#milkymist IRC log for Monday, 2011-07-04

GitHub3[extras-m1] yizhangsh pushed 1 new commit to master: http://bit.ly/mQIbcZ01:28
GitHub3[extras-m1/master] add a new design - Yi Zhang01:28
GitHub50[flickernoise] xiangfu pushed 1 new commit to master: http://bit.ly/meIHGs02:17
GitHub50[flickernoise/master] delete useless extern _res, already include in resolv.h - Xiangfu Liu02:17
wolfspraulximian: ok I'm testing the update-hang with 07-01 a bit more02:46
wolfsprauloops, I meant xiangfu  :-)02:46
wolfspraulxiangfu suggested that I remove the keyboard, and indeed without keyboard it boots into simple mode02:47
wolfspraulmy keyboard is a silicone keyboard and lying here a bit rolled up, maybe a key is pressed permanently. I will now put it flat and reboot again02:47
xiangfuthe silicone keyboard?02:48
wolfspraulxiangfu: something must have changed on that end between 05-09 and 07-01, because I definitely booted the exact same m1 (with kbd inserted) with 05-09 and it worked fine02:48
wolfspraulahh. now that I unrolled the keyboard, it boots into render mode02:48
wolfspraulok I will now try to press a key permanently when booting02:49
wolfspraulhe, now the bug is gone, cannot reproduce anymore even if I keep keys pressed down02:51
wolfspraulxiangfu: so maybe I just keep an eye on it, but for now we can assume this issue to be closed?02:52
wolfspraulI had my silicone keyboard connected while it was in a somewhat contorted position, so it may have sent unusual signals to the m102:52
xiangfuyes. we can close the issue.02:53
wolfspraulafter unplugging and replugging the problem was still there, but after putting the keyboard down flat it went away and now I cannot get it back02:53
wolfspraulalright then, I keep an eye on it02:53
kristianpaulhe, avalon bus looks simpler than wishobne, shame is from altera03:15
GitHub106[extras-m1] yizhangsh pushed 1 new commit to master: http://bit.ly/lzgtJY03:41
GitHub106[extras-m1/master] add new m1 transparent file - Yi Zhang03:41
GitHub47[extras-m1] yizhangsh pushed 1 new commit to master: http://bit.ly/mznc1c03:49
GitHub47[extras-m1/master] locked die cutting outline - Yi Zhang03:49
GitHub88[extras-m1] shiyele pushed 1 new commit to master: http://bit.ly/kA5EOq06:34
GitHub88[extras-m1/master] added v8 case from roh - Wolfgang Spraul06:34
lekernel-                       return -1;07:48
lekernel+                       ret -1;07:48
xiangfuouch, my mistake, gcc not give me warning .07:50
xiangfusent out new one07:53
lekernel+       s = strchr(name, '/');07:58
lekernel+       *s = '\0';07:58
lekernelwhat if strchr returns NULL?07:58
lekernel(in ycb_mknod)07:58
wpwraklekernel: do you get a segfault then ?08:15
larscon lm32 probably not08:19
lekernelwhen properly configured, which I have had no time for yet, it can08:20
wpwrakgreat. that may be very useful for catching bugs that now just cause some silent misbehaviour.08:21
xiangfuso better add:if(s != NULL) *s = '\0';08:21
xiangfu 08:21
lekernelyeah moment, writing email08:22
lekernelwpwrak, qemu should catch those already, and I didn't find prominent ones08:22
Action: xiangfu fixed 'iSerial' not working in ftdi_eeprom08:23
wpwraklekernel: ah, good.08:23
wpwraklekernel: by the way, it seems M1 takes quite a while before it displays anything. is there any hope this time could be reduced ? it's always bad when people try to power on and nothing seems to happen.08:25
lekernelit's intentional, boot screens are usually unwelcome when used in parties08:26
lekernelyou can display a splash screen by holding the power button a bit longer08:26
wpwraklekernel: maybe just display a little logo/indicator in a corner ?08:27
lekernel(though it's cleared after control is passed to RTEMS... one more of those trivial things I'm tired of fixing)08:27
wpwraklekernel: (tired) just a little more effort. you're almost there ! ;-)08:27
xiangfulekernel: [PATCH v3] on mkdir '/' ...  http://dpaste.com/562776/08:46
lekernelpath1 is not freed when returning from line 3808:47
xiangfuthen also line 41 :(08:48
lekernelline 41??08:49
lekernelfor me line 41 is *name = s + 1;08:49
lekerneli'm using the dpaste line numbers08:49
xiangfuline 4708:49
lekernelbut line 47 yes08:49
lekerneli wonder how many memory leaks are lurking in all this code... unfortunately there's no valgrind for rtems :(08:50
xiangfuanother one in mknod :(08:51
GitHub158[extras-m1] yizhangsh pushed 1 new commit to master: http://bit.ly/k2COj608:53
GitHub158[extras-m1/master] made changes - Yi Zhang08:53
xiangfulekernel:  http://dpaste.com/562779/ line: 40, 50, 82 add three free()08:53
lekernelcommit it08:56
lekernelhave you tested it well?08:56
xiangfuI think so.08:57
xiangfulet me test again08:57
lekernela good test is to transfer some complex filesystem tree over FTP08:58
lekerneland then move the root directory of that tree back and forth between the ramdisk and the flash08:59
xiangfuoh. yes.08:59
wolfspraulwpwrak: the m1 boot time was excellent at the beginning (2-3 seconds)13:38
wolfspraulthen it started to grow, for all sorts of reasons13:39
wolfspraulthe final hit was that simple mode (which it normally boots into) will compile all patches before starting rendering13:39
wolfspraulthere is no caching at all right now13:39
wolfspraulso the boot time now is 35 seconds or so13:39
wolfspraulthat's a fairly high priority, at least on my list, to bring down. the easiest is probably to do some form of simple caching, with a timestamp or so to catch patches that were edited13:40
wolfspraulI believe with some simple caching, we can bring it down to 15 seconds or less13:41
wolfspraulthen maybe everything to about 10 is still long hanging fruits, after that it probably gets harder, but... it was 2-3 seconds at the beginning so there are probably ways to bring it back there, if we prioritize this high enough...13:41
wolfspraulthe good news is that it seems all fairly easy to bring down, absolutely nothing that is hard should be in the way...13:42
wpwrakwolfspraul: sounds good. being able to get quickly into its regular operation would indeed be best. else, some indicator that it has really activated would also help. it was a little odd to hear rejon talk about how it's "instantly on" yet i had seen him boot a number of times before and there was always a long period of uncertainty if the device was even running.13:50
wolfspraulyes I know13:54
wolfspraulyou know how it is13:54
wolfspraulover time, boot time seems to know only one direction...13:54
wolfspraulevery down to 15 or 10 should be fairly easy, everything down to 5 seconds still realistic probably. so there is hope...13:55
lekerneltimestamp cannot work, there's no rtc13:56
lekernelhashing, maybe13:56
wolfspraulah true, sorry13:58
wolfspraulyou said that already earlier13:58
wpwraklekernel: couldn't you just delete/regenerate the cache when a new patch is downloaded ?13:59
lekernelpeople can edit patches, too13:59
lekernelmodify files, etc.13:59
wolfspraulbut my point still stands, we can bring the boot time down to 15 seconds or less with relatively little effort, as it just grew recently, I don't expect particularly hard stuff when bringing it back down14:00
lekernellet's ship devices happily booting in 30 seconds _first_14:00
lekernelthen we see14:00
wpwraklekernel: hm. how about a timestamp composed of boot number and uptime ?14:00
wolfspraullekernel: I think it's already over 30 now, and probably keeps growing as we add patches14:01
wolfspraulbut yes, I agree. this is not a critical issue right now, but I do keep my eye on it.14:01
wolfspraulback when it was booted in 2-3 seconds, that was really cool14:02
lekernelthere was no gui, no filesystem, no networking, etc.14:02
wolfspraulcaching of patch compilation alone will bring it back to about 15 seconds14:03
lekerneland it froze every time you switched patches (the patch compilation was here to prevent this)14:03
lekerneland people _did_ have a bad impression because of that switching delay14:03
wpwraklekernel: the problem (for the user) isn't the boot time per se but the indication whether it is actually starting or not. maybe you can see this from the LEDs. are they a clear enough indication that users will recognize them as such without being told by documentation ?14:03
lekernelyes, there is a second LED switching on14:03
lekernelthat says "ON" ...14:04
lekernelso it's a trivial issue, really14:04
lekerneli'd rather worry about how the production goes ...14:04
wpwraklekernel: hmm, okay. maybe it's just a documentation problem then ;-)14:04
wolfspraulno it's not trivial14:04
wolfspraulbut we don't need to do it now14:04
wolfspraulit's not trivial at all14:04
wolfspraulthere are few things as important as quick boot time14:05
wolfspraulbut... a great web update is even far more important14:05
lekernellet's see. for now i'm tired of building features only 3-4 people care about. the way over this is getting the run 3 done and promoting it.14:05
wolfspraulsure, only 3-4 are active from the rc2 bunch :-)14:06
wolfspraulbut really, believe me, for end users boot time is crazy important, it's their first impression of the product14:06
wolfspraulspeed is addictive14:06
wolfspraulbut I am _NOT_ worried about m1 boot time, at all14:06
wolfspraulwe can bring it down with a little targeted engineering, at any time later if we have a good update procedure14:07
lekernelthe update should work now14:07
lekerneli spent my weekend on it ...14:07
wpwrakwolfspraul: agreed on the boot time. also, rejon is emphasizing the "instant on" quite a bit, so you have to decide whether he should keep on advertizing it or not ?14:07
wolfspraulwe can advertize it14:07
wolfspraulwe will bring it back14:07
wolfspraul'instant-on' is almost a must have nowadays, say 5 seconds or so14:08
wolfspraul15 is barely ok, 30 is the total cut-off for consumer products14:08
wolfspraulthat's how it is, those are not my numbers but expectations of people14:08
kristianpaullekernel: there was no lzma compression for flickernoise binary either ;-)14:08
lekernelit's disabled atm14:09
kristianpaulbut even with gui my M1 as arrived took less than 5 seconds to boot i remenber, was reallt fast14:09
kristianpaulah, ok14:10
kristianpaulgood :-)14:10
lekernelno, it's bad it's so slow14:10
kristianpaul5 seconds still because png loading right?14:11
kristianpaulwhatever wallpaper format is.14:11
wpwrakwolfspraul: i see the biggest issue in it taking so long before there's a video signal. that's annoying if you're in a setup where you're not sure if things work. that should be particularly bad for a device like the M1, when used by a traveling VJ.14:12
lekernelthe main things which take time atm14:12
lekernel1) failed dhcp attempts14:12
lekernel2) compilation of patches14:13
lekernelfor 1) the problem is a) rtems blocks startup until dhcp timeout b) mdio doesn't work (i've filed an issue against that) so there's no way to disable it when ethernet is unplugged14:13
kristianpaulabout dhcp i could be done before all the gui is up?14:13
wpwrakwolfspraul: even a little M1 logo in a corner that shows that it is alive would help a lot14:14
kristianpaulwpwrak: there was big logo from bios, but i think there was another blank screen issue just after the gui booted i remnber14:14
Action: kristianpaul long time since not use flickernoise14:15
lekernelkristianpaul, yeah so I'm not motivated to fix that.14:15
kristianpaulso, if you even call dhcp after mtk/gui loaded there will be a freeze?14:17
wpwrakkristianpaul: lekernel said he didn't want to put a big logo. but it wouldn't have to be a big fancy splash screen. actually something small is better, psychologically, because a big fancy screen suggests that the system is wasting resources on the splash screen instead of working on getting booted.14:17
lekernelno, it's bad because everyone sees it14:18
kristianpaulwpwrak: sure the problem will be when your small logo dissapier to blank screen for seconds..14:18
kristianpaulis not that end user scaring?14:18
wolfspraulbtw, Sebastien's brochure that I heard very positive feedback for, on page 3 says "ready in seconds" as one of the big bullet points14:19
wpwraklekernel: that's an even worse situation: VJ gets setup while the crowd is already there. a high-pressure situation. then quick feedback whether the video actually works is even more important.14:19
wolfspraul"Lightweight" "very low latency" "no computer needed" "ready in seconds"14:19
wolfspraulindeed I like those 4 bullet items, they are good14:19
wpwraklekernel: also, do clubs have special projectors that don't helpfully display "NO SIGNAL" or the vendor's logo in that time ?14:20
lekernelwpwrak, clubs tend to avoid those projectors for the same reason I avoid the M1 splash screen14:20
wpwraklekernel: good ;-)14:20
wolfspraulwhen you ask people how many seconds "ready in seconds" means (in this context), I'm sure the answer would be 'around 5'14:20
lekernelwolfspraul, ok, I got it, right? but we ship the run 3 first.14:21
wolfspraulnot 200, for example as with the FreeRunner :-)14:21
lekernel30 seconds is still acceptable14:21
wolfspraullekernel: yes, I'm totally relaxed about it14:21
wolfspraulyou seem to misunderstand me14:21
wpwraklekernel: you could also do something less obvious, e.g., just a frame at the screen edges14:21
wolfspraulplus the work into bringing this down is easy, don't worry someone will get it done14:21
wolfspraulso I did not complain about this bullet item14:21
wolfspraulleave it there14:21
wolfspraulsame as Jon's 'instant-on'14:22
wpwraklekernel: you're going from a blank screen, so a bit of a build-up would actually seem appropriate :)14:22
wolfspraulI am just trying to explain the situation to werner14:22
wolfspraulso everybody stay calm. Jon can continue with instant-on, the bullet item is good, and we will get this back down to 10 seconds or so in due time14:22
wolfspraulwithout Sebastien, who has already been doing 1000 things14:22
wolfspraulhave you seen me bitching about boot time? no :-) I'm just giving the big picture to Werner...14:24
wpwrakwolfspraul: i think "ready in seconds" is very good. just that a long "dark" start time matches this badly. but yes, if this gets fixed in time, there's obviously no problem.14:24
wolfspraulwpwrak: it can and will be brought down again14:25
wolfspraulso those items should stay14:25
wolfspraulI make a 'not yet' label and glue it over the brochue, ha :-)14:25
wolfspraul"ready in seconds" - NOT YET! :-)14:25
wpwrakwolfspraul: (bitchin) yeah, just my observation from watching rejon fight with it. everything else looks very nice and smooth, so this one ugly spot sticks out quite a bit.14:25
wolfspraulyes I totally agree with you, but I think you understand the plan now14:26
wpwrakwolfspraul: ready in more seconds ;-)14:26
wolfspraulfirst - we are talking about the speed in marketing material already now14:26
wolfspraulsecond - we have more focus on easy update procedure first14:26
wolfspraulthird - we are confident that we can bring the boot time back down to 15 seconds or less, because we understand the big blocks that only very recently drove it up from 5 seconds14:26
wpwrakwolfspraul: by the way, when will you expect to start selling these critters ?14:27
wolfspraulfourth - Sebastien doesn't need to worry about this at all because afaik the blockers to bring it down are not very tough technically, so I am optimistic xiangfu or others can get it done14:27
wolfspraulstart selling, maybe in 2-3 weeks?14:28
wolfspraulsmt is next week14:28
wolfspraulremot controls just arrived14:28
wolfspraulpower supplies still 'in the making'14:28
wpwrakvery good. several people have inquired when it will sell.14:28
wolfspraullabels - I believe print job is almost out or out14:28
wpwrak(power supply) does it have an (S) for argentina printed on it ? :)14:29
wolfspraulbox - job not sent out yet, but I hope in the next few days, we need to settle on a design, maybe some more minor tweaks14:29
lekernelyi's design is good btw14:29
lekerneljust add the creative commons credits near the icons14:29
wolfspraulcases - talked to roh earlier today, he still has a lot of work14:30
wolfspraullet's hope this week the lasering is done14:30
wolfsprauland the buttons14:30
wolfspraulyeah, what else14:30
wolfspraulthat's about it I think14:31
wolfspraulso... 2-3 weeks14:31
wolfspraullekernel: which design? (which filename)14:31
wolfsprauland what credits do you want there?14:31
lekerneland see credits on the brochure where i'm using those icons as well14:32
wolfspraulah ok, yes I see it14:33
wolfspraulright now it looks like all of those things like power supplies, box, cases may come in a little late14:34
wolfspraulmy bad14:34
wolfspraul'a little' means about a week or two later than in a perfect world14:35
lekernelwpwrak, about the lamp, it's just that I could not find a better clipart in a timely fashion14:35
wolfspraulpower supplies - I underestimate the lead-times in that type of category. the PO is out long ago, but... 2 months is nothing in power supply land. oh well, lesson learnt :-)14:35
wolfspraulbox - bah. design is always slower than I estimate. will also learn.14:36
lekerneljust add the CC credits and send out for printing14:36
wolfspraulcases - well, roh is juggling between different clients, raumfahrtagentur moved, and we added extra jobs like button assembly14:36
wolfspraullekernel: yes correct14:36
wolfspraulthat's what will happen14:36
wolfspraulat least this will give Adam ample time to test the boards, which is good14:38
wpwraklekernel: (lamp) could you just color it ? the lamp as such looks okay. or is this too difficult ?14:38
wpwrak(button assembly) i hope roh makes a support for that. should be very easy with the mill - two concentric cylindrical cuts in pretty much any material.14:40
wolfspraulthat's exactly what he will do14:40
wpwrakwolfspraul: good :)14:40
wolfspraulbut after the move the mill was dysfunctional, I believe until today or so14:40
wpwrakah, that sucks14:41
wpwrakthe drawback of having a really nice big mill :)14:41
wolfspraulit's so strange14:42
wolfspraulfor cables, lead times is a few days14:42
wolfsprauleven if you have custom length14:42
wolfspraulthey seem to always be able to throw in a small job, like 100 cables, very quickly14:42
wolfspraulbut for power supplies, they live and think in months14:43
wolfspraul2 months, 3 months, etc.14:43
wolfsprauleven though the power supplies are totally standardized as well of course14:43
wpwrakwolfspraul: what will you do about the various plug shapes in different countries ?14:44
wolfspraulI don't know why that is so, but I learnt it. the power supplies should have been in stock long ago, like the cables. now they come in as one of the last pieces.14:44
wolfspraulwe will be shipping with small plastic adapters for US/Japan, EU, UK and Australia14:44
wolfspraulthe rest of the world needs to buy a local adapter14:45
wolfsprauland no Argentina (S) either, unfortunately14:45
wpwrakwill your logistics allow you to remove the power supply before shipping ?14:45
wolfsprauloh sure14:45
wolfspraulfor now yes, no problem14:45
wpwrakgood. this may save a few people some afternoons at customs.14:46
wpwrakwhat shipping options do you have ? just fedex or also others, e.g., EMS ?14:46
wolfspraulhaven't really thought about it14:46
wolfspraulfirst fedex only14:46
wolfspraulbut we are flexible14:47
wolfspraulall m1 will be shipped from Adam initially14:47
wolfspraulnot from the logistics center in Hong Kong14:47
wolfspraulthat's for later...14:47
wpwrakokay. this gives you quite some flexibility.14:47
kristianpaulEMS may save some to pay customs but it really take time to deliver..14:49
wpwrakkristianpaul: EMS may also be friendlier when it comes to sending power supplies to argentina. EMS goes through the regular postal mail process. so a lot of personal parcels take that route. couriers are considered more for commercial shipments, so they may get pickier treatment.16:01
wpwrakkristianpaul: but i haven't done any extensive studies. should send, say, five parcels either way, and see what acceptance rate they have. also couriers shipments don't always get molested. e.g., i once bought some power supplies from digi-key, along with some other electronic components, and didn't have problems.16:03
wpwrakkristianpaul: alas, making such a test would be quite expensive ...16:03
kristianpaulthat remenbers me i need to find something like digikey that uses first class mail16:14
kristianpaulso far i never had problems with first class mail from the US, even i had good surprises of deliver times in days and not weeks16:14
kristianpaulbut yeah, EMS from china is another history i guess, most if you alreayd paid 500usd for a milkymist, why not paid some more dollars for Fedex good delivery :)16:15
kristianpauldeal extreme in the other side is odd shipiment, how they could offer tracking for  jsut 1 dolar and at the end it still first class mail, but deliver still take at least 3 weeks..16:17
kristianpaulhum, may be a good test for deal extreme will be send one item to argentina16:17
kristianpaullekernel: for rc3 there are plans to reach US market for MM1?16:20
wpwrakkristianpaul: EMS from china is quite slow. takes more than a week.16:21
kristianpaulwpwrak: more than a week? ha!, xiangu send me some memory cards thatwas a month, also when openmoko send me a memory card replacement for the wikireader16:22
wpwrakkristianpaul: there is tracking on regular mail as well. it's just a bit spottier than with the couriers. e.g., when i send out a regular parcel, i can track it all the way in argentina, and sometimes even with the postal system at the other end.16:22
wpwrakkristianpaul: (other postal system) e.g., in the US, it worked. in germany, it didn't.16:23
wpwrakkristianpaul: i have something like 10 days in my mental notebook. i don't think EMS took more than two weeks TW -> AR. not sure about CN16:24
kristianpaulwpwrak: yeah, thats true, here in colombia and i noticed when i call postal office about asking tracking for first class mail,i  can feel/listen from ther voice they know in wich part of the country is but they just dont say it16:24
kristianpaulwpwrak: or ie, the mail is on the office, and they intenially delay it to you16:24
kristianpaulwpwrak: yeah TW is other history, i think there is an extra delay just when coming out from CN16:26
wpwrakkristianpaul: when you send something with air mail here in argentina, it gets a tracking number. i think that's automatic for all international shipments.16:27
kristianpaulbut OM was from TW, and took a month as well.. i think for some reason that packages arrive sometimes to venezuela, and thats anoyher week delay..16:27
wpwrakkristianpaul: (CN) customs there may be slower, yes16:27
wpwrakkristianpaul: (VE->CO) hehe, it's nice that they don't just confiscate it ;-)16:28
kristianpaulthe open it btw16:28
kristianpaulwell i just noticed that on a shipment, because there was a label indicating that16:29
wpwrakkristianpaul: so make sure you don't ship around any CDs with reggaeton, or they'll keep them ;-)16:33
kristianpaulbetter they keep it, i'm no interested on that, have _enought_ with noisy neihgboords !!16:34
wpwrakwith the M1, you can now fight back with video :)16:35
kristianpauli need a video projetctor16:37
kristianpaulbut yeah,for halloween and christmas i was thinking to project some animations on of the windows16:37
kristianpauland feed the M1 with the sound land scape coming from street16:38
kristianpaulactually i had a crazy idea last time feed m1 from ogg streams, like the ones from locusonus project16:39
wpwrakyou could make an animated christmas tree :-)16:40
kristianpaulactually even !! a guy i met in labsulab he is from france i think and want to make something like locusonus but with raw readings from sensors16:40
kristianpauland at the end he will provide  OSC server, so just a unique OSC source of effects for flickernoise :-)16:41
kristianpaulthat remenber the question about tweets in FISL mm talk, i think Jon missundertood tweets are just for wall16:43
kristianpaulor he see in the future tweetw can also modify ariables for running patches?16:43
wpwrakhmm, i didn't notice him say anything incorrect there. ambiguous ? and well, i guess tweets could modify variables. don't sound very difficult :)16:44
kristianpaulsure, just they guy who asked seemed to relate twittes with variables than with wall feature i think16:45
kristianpaulnot incorrect16:45
kristianpaulambiguous, yes i think16:46
kristianpaulgah, define and connect 10 wires and 10 registers between 3 modules is something a human should not do!16:56
mwallelekernel: rtems shouldn't jump to DEBA+offset16:56
lekernelwhat to do in case of e.g. divide by zero error, then?16:57
lekernelinfinite loop in the exception handler?16:57
mwalledo a register dump, print sth like "connect debugger" and loop16:57
mwallefor all unhandled exceptions16:58
mwalleexcept breakpoint and watchpoint, which always go to DEBA16:58
GitHub35[extras-m1] yizhangsh pushed 1 new commit to master: http://bit.ly/kgj6Ji18:28
GitHub35[extras-m1/master] add another box design variation - Yi Zhang18:28
kristianpaullekernel: hmm, yes i think i forgot that there is also (Y) code in L1, so thats not implemented in namuru but it could be as well21:31
--- Tue Jul 5 201100:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!