_ade_ | freemangordon: I don't think we need to have libc6 depend on libcityinfo0-0. A new libcityinfo0 should have the right libc6 dependency. | 00:13 |
---|---|---|
_ade_ | If the old libcityinfo is in place, some references may be out of date, but nothing serious. | 00:14 |
freemangordon | _ade_: sure | 00:14 |
_ade_ | sorry for any confusion | 00:14 |
*** _ade_ has quit IRC | 00:27 | |
*** LauRoman|Alt has joined #maemo-ssu | 01:01 | |
DocScrutinizer05 | maybe we should not only make sure that all thumb-binaries reliably bail out the very moment they got started when kernel/system doesn't support thumb | 01:03 |
DocScrutinizer05 | we alsi should define a minimum requirements for system to work even when a full thumbification gets busted by wrong kernel - it still shouldn't result in a bootloop that's not recoverable | 01:04 |
*** LauRoman has quit IRC | 01:04 | |
DocScrutinizer05 | I think it might be a reasonable requirement that a system in such pathological state still offers a self-contained way to recover, e.g by unconditionally loading HAM which mustn't segfault and must allow to install thumb-enabled kernel or in any other way successively recover from that situation | 01:06 |
DocScrutinizer05 | all thumbification then had to make sure that such recovery is always possible | 01:07 |
DocScrutinizer05 | i.e. it shouldn't be allowed to thumbify the init process (upstart, eh?) and the init shell | 01:08 |
DocScrutinizer05 | then the init shell can test for proper system support of thumb and branch to an emergency recovery scheme when that support is missing | 01:11 |
freemangordon | DocScrutinizer05: hmm, that one sounds sane | 01:17 |
DocScrutinizer05 | thanks! :-) | 01:17 |
freemangordon | though... I was thinking of thumb-compiling glibc, so maybe a way to achieve ^^^, is to have non-thumbified libc6 (and all other needed libs) somewhere in /lib/non-thumb/ | 01:19 |
DocScrutinizer05 | or bind statically | 01:20 |
freemangordon | busybox? | 01:20 |
freemangordon | might work | 01:20 |
DocScrutinizer05 | umm | 01:20 |
freemangordon | preinit script is executed by /bin/sh afaik | 01:21 |
freemangordon | which is BB imo | 01:21 |
freemangordon | but should be possible to be done | 01:21 |
DocScrutinizer05 | yeah, you might pull in a statically linked non-thumb miinimal aka standard version of busybox to do that initial testing. Could exec to busybox-power-thumb when test been success | 01:22 |
freemangordon | exactly | 01:22 |
*** NIN101 has quit IRC | 01:22 | |
freemangordon | DocScrutinizer05: no -power- :) unless it enters cssu some day :) | 01:23 |
freemangordon | anyway, I'll think about some runtime protection | 01:24 |
DocScrutinizer05 | for that thumb version that gets explicitly started by init scripts and is not needed to run nit initially, I don't mind | 01:24 |
freemangordon | yeah, should make no difference | 01:24 |
DocScrutinizer05 | can be busybox-super-mega-power and live on eMMC and eat 200MB there | 01:25 |
freemangordon | no | 01:25 |
freemangordon | this must be on rootfs | 01:25 |
freemangordon | nothing elese is mounted at that point | 01:25 |
freemangordon | *else | 01:25 |
DocScrutinizer05 | yeah, you have postpone the exec then | 01:25 |
DocScrutinizer05 | until eMMC mounted | 01:25 |
freemangordon | checking for thumb support in the kernel should be the first think to be done in preinit script IMO | 01:26 |
freemangordon | *thing | 01:26 |
DocScrutinizer05 | you could even bindmount the bb-supadupa from eMMC on top of the original one in rootfs | 01:26 |
freemangordon | I can't as it will be the thumbified bb that will contain the mount and it could sigill ;) | 01:27 |
DocScrutinizer05 | for those who think bb-supadupa is absolutely mandatory for init, for whatever reason | 01:27 |
freemangordon | so the first sh to run should be non-thumb | 01:27 |
freemangordon | sh/bb | 01:27 |
DocScrutinizer05 | freemangordon: you should consider if you want to start a business with neo900 group | 01:28 |
freemangordon | hmm, business as in? | 01:28 |
DocScrutinizer05 | we need or rather want an OS | 01:28 |
DocScrutinizer05 | it should have added value that's stuff like VPN, cryptfs, another few goodies, and it should have support from maintainer for registered users, and we would ship it with each Neo900 and pay you per device sold | 01:30 |
DocScrutinizer05 | you could even offer monthly or per-year registration to the servers offering the updates and stuff | 01:30 |
DocScrutinizer05 | of course I think it should be based on fremantle ;-) | 01:32 |
DocScrutinizer05 | basically you could run a SSU | 01:32 |
DocScrutinizer05 | of course we don't expect new software getting developed, just distro maintenance basically | 01:35 |
DocScrutinizer05 | you package a distro (much like SuSE did in former times, and still does with SLES etc) and you charge for that service | 01:36 |
freemangordon | DocScrutinizer05: the problem is that I can work on that project only in my spare time, which is a bit limited | 01:36 |
DocScrutinizer05 | do you think you could coordinate a group of developers all willing to work on it parttime and share the dollars? | 01:37 |
freemangordon | I think yes | 01:37 |
DocScrutinizer05 | do you think you could *find* such group of developers? | 01:37 |
freemangordon | though, keep in mind, I am not doing all this stuff for money ;) | 01:37 |
freemangordon | no, this is a big problem | 01:38 |
freemangordon | I can;t even find good developers for my dayly job | 01:38 |
* dos1 is interested | 01:38 | |
DocScrutinizer05 | freemangordon: see? ^^^ :-) | 01:38 |
freemangordon | yeah :) | 01:38 |
DocScrutinizer05 | freemangordon: maybe just try a tiny bit harder ;-) | 01:38 |
dos1 | I'm available at around half-time next semester ;) | 01:39 |
DocScrutinizer05 | of course we (Neo900) are willing to place adds to the website and do whatever we can to find devels for your group | 01:40 |
freemangordon | DocScrutinizer05: see, if I am going to get money for a job, I should be involved more than my free time allows. remember, there are other people that count on me for a job to be done :) | 01:40 |
freemangordon | DocScrutinizer05: anyway, let me have some time to think about that stuff | 01:41 |
DocScrutinizer05 | sure | 01:41 |
freemangordon | (and to finally find the reason why gcc4.7.2 screwes glibc locales) | 01:42 |
DocScrutinizer05 | we maybe can slightly modify the whole business model when you e.g. say you don't want to run SSU infra, we can find somebody else for doing that | 01:42 |
freemangordon | maybe some of the ex-nokians would be interested | 01:43 |
DocScrutinizer05 | quite possible | 01:43 |
*** hbib has quit IRC | 01:46 | |
DocScrutinizer05 | if anybody else is listening and thinking "Umm, *I* could do *that*" - don't hesitate to ping me or send a mail to contact at neo900.org. We are not picky about who's "boss" of such project | 01:46 |
freemangordon | dammit, seems like a bug in gcc :( | 02:02 |
*** kolp has quit IRC | 02:06 | |
*** BCMM has quit IRC | 02:06 | |
DocScrutinizer05 | wut? you found a bug in gcc? incredible! | 02:26 |
DocScrutinizer05 | ;-) | 02:26 |
*** arcean has quit IRC | 03:03 | |
*** aap has quit IRC | 03:38 | |
*** aap has joined #maemo-ssu | 03:38 | |
*** okias has quit IRC | 03:43 | |
*** LauRoman|Alt has quit IRC | 04:11 | |
*** sixwheeledbeast has quit IRC | 04:16 | |
*** M4rtinK has quit IRC | 04:18 | |
*** dafox has quit IRC | 04:24 | |
*** dos1 has quit IRC | 04:33 | |
*** nox- has quit IRC | 04:45 | |
*** javispedro has quit IRC | 05:02 | |
*** amiconn has quit IRC | 05:42 | |
*** amiconn_ has joined #maemo-ssu | 05:42 | |
*** amiconn_ is now known as amiconn | 05:43 | |
*** ototo is now known as zz_ototo | 09:12 | |
*** cappu has quit IRC | 09:25 | |
*** cappu has joined #maemo-ssu | 09:29 | |
*** sixwheeledbeast has joined #maemo-ssu | 09:55 | |
*** NIN101 has joined #maemo-ssu | 10:14 | |
*** LauRoman has joined #maemo-ssu | 10:17 | |
*** trx has quit IRC | 10:36 | |
*** trx has joined #maemo-ssu | 10:40 | |
*** trx has joined #maemo-ssu | 10:40 | |
*** luf has joined #maemo-ssu | 11:01 | |
*** kolp has joined #maemo-ssu | 11:43 | |
*** dafox has joined #maemo-ssu | 11:53 | |
*** dafox is now known as Guest51131 | 11:54 | |
*** NIN102 has joined #maemo-ssu | 11:57 | |
*** NIN101 has quit IRC | 11:59 | |
*** luf has quit IRC | 12:30 | |
freemangordon | kerio: http://talk.maemo.org/showpost.php?p=1407462&postcount=1917 | 12:48 |
kerio | mh | 12:49 |
kerio | it could be a problem | 12:49 |
freemangordon | hmm? | 12:49 |
kerio | i still have no decent internets on my n900 | 12:49 |
kerio | actually ima backup | 12:49 |
freemangordon | so? | 12:49 |
kerio | idk its a bit annoying | 12:49 |
kerio | i'll do it tho, need to backup first | 12:49 |
freemangordon | anyway you should download the .deb by hand | 12:49 |
freemangordon | ok | 12:49 |
kerio | fuck moving files around | 12:50 |
kerio | i wget | 12:50 |
kerio | or bust | 12:50 |
freemangordon | kerio: BTW I have the impression that everything is 2 times faster. could be a placebo effect though | 12:50 |
kerio | :3 | 12:50 |
kerio | also how's it possible that a 2600eur laptop only has 2 usb ports i dont get it | 12:51 |
freemangordon | kerio: wait, what?!? you paid 2600 euros for a laptop? | 12:51 |
kerio | well i paid 2400 for it | 12:51 |
freemangordon | are you crazy man? | 12:51 |
kerio | but it's so fucking cool | 12:52 |
freemangordon | does it do blowjobs? I guess for that price :P | 12:52 |
kerio | ...why the hell does rescueos need a login via telnet | 12:53 |
kerio | instead of giving you a shell | 12:53 |
* ShadowJK 's obnoxiously expensive laptop came with 5 usb ports | 12:54 | |
kerio | fwiw they can't really fit i guess | 12:54 |
kerio | meh | 12:54 |
kerio | this shits like crazy small | 12:54 |
freemangordon | ShadowJK: are you on -thumb? | 12:55 |
ShadowJK | yes | 12:55 |
freemangordon | wanna help me test thumb-compiled glibc6? | 12:55 |
ShadowJK | uh, okay | 12:55 |
ShadowJK | sounds scary :) | 12:55 |
kerio | i'm backupping dis | 12:56 |
* freemangordon is afk for a while | 12:56 | |
kerio | stupid question: does it work on *your* device? | 12:58 |
freemangordon | kerio: see the link I posted ;) | 12:58 |
freemangordon | kerio: ofc it does | 12:58 |
kerio | alright i trust you | 12:58 |
freemangordon | how do you think I decided that: | 12:58 |
freemangordon | it is 2 times fatser | 12:59 |
kerio | idk it's something i'd do :3 | 12:59 |
kerio | "can you please test this it might wreck your n900, i don't wanna test it on mine" | 12:59 |
freemangordon | kerio: you're not me :) | 13:09 |
freemangordon | I never, ever, ask lusers to install thumb-related stuff on their devices without first installing it on my daily and had a reboot | 13:11 |
kerio | mh, why isn't internet sharing working as intended here | 13:12 |
kerio | meh | 13:12 |
freemangordon | ShadowJK: hmm, why scary, it is just another library :) | 13:15 |
kerio | gaaah why isn't this working | 13:21 |
kerio | fucking osx | 13:21 |
kerio | meh | 13:22 |
kerio | "just works" my ass | 13:22 |
*** BCMM has joined #maemo-ssu | 13:23 | |
kerio | freemangordon: wait, why are you using an external file storage thing | 13:26 |
kerio | instead of your space on merlin's server | 13:26 |
freemangordon | hmm, good question :D | 13:27 |
kerio | i can't wget that :s | 13:27 |
freemangordon | however, now my desktop is finally on linux, I'll set up web server and debian repo soon | 13:27 |
freemangordon | kerio: download it with microb | 13:28 |
kerio | also i *really* need to find a way to use my home connection | 13:28 |
kerio | i need to update 22MB of things | 13:28 |
kerio | 22 is a lot | 13:28 |
kerio | ...meh fuck it | 13:28 |
freemangordon | kerio: what is the problem with your home connection? | 13:28 |
kerio | i can't use it | 13:28 |
freemangordon | why? | 13:28 |
kerio | because my n900's wifi doesn't work | 13:28 |
freemangordon | ooh | 13:28 |
freemangordon | broken? | 13:29 |
kerio | i think so, yeah | 13:29 |
freemangordon | hmm | 13:29 |
kerio | now, with snow leopard, internet sharing over usbnet used to work | 13:29 |
kerio | but now, nope | 13:29 |
kerio | i wonder if i can use bluetooth pan | 13:30 |
kerio | i'll tell you, it's really pretty fucking annoying | 13:32 |
kerio | i should probably set up some "stable" usb networking on my home server actually | 13:32 |
kerio | i really need a powered usb hub for that, tho | 13:32 |
kerio | the sheevaplug has no protection circuits on the usb port | 13:33 |
kerio | so it's easy to overdraw | 13:33 |
kerio | and wreck the power supply | 13:33 |
freemangordon | why don;t you by another device with everything working? | 13:33 |
kerio | i'm not gonna buy another n900 | 13:33 |
kerio | seriously | 13:33 |
*** dos1 has joined #maemo-ssu | 13:34 | |
freemangordon | hmm, why? | 13:34 |
kerio | the one i have works! | 13:34 |
freemangordon | but, but,... you said wifi dowsn't wokr | 13:35 |
freemangordon | omg, what a typing skills | 13:35 |
kerio | freemangordon: yeah but it makes phone calls | 13:35 |
freemangordon | you lost me on that one | 13:35 |
kerio | it does serve its main purpose | 13:37 |
*** _rd has joined #maemo-ssu | 13:37 | |
kerio | freemangordon: ok here goes nothin' | 13:39 |
trx | freemangordon installed, rebooted | 13:39 |
trx | so far so good | 13:39 |
kerio | rebooting | 13:39 |
kerio | h-d isn't crashing | 13:40 |
kerio | so far so good | 13:40 |
freemangordon | good | 13:40 |
trx | it actually is faster | 13:40 |
freemangordon | trx: ain't? | 13:41 |
trx | or placebo :) | 13:41 |
freemangordon | yeah :D | 13:41 |
trx | but it is | 13:41 |
kerio | the point of placebo isn't that it does nothing | 13:41 |
kerio | the point of placebo is that it works | 13:41 |
kerio | :3 | 13:41 |
kerio | anyway, it works | 13:41 |
trx | it started up faster, and opens shit faster | 13:41 |
freemangordon | :nod: | 13:41 |
freemangordon | same feeling here | 13:41 |
trx | good job :) | 13:41 |
kerio | feel faster but it could be suggestion | 13:41 |
freemangordon | will compare with my GF's device when she is here | 13:42 |
kerio | let's get DocScrutinizer into this | 13:43 |
ShadowJK | main purpose is wifi+3g for irc, surely | 13:43 |
kerio | ShadowJK: to be honest, yeah | 13:44 |
kerio | but meh, i have a laptop | 13:44 |
kerio | so wifi+irc isn't *that* useful | 13:44 |
kerio | and irc uses so little traffic anyway | 13:44 |
freemangordon | hmm, ok, will test it for a week and will issue a new update | 13:46 |
*** trx has quit IRC | 14:11 | |
*** M4rtinK has joined #maemo-ssu | 14:13 | |
*** okias has joined #maemo-ssu | 14:16 | |
*** _rd has quit IRC | 14:17 | |
*** Guest51131 has quit IRC | 14:17 | |
*** hbib has joined #maemo-ssu | 14:36 | |
Ashley` | freemangordon, I am not sure, but I think my N900 with your libc started up faster :P | 14:42 |
*** dafox has joined #maemo-ssu | 14:50 | |
*** _rd has joined #maemo-ssu | 14:55 | |
*** _rd has quit IRC | 15:24 | |
kerio | it probably did | 15:29 |
kerio | i mean, libc is kind of used everywhere | 15:29 |
kerio | and this is a better version | 15:29 |
*** NIN102 has quit IRC | 15:33 | |
*** NIN102 has joined #maemo-ssu | 15:34 | |
*** scoobertron has quit IRC | 15:41 | |
*** scoobertron has joined #maemo-ssu | 15:41 | |
*** BCMM has quit IRC | 15:51 | |
*** scoobertron has quit IRC | 15:57 | |
*** scoobertron has joined #maemo-ssu | 16:01 | |
*** _rd has joined #maemo-ssu | 16:05 | |
*** _rd has quit IRC | 16:22 | |
*** javispedro has joined #maemo-ssu | 16:24 | |
*** _rd has joined #maemo-ssu | 16:39 | |
*** trx has joined #maemo-ssu | 16:54 | |
*** trx has joined #maemo-ssu | 16:54 | |
*** _rd has quit IRC | 16:56 | |
*** _rd has joined #maemo-ssu | 17:06 | |
freemangordon | hmm, I tried to compare application startup times between my device with -thumb glibc and my GF's device, with cssu's one. Also compared the devices startup times. My device is OC to 805, her is @ 600. Surprise, surprise: | 17:16 |
freemangordon | There is no difference :D | 17:17 |
freemangordon | so, that means that my test cases are flawed and we're still flash/swap speed bound | 17:17 |
freemangordon | I am open to suggestions on how to test thumb/non-thumb glibc speed | 17:18 |
freemangordon | DocScrutinizer05: ^^^ | 17:18 |
freemangordon | all others too ofc: ^^^ | 17:19 |
*** _rd has quit IRC | 17:24 | |
*** trx has quit IRC | 17:31 | |
*** hbib1 has joined #maemo-ssu | 17:31 | |
*** _rd has joined #maemo-ssu | 17:33 | |
*** hbib has quit IRC | 17:34 | |
DocScrutinizer05 | I think there is no better test than real usecase like boot time and app startup. It's probably hard to create and run real standardized application performance tests | 17:41 |
*** trx has joined #maemo-ssu | 17:48 | |
kerio | DocScrutinizer05: install the glibc! | 17:51 |
*** dhbiker has joined #maemo-ssu | 17:53 | |
*** arcean has joined #maemo-ssu | 17:53 | |
freemangordon | DocScrutinizer05: yes, but if the speed is IO bound, such test tells nothing about glibc speed | 18:19 |
DocScrutinizer05 | well, then glibc speed is irrelevant? | 18:20 |
DocScrutinizer05 | when you find your PC has a bottleneck on IO/HDD, will you throw massive amounts of money at a octocore instead of the dualcore it may have right now, when you consider improving it? | 18:28 |
*** _rd has quit IRC | 18:30 | |
*** _rd has joined #maemo-ssu | 18:34 | |
freemangordon | DocScrutinizer05: come on, I am not 13 years old :P. If glibc is faster that means slightly better battry life, ain't? | 18:39 |
freemangordon | DocScrutinizer05: I want to measure glibc performance, not the system performance. But have no idea how | 18:40 |
DocScrutinizer05 | umm, right. then that's the way to go for glibc evaluation, no? simply check system load average last 5 min, while doing exactly same tasks on two systems | 18:40 |
dos1 | aren't thumb binaries actually slower? | 18:40 |
DocScrutinizer05 | supposed to be slower, yes | 18:41 |
dos1 | I thought that speed up on cssu-thumb came from decreased memory usage | 18:41 |
DocScrutinizer05 | :nod: | 18:41 |
DocScrutinizer05 | I'm unclear if this whole thing is about thumb though | 18:41 |
freemangordon | dos1: but the catch is that gcc used to build them is waaay better than stock | 18:42 |
DocScrutinizer05 | might be related to a newer improved glibc version, unrelated to thumb | 18:42 |
dos1 | freemangordon: ah, ok | 18:42 |
DocScrutinizer05 | :-) | 18:42 |
DocScrutinizer05 | makes sense | 18:42 |
freemangordon | dos1: for example - some openssl benchmarks are 2 times faster with thumb and gcc 4.7.2 | 18:43 |
DocScrutinizer05 | freemangordon: check system load | 18:43 |
DocScrutinizer05 | make sure you're not suffering from IO lagging | 18:43 |
freemangordon | DocScrutinizer05: hmm, ok. will ask google too | 18:43 |
freemangordon | it is not a lag, might be related to rootfs compression | 18:43 |
DocScrutinizer05 | e.g game playing might be a good way to check for CPU load | 18:43 |
freemangordon | don;t think so | 18:44 |
DocScrutinizer05 | nah, IO spoils load figures while CPU actually might be idle | 18:44 |
freemangordon | yep | 18:44 |
DocScrutinizer05 | thus do something that's CPU heavy and IO light | 18:44 |
DocScrutinizer05 | and then check system load | 18:45 |
DocScrutinizer05 | A vs B | 18:45 |
freemangordon | hmm, what about that one? http://www.etalabs.net/libc-bench.html | 18:46 |
DocScrutinizer05 | time for ((x=0; x<10000000; x++ )) do :; done | 18:46 |
DocScrutinizer05 | well, maybe 2 "0" less | 18:48 |
freemangordon | lets see | 18:49 |
DocScrutinizer05 | IroN900:~# time for ((x=0; x<100000; x++ )) do :; done; cat /proc/loadavg | 18:51 |
DocScrutinizer05 | real 0m15.943s | 18:51 |
DocScrutinizer05 | user 0m15.344s | 18:51 |
freemangordon | hmm, actually it takes about a second for every test, lets not OK | 18:51 |
DocScrutinizer05 | sys 0m0.289s | 18:51 |
DocScrutinizer05 | 0.23 0.07 0.02 1/202 22962 | 18:51 |
freemangordon | OC even | 18:51 |
DocScrutinizer05 | maybe I need to add one 0 | 18:51 |
freemangordon | use how it is, the whole bench takes about 40 seconds | 18:51 |
freemangordon | here :) | 18:51 |
freemangordon | real0m 35.85s | 18:52 |
freemangordon | user0m 17.40s | 18:52 |
freemangordon | sys0m 14.48s | 18:52 |
freemangordon | device @ 600 MHz, glibc-thumb | 18:53 |
freemangordon | this is for libc-bench, lets see how it will perform on cssu glibc | 18:53 |
DocScrutinizer05 | IroN900:~# time for ((x=0; x<1000000; x++ )) do :; done; cat /proc/loadavg | 18:56 |
DocScrutinizer05 | real 2m41.718s | 18:56 |
DocScrutinizer05 | user 2m36.547s | 18:56 |
DocScrutinizer05 | sys 0m2.336s | 18:56 |
DocScrutinizer05 | 0.94 0.46 0.17 1/202 22963 | 18:56 |
freemangordon | real0m 35.85s | 18:57 |
freemangordon | user0m 17.40s | 18:57 |
freemangordon | sys0m 14.48s | 18:57 |
freemangordon | device with cssu libc @ 600 MHz | 18:57 |
freemangordon | oops | 18:57 |
freemangordon | real0m 35.60s | 18:57 |
freemangordon | user0m 17.16s | 18:57 |
freemangordon | sys0m 14.38s | 18:57 |
freemangordon | that one ^^^ | 18:57 |
freemangordon | DocScrutinizer05: -sh: syntax error: "(" unexpected | 18:58 |
freemangordon | do you have bash by chance? | 18:58 |
DocScrutinizer05 | use a proper shell | 18:58 |
freemangordon | oh, come on | 18:58 |
freemangordon | I won't install bash on my GF's device just to please you :) | 18:59 |
DocScrutinizer05 | then rewrite the above oneliner to basicbourne | 19:00 |
* freemangordon is C/++ developer :P | 19:00 | |
DocScrutinizer05 | and i'll happily c&p and echo the results | 19:00 |
freemangordon | not a linux admin | 19:00 |
freemangordon | seriously, it'll take me 15 minutes to figure it out | 19:00 |
DocScrutinizer05 | the obvious use of "for i in `seq 1 1000000`" might fail epically | 19:01 |
DocScrutinizer05 | you'll have to unroll the for-statement into a proper conditional sequence wrapped into a while loop | 19:02 |
DocScrutinizer05 | sth like i=0; while [ i -lt 100000 ]; do :; done | 19:03 |
DocScrutinizer05 | err | 19:03 |
DocScrutinizer05 | sth like i=0; while [ i -lt 100000 ]; do $(( i++ )) :; done | 19:03 |
freemangordon | DocScrutinizer05: scratch that , this libc-bench tests various stuff, like ptherds etc | 19:04 |
freemangordon | seems glibc-thumb is neither faster nor slower than cssu glibc | 19:04 |
DocScrutinizer05 | that's for sure nice but far from my every day reality | 19:04 |
DocScrutinizer05 | well, THAT result however is pretty close to my reality ;-P | 19:05 |
freemangordon | ok, maybe 2-5% faster, but i guess that is not relevant | 19:05 |
DocScrutinizer05 | yeah, <5% is probing error | 19:05 |
freemangordon | what counts is that gliibc alone is 1.2 MB in cssu and ~800k thumbified | 19:05 |
DocScrutinizer05 | hmm, so we saved 400kB | 19:06 |
DocScrutinizer05 | which is nice though not groundbreaking | 19:06 |
freemangordon | more than that, there are more libs and bins in glibc | 19:09 |
freemangordon | maybe aboiut 600-700KiB | 19:09 |
*** _rd has quit IRC | 19:20 | |
*** FlameReaper has quit IRC | 19:22 | |
DocScrutinizer05 | really nice, but still... It's a shared lib | 19:33 |
DocScrutinizer05 | I guess the best thing to say about new glibc would be that it's basically not distinguishable from old one | 19:34 |
DocScrutinizer05 | 100% compatibility is of paramount importance. everything else is just a bonus on top | 19:36 |
DocScrutinizer05 | I've seen gcc or linker fsckup symbol names so stuff broke with no apparent reason | 19:37 |
DocScrutinizer05 | and of course it did that for one variable type only, in one particular situation | 19:37 |
DocScrutinizer05 | sure apps liked against new lib used new symbol name, but "old" apps suddenly stopped working with "symbol not found" despite the particular function didn't change at all in new lib version | 19:40 |
DocScrutinizer05 | linked* | 19:41 |
*** _rd has joined #maemo-ssu | 19:41 | |
DocScrutinizer05 | and since no proper testframe came with the whole thing, the problem only showed up much later, when first app not only referred to but actually used that particular function (iirc) | 19:42 |
DocScrutinizer05 | must have been like 6 years or more ago, so sorry when my memory is sparse | 19:43 |
DocScrutinizer05 | iirc it been in context of twinkle and probably libboost or the rtp-libs | 19:44 |
DocScrutinizer05 | or maybe even qt | 19:44 |
freemangordon | DocScrutinizer05: just look at lines 39-40 here https://gitorious.org/community-ssu/glibc/commit/c3afe0ae3f658b947126e27950ad0cf41410ae23 | 19:55 |
freemangordon | that is what fixes the non-working locales :D | 19:55 |
freemangordon | re: "I've seen gcc or linker fsckup symbol names so stuff broke with no apparent reason" | 19:56 |
freemangordon | without that patch _nl_category_name_idxs[category] always evaluates to 0, no matter the index | 19:57 |
freemangordon | gcc screwed it badly this time | 19:57 |
freemangordon | or /me knows nothing about C | 19:58 |
*** ravelo has joined #maemo-ssu | 20:28 | |
*** _rd has quit IRC | 20:33 | |
*** M4rtinK has quit IRC | 20:33 | |
DocScrutinizer05 | err, looks pretty much gibbersih to me. | 20:39 |
DocScrutinizer05 | anyway that one actual c line that's in there looks semantically changed to me | 20:40 |
DocScrutinizer05 | _nl_category_names.str looks like an object, while _nl_category_name_idxs[category] seems an array | 20:42 |
DocScrutinizer05 | no idea what happens when adding those | 20:42 |
DocScrutinizer05 | to make matter worse, the whole thing is a multiline #def aiui | 20:45 |
DocScrutinizer05 | which renders the whole thinh subject to macroprocessor parsing. No idea what the macroprozessor creates out of it | 20:47 |
DocScrutinizer05 | thing* | 20:47 |
DocScrutinizer05 | a trailing space can already mess up stuff I guess | 20:47 |
DocScrutinizer05 | I seem to recall it's not first time I seen "( )" used in macros, for obscure reasons | 20:49 |
DocScrutinizer05 | though the major diff is the deref* | 20:49 |
DocScrutinizer05 | and all this needs checking if it might mean something special to the macroprocessor | 20:50 |
DocScrutinizer05 | don't ask me about the makefile sed commands, completely obscure | 20:51 |
DocScrutinizer05 | +nscd-cflags += -fstack-protector $(common-objpfx)$(binfmt-subdir)/ld-linux.so.3 errwutevr | 20:52 |
*** _rd has joined #maemo-ssu | 20:57 | |
freemangordon | DocScrutinizer05: scratch the makefiles change, I pointed you to lines 39-40 | 21:21 |
freemangordon | - _nl_category_names.str + _nl_category_name_idxs[category] | 21:21 |
freemangordon | 40 + (_nl_category_names.str + *(_nl_category_name_idxs + category)) | 21:22 |
freemangordon | this is one and the same thing | 21:22 |
freemangordon | but gcc fails to properly compile the old one | 21:22 |
freemangordon | that "_nl_category_names.str + _nl_category_name_idxs[category]" one | 21:22 |
freemangordon | this always results in like "category" is 0 | 21:23 |
freemangordon | I put the brackets as I am used to always surrond macro expressions in brackets | 21:24 |
freemangordon | *surround | 21:24 |
ShadowJK | I forget if it depends on how the type whether foo[bar] is same as * (foo + bar) or not | 21:28 |
*** M4rtinK has joined #maemo-ssu | 21:30 | |
*** zz_ototo is now known as ototo | 21:30 | |
freemangordon | ShadowJK: index is always converted to in and whether it is *(p+i) or p[i] makes (well, should not make) any difference | 21:32 |
freemangordon | s/in/int/ | 21:32 |
infobot | freemangordon meant: ShadowJK: intdex is always converted to in and whether it is *(p+i) or p[i] makes (well, should not make) any difference | 21:32 |
freemangordon | hmm? | 21:33 |
freemangordon | anyway, read "converted to int" | 21:33 |
freemangordon | ShadowJK: both dereference the same pointer type | 21:34 |
freemangordon | well, it is not the first time in the history when the compiler oversmarts itself | 21:34 |
*** NIN102 has quit IRC | 21:35 | |
freemangordon | ShadowJK: think of *p and p[0] | 21:35 |
DocScrutinizer05 | freemangordon: is this c or c++? | 21:42 |
*** NIN101 has joined #maemo-ssu | 21:42 | |
DocScrutinizer05 | aiui _nl_category_names.str is an object | 21:42 |
freemangordon | DocScrutinizer05: this is C | 21:42 |
freemangordon | glibc, remember :P | 21:42 |
DocScrutinizer05 | ooh | 21:42 |
DocScrutinizer05 | :-P | 21:42 |
DocScrutinizer05 | well, then it's actually just a question if it maybe is a macropreprocessor hiccup | 21:43 |
DocScrutinizer05 | you qould try kick out optimization, just for test | 21:44 |
DocScrutinizer05 | could* | 21:44 |
DocScrutinizer05 | -O0 ? | 21:44 |
*** FlameReaper has joined #maemo-ssu | 21:44 | |
DocScrutinizer05 | check preproc output of both versions | 21:45 |
DocScrutinizer05 | check assembly of both versions | 21:45 |
DocScrutinizer05 | will tell a story, I'm sure | 21:45 |
DocScrutinizer05 | no doubt it's a bug or highly undesirable "feature" of gcc | 21:46 |
ShadowJK | lol, I logged in to my server and did "nano test.c" and discovered previous test was also pointer arithmetic | 21:46 |
DocScrutinizer05 | nevertheless I'd want to check what exactly happening there | 21:47 |
DocScrutinizer05 | maybe compiler thinks _nl_category_name_idxs is an array of vars of size 0 | 21:48 |
DocScrutinizer05 | (yeah I know that doesn't make any sense) | 21:49 |
DocScrutinizer05 | nayway sizeof(_nl_category_name_idxs) would be first thing I'd check | 21:50 |
DocScrutinizer05 | as well as sizeof(_nl_category_name_idxs[0]) | 21:51 |
*** _rd has quit IRC | 21:55 | |
*** FlameReaper has quit IRC | 21:56 | |
*** M4rtinK2 has joined #maemo-ssu | 22:04 | |
DocScrutinizer05 | I'd almost *bet* compiler "thinks" the elements of _nl_category_name_idxs are size 0 | 22:05 |
DocScrutinizer05 | does it comsist of structs? | 22:06 |
DocScrutinizer05 | which are still "empty" at that time/phase of compiler run? | 22:07 |
*** M4rtinK has quit IRC | 22:07 | |
freemangordon | DocScrutinizer05: glibc can't be compiled without optimization | 22:08 |
freemangordon | there is a compile-time check | 22:08 |
DocScrutinizer05 | and I bet it optimizes by compile-time solving of constant arithmetic expressions | 22:08 |
DocScrutinizer05 | and for some reason thinks sizeof(_nl_category_name_idxs) == 0 | 22:09 |
DocScrutinizer05 | thus the whole "+ _nl_category_name_idxs[category]" term is const 0 | 22:10 |
DocScrutinizer05 | --> optimize it out | 22:10 |
*** _rd has joined #maemo-ssu | 22:16 | |
*** M4rtinK2 has quit IRC | 22:27 | |
*** _rd has quit IRC | 22:32 | |
*** _rd has joined #maemo-ssu | 22:33 | |
*** arcean has quit IRC | 22:54 | |
*** _rd has quit IRC | 22:56 | |
*** LauRoman has quit IRC | 23:01 | |
*** hbib1 has quit IRC | 23:03 | |
*** M4rtinK2 has joined #maemo-ssu | 23:20 | |
kerio | freemangordon: have you seen luf lately? | 23:20 |
*** nox- has joined #maemo-ssu | 23:22 | |
*** xes has joined #maemo-ssu | 23:29 | |
*** okias has quit IRC | 23:41 | |
*** ravelo has quit IRC | 23:48 | |
*** NIN101 has quit IRC | 23:48 |
Generated by irclog2html.py 2.15.1 by Marius Gedminas - find it at mg.pov.lt!