwell, then go to the sourcecode and find what else is causing the one-off12:58
DocScrutinizerthe statistical nature of this bug very much makes me believe it is related to alignment, maybe even at execution time12:58
DocScrutinizerwe had similar issues like 3 years ago in OM200812:59
DocScrutinizerwith a few pkgs12:59
DocScrutinizercan't recall how it got fixed12:59
freemangordon_what exactly is not clear so far? I think that MohammadAG had made a build based on remnants from previois builds. Without doing git clone first. And maybe without cleaning old builds remnants.13:00
freemangordon_Not that gcc is perfect13:01
freemangordon_MohammadAG: ^^^13:01
freemangordon_Is that true or not?13:01
DocScrutinizerwhat exactly *is* clear so far and not speculation?13:02
*** M4rtinK has joined #maemo-ssu13:02
freemangordon_the fact that build based on git clone seems forking so far13:02
freemangordon_*working13:03
DocScrutinizercouldn't modest miss some //#META:ALIGN statement in source sinde 2 years?13:03
DocScrutinizersince*13:03
freemangordon_have no clue13:03
DocScrutinizerI can't see how a .o that was ok now suddenly gets misaligned just because main or sth got recompiled13:04
freemangordon_we had such case with h-d once13:04
freemangordon_MohammadAG: remember? ^^^13:04
DocScrutinizerunless linker soesn't cope with alignment due to missing directive13:05
freemangordon_he pull the commits from gitorious and made dpkg-buildpackage -nc13:05
freemangordon_and shouted "h-d is not working"13:05
freemangordon_building from scratch shown that it *IS* working13:06
DocScrutinizerI'm playing devil's advocate once more and say h-d has same bug of missing linker directive13:06
*** NIN101 has joined #maemo-ssu13:07
freemangordon_and the winner is?13:07
DocScrutinizerhttp://bugs.debian.org/cgi-bin/bugreport.cgi?bug=64704913:07
freemangordon_it is different, there is a register based relative offset here, which easily could be caused by incorrect typecast.13:10
freemangordon_while the code in modest is "modest_mail_operation_get_msg (mail_op, info->header, TRUE, view_msg_cb, info->row_reference);"13:11
freemangordon_no relatives, just naked function name13:12
freemangordon_view_msg_cb13:12
freemangordon_is away for a while,bbl13:15
DocScrutinizergot some link to the sourcecode?13:16
*** lizardo has joined #maemo-ssu13:34
*** neal has joined #maemo-ssu13:37
MohammadAGfreemangordon_: I already said THIS is a clean build13:48
*** ekze-nyan is now known as ekze13:56
*** scoobertron has joined #maemo-ssu13:56
*** ekze-nyan has joined #maemo-ssu13:59
*** ekze has quit IRC14:00
*** ekze-nyan is now known as ekze14:00
*** psycho_oreos has quit IRC14:05
*** ekze-nyan has joined #maemo-ssu14:16
*** ekze-nyan has quit IRC14:17
*** ekze has quit IRC14:20
*** ekze has joined #maemo-ssu14:20
*** ekze-nyan has joined #maemo-ssu14:40
*** ekze has quit IRC14:42
*** ekze has joined #maemo-ssu14:50
*** ekze-nyan has quit IRC14:54
*** mece has quit IRC14:57
*** scoobertron has quit IRC15:02
freemangordon_DocScrutinizer: http://gitorious.org/community-ssu/modest/blobs/master/src/widgets/modest-msg-view-window.c#line197815:08
*** mase76 has joined #maemo-ssu15:36
*** mase_76 has joined #maemo-ssu15:42
freemangordon_MohammadAG: So, what's broken then. toolchain? headers?15:42
freemangordon_filesize of modest build by me is 911844 bytes while that in CSSU repo is 694744 bytes.15:47
Lava_CroftI bet this fuck up will make for a great story15:48
freemangordon_that is about 200k ?!?15:48
*** scoobertron has joined #maemo-ssu15:53
*** NIN101 has quit IRC15:55
*** xnt14 has quit IRC15:55
*** MohammadAG has quit IRC15:55
freemangordon_DocScrutinizer: ^^^16:13
*** scoobertron has quit IRC16:55
DocScrutinizerextremely suspicious16:58
*** NIN101 has joined #maemo-ssu16:59
DocScrutinizerif both are same strip-level then sth got severely screwed16:59
*** DocScrutinizer is now known as DocSalarySlave17:00
merlin1991freemangordon_: I'm starting my scratchbox, let's see what bytecount I can achieve17:00
*** scoobertron has joined #maemo-ssu17:01
*** scoobertron has quit IRC17:09
*** Kaptenen has quit IRC17:22
freemangordon_merlin1991: ok, try with both source from CSSU repo and with git clone17:24
*** Kaptenen has joined #maemo-ssu17:24
merlin1991did you check filseize of modest of tinymail?17:25
freemangordon_modest17:25
freemangordon_only17:25
Lava_Croft200k is ridiculous17:25
merlin1991and wich files from tinymail do I have to install on my scratchbox to compile modest properly?17:25
freemangordon_no,no17:25
freemangordon_wait17:25
freemangordon_do that only if you have scratchbox which is different from CSSU stable one, as you need to git clone tinymail, build it and install resultant -dev files17:26
freemangordon_merlin1991 ^^^17:27
merlin1991freemangordon_: no tinymail / modest in stable17:27
merlin1991I won't break anything for stable compiled17:28
merlin1991s/compiled/compiles/17:28
freemangordon_ok then17:28
freemangordon_git clone git://gitorious.org/community-ssu/tinymail.git17:28
freemangordon_after compilation is complete dpkg -i *-dev17:29
freemangordon_and do a modest build from git sources, not from CSSU tarball17:30
*** jonwil has quit IRC17:31
merlin1991freemangordon_: do I have to replace debian/control with debian/control.maemo-fremantle?17:36
freemangordon_for tinymail?17:36
freemangordon_merlin1991: ^^^17:36
merlin1991modest17:36
freemangordon_not sure, most probably not17:37
freemangordon_merlin1991: you need to do that for tinymail17:38
merlin1991I did it for tinymail17:38
freemangordon_ok17:38
freemangordon_merlin1991: wait, it seems modest is thumb compiled17:39
freemangordon_i am compiling it now too and see -mthumb in compilation swithches17:40
freemangordon_for some reason DEB_BUILD_OPTIONS=maemo-launcher,thumb,vfp17:43
freemangordon_most probably the same thing has happened to MohammadAG17:43
freemangordon_and that could explain 200k difference in binary17:43
merlin1991andre__: ping17:44
Lava_Croftthanks for making me learn something freemangordon_17:44
Lava_Croft'thumb' is my word of the day17:44
Lava_Croftnever knew about it17:44
andre__merlin1991, pong17:44
merlin1991I want to reopen bug 12306, wich status would I give it?17:45
povbotBug https://bugs.maemo.org/12306 Conflict problems for libqt4-test17:45
andre__merlin1991, UNCONF or NEW17:45
merlin1991I can only give it resolved, verified, closed or unconfirmed17:45
andre__there's nothing else anyway :P17:45
merlin1991but new isn't there17:45
andre__yes, because it was not NEW before17:45
merlin1991and imo new would be the right status17:45
andre__so first UNCONF, then you can change that to NEW17:45
merlin1991oh ffs17:46
freemangordon_I can bet we've been running thumb compiled modes all the time :D :D :D17:51
freemangordon_modest*17:51
Lava_Croftheh17:51
merlin1991freemangordon_: why is your compile not thumb compiled? to have better coredumps?17:52
merlin1991also how did you compile without thumb?17:52
freemangordon_because i have 4 cores ;)17:52
freemangordon_and usually do export DEB_BUILD_OPTIONS=parallel=4 before every build17:53
merlin1991ah :D17:53
merlin1991so the deb_build_options is sit somewhere in rules?17:53
freemangordon_at least that is what i guess17:53
merlin1991omfg bloody plasma shell went completely m00t here17:53
merlin1991won't even start dolphin now17:54
freemangordon_maybe configure script checks for target CPU capabilities, who knows17:54
Lava_Croftopen a term and type mc17:54
merlin1991mc does what?17:54
Lava_Croftstarts midnight commander17:54
Lava_Croftim being silly17:54
Lava_Croftsorry17:54
merlin1991hehe17:55
merlin1991btw thumb on arm comes from deeper layers of hell :D17:56
merlin1991thumb2 seems to be ok though as far as I've heard17:56
Lava_Crofti bet it servers a purpose:P17:56
Lava_Croftserves*17:56
freemangordon_anyway, there is an explanation what has happened17:56
merlin1991thumb fsckd things?17:57
merlin1991freemangordon_: modest ended up size 912532 here17:59
freemangordon_sure17:59
freemangordon_same here17:59
freemangordon_91184417:59
merlin1991hm why didn't it to thumb here then?18:00
freemangordon_have no clue18:00
freemangordon_did you open new scratchbox login?18:00
freemangordon_maybe some remnants from timymail build18:01
merlin1991I did not open a new scratchbox login18:03
freemangordon_merlin1991: BTW why is there configure script18:03
freemangordon_?18:03
merlin1991what do you mean?18:04
freemangordon_it should be created via autoconf scripts, don't have to be in source package18:05
merlin1991dpkg-buildpackage calls clean before creating the source afaik, I guess the configure script doesn't get removed and therefore ends up in the file18:07
freemangordon_yeah, but that means build scripts are fskd up18:09
freemangordon_as configure script (along with configure.timestamp and others) should not be distributed in the source package18:10
freemangordon_AIUI18:10
freemangordon_BTW DEB_BUILD_OPTIONS remains set when you break a compilation.18:11
merlin1991okay neither tinymail nor modest set DEB_BUILD_OPTIONS18:11
merlin1991hm modest from source tarball ended up 91250418:13
freemangordon_Have no idea why it was set here, i was compiled some hildon-input-method portrait fkb before18:14
freemangordon_modest18:14
merlin1991well I guess first step is to ask Mag if he has anything set in deb_build_options18:14
freemangordon_will you check again what is the size in CSSU repo, maybe i made a mistake18:14
merlin1991and if not have him compile it again to see if it's now different18:15
freemangordon_merlin1991: ^^^18:15
freemangordon_will you open modest .deb from CSSU and check again filesize18:15
DocSalarySlavemerlin1991: freemangordon_: I seem to recall I heard thumb is instable18:16
freemangordon_yes, there is your proof :D18:16
merlin1991modest deb from cssu is 69474418:17
freemangordon_yep18:17
freemangordon_it is deffinitely thumb compiled18:17
freemangordon_ok, i think we MUST disable thumb check in debian/rules18:17
merlin1991wait for your user if his crashes stop18:18
merlin1991if yes, then go and do that18:18
freemangordon_what is the point?18:18
merlin1991true18:18
* merlin1991 headdesks a lil rethinking his last statement18:19
freemangordon_it does not make sense to wait, "my" user was using modest with not a single crash for the last two months18:19
merlin1991yeah go ahead and get rid of thumb in rules for tinymail and modest18:19
merlin1991I really want to know though wich source set the DEB_BUILD_OPTIONS to thumb18:19
merlin1991freemangordon_: do you have commit access on gitorious?18:20
freemangordon_no, that was the other thing i needed MohammadAG18:21
freemangordon_i don't have *ANY* rights on gitorious18:21
* DocSalarySlave dances the killed-bug jig18:21
freemangordon_a bug ?!? this is an elephant man18:21
merlin1991Doc salaryslave? what's the occasion? :D18:21
DocSalarySlavecontract signing mambo jumbo18:22
merlin1991now we just need MAG18:22
* Lava_Croft smiles @ bufixers18:23
merlin1991okay fsckd plasma shell seems to behave now18:23
merlin1991(after 2 reboots)18:23
merlin1991Kubuntu is the suck18:23
merlin1991I think I'll get gentoo or something next time18:23
merlin1991though that's going to be some fun the get scratchbox up and running there :/18:24
merlin1991I don't get it how scratchbox can have a binary only repo18:24
merlin1991their packages contain stuff like gcc18:25
freemangordon_BTW stable thumb on n900 is my next thing TODO. After stable SR18:26
freemangordon_I steal dream of thumb compiled Qt18:26
freemangordon_still*18:26
DocSalarySlaveI dream of Qt not using friggin stupid fp18:29
DocSalarySlaveI mean the screen is max 1280 wide, so why would Qt need floats?18:29
DocSalarySlaveunless some coders skived their math lessons18:30
DocSalarySlaveor were simply lazy18:30
DocSalarySlavenever use float when you can do the job with fixed point arithmetics18:31
DocSalarySlaveif you don't know how to do proper fixed point, get another job!18:32
freemangordon_sure, but i don't think that could be achieved.18:32
freemangordon_while thumb should be possible18:32
freemangordon_actually they use it for simbian builds18:32
DocSalarySlaveobviously thumb is from hell (duh, sb said this before, no?)18:33
freemangordon_symbian*18:33
freemangordon_so what?18:33
DocSalarySlaveso don't use it, as it causes coredumps ;-D18:34
DocSalarySlavewell, that's all hearsay18:34
merlin1991freemangordon_: what's the point of having thumb qt?18:34
DocSalarySlaveactually I have no own notion regarding thumb18:35
nealIs the claim that gcc is generating bad thumb code?18:35
merlin1991thumb is smaller but slower, and qt should be in memory only one anyway18:35
nealor, what's the issue?18:35
nealmerlin1991: Why is thumb slower?18:35
merlin1991because of gcc18:35
freemangordon_neal: yes, seems so18:35
freemangordon_merlin1991: it should be faster18:36
nealI agree with freemangordon_18:36
merlin1991freemangordon_: theory and praxis :D18:36
nealyou get better instruction density => better cache performance18:36
freemangordon_and having in mind the size of qt libs, we can achieve both memory and performance gain18:36
DocSalarySlaveshorter code usually is supposed to be faster, though you can show yu can speed up code by bloating it18:36
freemangordon_bye for now guys, bbl18:37
*** freemangordon_ has quit IRC18:37
* DocSalarySlave waves in freemango's general direction18:38
nealscratchbox uses gcc 4.2.1, which is already 4 years old18:38
nealI suspect that newer versions would correct these issues.18:38
merlin1991I suspect you're right18:39
merlin1991but our environment is gcc 4.2.118:39
DocSalarySlave(speed of code) though, you recall DMA on 386/486? is was like 5 times faster in sw than when using hw-DMA18:46
DocSalarySlavethumb might be a similar relic, implemented in cortex-A8 for mere backward compatibility18:48
merlin1991A8 should have thumb2 afaik18:49
DocSalarySlaveNFC, couldn't even tell the difference. Anyway nothing in offending DEB_BUILD_OPTIONS seems to suggest compiler will use thumb218:51
DocSalarySlaveor is that just a speedier architecure of that core, with identical opcode set?18:52
* DocSalarySlave googles a bit18:53
andre__hmm, in latest testing I don't see any "Accounts" anymore in status bar or settings panel (like gmail or skype). is that a feature? :P18:54
merlin1991andre__: that defenitely is not a feature18:56
merlin1991I don't have testing atm though so I can't say it's because of cssu18:57
DocSalarySlavehttp://de.wikipedia.org/wiki/ARM-Architektur#Thumb-Befehlssatz18:59
DocSalarySlavethumb is smaller but *slower*18:59
DocSalarySlave>>Der Hersteller NXP gibt in den Datenblättern seiner LPC2000-Controller einen Geschwindigkeitsverlust von 30 % an.<<19:00
DocSalarySlaveandre__: settings->VoIP+Chat ?19:02
DocSalarySlaveWFM19:03
andre__DocSalarySlave, ah, darn, yeah, that one is still there. but wasn't that also always in the title bar dropdown (battery, clock, internet, profile, bluetooth)?19:03
andre__maybe my mind plays tricks on me19:03
DocSalarySlaveavailability?19:04
andre__"Availability"19:04
andre__yeah19:04
DocSalarySlaveWFM19:04
andre__hmm. now after enabling and disabling an account it's still there.19:04
andre__no idea why it was missing before.19:04
andre__how weird.19:04
*** freemangordon has joined #maemo-ssu19:10
*** andre__ has quit IRC19:12
*** scoobertron has joined #maemo-ssu19:17
*** rd has joined #maemo-ssu19:49
*** andre__ has joined #maemo-ssu20:10
*** MohammadAG has joined #maemo-ssu20:18
*** xnt14 has joined #maemo-ssu20:19
freemangordonMohammadAG: ping20:21
merlin1991MohammadAG: we've found why modest acts up20:27
* merlin1991 proposes h-d/modest/tinymail update asap20:27
freemangordon(nod)20:27
merlin1991aslo bug 1230620:28
povbotBug https://bugs.maemo.org/12306 Conflict problems for libqt4-test20:28
merlin1991s/also/also20:28
* DocSalarySlave proposes to check other packages for same ooopsie as well21:20
*** DocSalarySlave is now known as DocScrutinizer21:22
*** NIN102 has joined #maemo-ssu21:27
*** keel has joined #maemo-ssu21:27
*** NIN101 has quit IRC21:28
*** StyXman has joined #maemo-ssu21:46
amiconnDocScrutinizer: Thumb instructions as such aren't slower than ordinary arm instructions, the difference comes from the fact that since thumb instructions are less flexible you need slightly more instructions.22:00
amiconnThis is where the ~30% slowdown come from22:00
amiconnMore instructions for doing the same work that is. Of course the 30% are just an average22:01
*** rd has quit IRC22:03
*** NIN101 has joined #maemo-ssu22:07
DocScrutinizeryes, that's what the wikipage says22:09
DocScrutinizeresp they mention missing conditional opcodes except jump22:09
*** NIN102 has quit IRC22:09
DocScrutinizerso opcodes of thumb are 16 instead 32bit, they execute with same speed and you need 30% more to do same work, but still you save 20% size from 32 down to 16*1.322:11
DocScrutinizernot worth the effort in my book22:12
DocScrutinizeraah, forgot: the flushing of cmd pipeline on each jump is a huge bottleneck for execution speed as well. So maybe you don't even need 30% more code, it's actually the speed of execution that goes down 30%, due to a bit more opcodes plus worse execution optimization due to more jumps22:16
DocScrutinizerso size savings could be more than the 20% I mentioned above22:16
DocScrutinizermaybe 30..40%22:16
amiconnflushing pipeline is only necessary if the branch prediction made an error22:18
DocScrutinizeryes22:18
amiconnAnother limitation is that there is no shifter operand, instead you need to issue separate shift instructions22:18
DocScrutinizernice22:18
DocScrutinizerso thumb is MEH22:19
freemangordonfor a device with 256 MBytes?22:19
freemangordonand Qt codesize of about 40?22:19
amiconnThumb is useful for gui code. Depending on typical loop sizes, it might even be faster in some cases22:19
freemangordonwow22:19
freemangordonanyway, please find MohammadAG and try to convince him to clean up the mess22:20
amiconnThe reduced size improves caching22:20
freemangordon(nod)22:20
DocScrutinizerwell, you got a point there. Where execution speed doesn't matter you *could* use thumb to save space22:20
freemangordonbbl22:20
amiconnFor performance critical code you shouldn't use thumb22:21
DocScrutinizeranyway, time for dinner22:21
amiconnWith thumb2 you get best of both, but I'm not sure whether cortex A8 already supports thumb222:21
merlin1991thumb2 is armv7 or armv8 I'm not sure22:28
merlin1991comes with armv722:28
MohammadAGWhat mess?22:32
merlin1991so the mess is that for some reason modest in cssu got thumb compiled22:33
merlin1991some package sets the DEB_COMPILE_OPTIONS that it includes thumb22:33
MohammadAGThe fuck22:33
merlin1991or maybe your DEB_COMPILE_OPTIONS are already set to contain thumb, maybe check that22:33
merlin1991and the modest rules file checks for that and adds -mthumb in case it exists22:34
merlin1991that's also the reason why the modest file in the repo is only 600k bytes instead of 900k22:34
merlin1991err 700k22:34
merlin1991could you check what your DEB_COMPILE_OPTIONS are in your scratchbox?22:35
merlin1991err DEB_BUILD_OPTIONS22:35
*** Kaptenen has quit IRC22:35
DocScrutinizerwould something from binutils tell if a binary is thumb or not? maybe check *all* pkgs in repo22:35
merlin1991and I'll push a commit to tinymail and modest wich prevents the thumb compiling22:35
DocScrutinizerif this shit is sticky from some prev build event, then other pkgs may be flawed as well22:36
merlin1991then we'd like a new cssu testing release with fixed modest, h-d and bug 1230622:36
povbotBug https://bugs.maemo.org/12306 Conflict problems for libqt4-test22:37
merlin1991MohammadAG: could we get that done today?22:37
DocScrutinizerdon't rush please22:37
*** Kaptenen has joined #maemo-ssu22:38
DocScrutinizerdouble-check22:38
merlin1991DocScrutinizer: me and freemangordon triple checked this22:38
DocScrutinizerI don't mean the bug, I mean the fix :-D22:39
DocScrutinizerthe fixed pkgs22:39
DocScrutinizerthe whole release22:39
DocScrutinizerwget cmd to get the thumb binary?22:40
DocScrutinizerooh wait, I got modest installed22:40
merlin1991you do have the binary :D22:40
DocScrutinizerso what exactly was the file that is bad?22:41
amiconnmerlin1991: I don't see why modest compiled for thumb would cause problems though (except perhaps being a little slower)22:41
merlin1991the fsckd one that is .D22:41
merlin1991amiconn: it does coredumps22:41
merlin1991whilst the non thumb one doesn't22:41
amiconnI wonder why22:41
DocScrutinizeris modest binary thumb?22:41
merlin1991yes22:41
merlin1991if you have current cssu22:42
DocScrutinizermodest.launch?22:42
merlin1991exactly that one22:42
amiconnBesides, I didn't experience crashes of modest22:42
freemangordonamiconn: because of gcc version we are using22:43
freemangordonit has buggy thumb support22:43
freemangordonand AFAIK CortexA8 has thumb2 support22:43
freemangordonit is gcc which is to blame, the same with Qt -O3 and broken raster graphics22:44
freemangordonagain, bbl22:44
freemangordon:D22:44
amiconnWhere's the actual modest binary so I can check? /usr/bin/modest is just a link to 'maemo-invoker', which seems to be some kind of launcher22:45
merlin1991/usr/bin/modest.launcher22:45
amiconnfreemangordon: Can't gcc be changed?22:45
merlin1991err modest.launch22:45
amiconn678kB, so I probably have the thumb version22:46
merlin1991MohammadAG: some comment?22:47
DocScrutinizeramiconn: why would we want to do that?22:47
merlin1991amiconn: just as a sidenote, modest in stock maemo isn't compiled with thumb either22:48
amiconnWell, if it's buggy and there is a fixed version, why not?22:48
DocScrutinizerthe fix obviously is to compile modest the way it is compiled on maemo-proper22:48
*** rd has joined #maemo-ssu22:48
DocScrutinizerwith a "fixed" gcc version you buy new bugs22:48
DocScrutinizerbugs that didn't appear on maemo-proper as they didn'T use that gcc version22:49
* DocScrutinizer can't find out how to tell modest.launch is thumb :-(22:51
DocScrutinizerobjdump is not helping22:51
DocScrutinizerfile is not helping22:52
DocScrutinizeris it mere opcode used?22:52
DocScrutinizerif so, for what opcode mnemonic would I grep in objdump -d ?22:53
*** lizardo has quit IRC23:08
merlin1991MohammadAG: can we get an update out today?23:10
*** mase_76 has quit IRC23:12
*** mase76 has quit IRC23:12
freemangordondoc, the way to check is to attach gdb and to see instructions size. if every instruction is 32 bits then it is arm, otherwise it is thumb23:23
*** mase76 has joined #maemo-ssu23:34
*** m0use has joined #maemo-ssu23:37
DocScrutinizergot that meanwhile23:40
DocScrutinizerobjdump /usr/bin/modest.launch -d|less +"/illeg"23:40
DocScrutinizervs23:40
DocScrutinizerobjdump /usr/bin/arecord -d|less +"/illeg"23:41
*** StyXman has left #maemo-ssu23:42
*** rd has quit IRC23:46
*** rd has joined #maemo-ssu23:47

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