| <<<Back 1 day (to 2013/01/28) | 2013/01/29 |
mlilja | Hi, I have a issue about files getting bigger after running ghostscript with "compression", I have an example-file that gets over 100 times bigger after running gs -sDEVICE=pdfwrite........ | 08:02.57 |
kens | mlilja I keep covering this point. THe pdfwrite device is *NOT* intended as a compression, optimising or 'cleaning' utility | 08:03.41 |
| There is absolutely no guarantee that a PDF file will be smaller after passing through interpretation and re-creation than it was before, and may well get bigger | 08:04.12 |
| On average the output files in our test suite are the same size or smaller, but some get bigger (THOUGH NOTHIGN LIKE 100 TIMES) | 08:04.59 |
| The command line parameters you choose will obviously have an influence as well, of course. | 08:05.27 |
chrisl | kens: when you have a minute, can you fire up your fedora VM and do "make --version" and let me know what version of make you've got, please? | 08:11.04 |
kens | OK one minute chrisl | 08:11.24 |
chrisl | Ta | 08:11.34 |
mlilja | ok, we are running the script for some newspaper-pages in pdf, usually it compresses about 60-70%, but sometimes there are some pages that get many times bigger instead....... here are the parameters we use: | 08:12.19 |
| gs -sDEVICE=pdfwrite -sBATCH -dNOPAUSE -dCompatibilityLevel=1.4 -dColorConversionStrategy=/sRGB -dProcessColorModel=/DeviceRGB -dUseCIEColor=true -sOutputFile=JP-37-130125-T-01-01-CMYK_compressed.pdf strulfiler/JP-37-130125-T-01-01-CMYK.pdf | 08:12.30 |
kens | Well using -dUseCIECOlors is a bad place to start | 08:12.49 |
mlilja | ok? | 08:13.05 |
kens | THat will cause everything to be converted into a CIE color space. As a result a number of grpahics primitives won't be possible to handle in that space, and will be decomposed ot images, or tiny filled rectangles | 08:13.35 |
kens2 | I think the bad weather is affecting my broadband.... | 08:15.06 |
| chrisl make --version gives me GNU Make 3.82 | 08:15.22 |
chrisl | kens2: thanks - we have a GNU make bug!! | 08:15.58 |
kens2 | mlilja I'd suggest you don't use -dUseCIEColor to start with. FOr anythign more I would need to see an example file. | 08:16.21 |
| Oh joy.... | 08:16.25 |
| Has it beenfixed ? | 08:16.33 |
chrisl | No, 3.82 is bleeding edge - I need to trawl through the GNU make bug tracker, but god knows how this would be reported..... | 08:17.26 |
mlilja | I removed the -dUseCIEColor but it still gets much bigger | 08:17.42 |
kens2 | I'm more than a bit surprised that I'm using a bleeding edge make on Fedora 14. | 08:17.52 |
chrisl | Well, 3.82 is the latest release - it may have been around for a while | 08:18.39 |
kens2 | mlilja well as I said there is NO guarantee that a file will get smaller. I'd need to see the file to tell you why its bigger. | 08:18.58 |
chrisl | Yes, since July 2010..... | 08:19.05 |
kens2 | chrisl yes, I thought iot might be 'stable' | 08:19.22 |
mlilja | Please tell me how can I send you the file? | 08:19.44 |
kens2 | mlilja I'm assuming you don't work for an Artifex customer ? | 08:19.53 |
mlilja | nop | 08:20.02 |
kens2 | Well, assuming the file is a sensible size you can send it to ken.sharp@artifex.com or make it available on a public URL. Please be aware that the GPL version of Ghostscript is supplied without warranty or support. | 08:20.46 |
mlilja | the filesize is about 1,5Mb, ok to email that? | 08:24.41 |
kens2 | Yes, that's OK, tens or hundreds of megabytes would be bad ;-) | 08:25.00 |
chrisl | kens2: Basically, in the "so" target we use recursive make calls, in this case, in the final one which builds and links the executables, make is misidentifying one file (pasbase.dev) as having changed, so it decides it needs to rebuild the shared library object - but, by this time, the compiler options we've passed in to the recursive make call are for building an executable, not a shared lib........ | 08:30.35 |
kens2 | Oh, oops, that's not good | 08:31.06 |
| <sigh> 3 emails from Aaron this morning asking for more help in coding, and one from Nicole (same company) wanting me to explain why somebody else's tool screws up. | 08:32.03 |
chrisl | Tell them to stuff it..... | 08:32.37 |
kens2 | mlilja I've received your file, I'll look at it when I've finished my morning email. | 08:32.40 |
| chrisl fortunately its a fairly simple question, so I'm going to answer it just for kudos. | 08:32.59 |
chrisl | tsk - setting a precedent.... | 08:33.21 |
kens2 | I think I've been there already..... | 08:33.40 |
chrisl | Right, I think I have a workaround/fix for the make bug | 08:34.22 |
kens2 | THat would be cool | 08:34.34 |
chrisl | I should get around to fixing the warnings in dxmain.c at some point.... | 08:36.21 |
mlilja | thanks kens2 | 08:40.24 |
chrisl | is highly unimpressed with GNOME 3..... | 08:41.52 |
kens2 | mlilja your file contains two shading dictionaries which cover apparently small areas. These are defined in a CMYK colour space and use , one radial, one axail. Because you ahve set the ColorConversionStrategy to /sRGB the functions generating the colours are inappropriate. [more] | 09:34.36 |
| They both use type 3 stitching functions which further rely on type 0 sampled functions. | 09:35.18 |
| These are 16 bit 1-in 4-out functions. In order to have a sRGB coplour space these have to be 1-in 3-out in a totally different colour space. | 09:36.20 |
| The conversion of the functions is difficult, and probably impossible in the general case, so we don't attempt it. Instead we fall back to generating an equivalent. | 09:37.05 |
| In this case, rather than generate an image, we create millions of filled rectangles. | 09:37.27 |
| WHich is why your output size *increases* | 09:37.36 |
| At present there isn't anything to be done about this. If it becomes a problem for a customer, or a serious issue for many free users, we may revisit it, but fundamentally, teh output is 'equivalent' to the input (given that you have changed the colour space). Shading dictionaries are fairly uncommon, and ones that cna't be easily converted are more uncommon still. | 09:40.00 |
| You could reduce the resolution (default is 720 dpi) which would reduce the problem, buyt won't solve it. At 72 dpi I still get a 3.6 Mb file. | 09:40.32 |
| But if I 'LeaveColorUnchaged' I get a 640kb file, ledss than half teh size of the original. | 09:41.04 |
| FWIW the shading seems to be the 'stepped' red at the top and left of the advert, and teh curved area where it says '53 03 Jonkoping 0770 77 87 87' near the bottom and the right. | 09:42.43 |
| Although the visible areas are small, the shadings actually extend over most of the page under the actual advert, and this is what causes the outpt to be so large. Had thos shadings been drawn with a clip to the final size the output would be a more reasonable size. Poor design by the designer. | 09:43.50 |
mlilja | ok, thanks for the analyze, I will try to understand it. Is there a way to find if the pdf "contains" shading and if so change to 'LeaveColorUnchaged' in that case? | 09:45.18 |
kens2 | Well you can check pages for a Shading resource, I'm not at all sure if we have any tools to report it. | 09:45.48 |
| By the way, the code later hits a colour space it *can't* convert, and reverts to LeaveColorUnchanged anywa, but by that time its already done the shadings, whic come eaarly. | 09:46.53 |
| So you're actually getting the worst possible output by doing this. | 09:47.05 |
sebras | paulgardiner: morning paul! | 10:09.22 |
paulgardiner | Hi sebras | 10:09.29 |
sebras | paulgardiner: did you read the logs? | 10:09.33 |
| paulgardiner: I made the app totally crazy last nigt. :) | 10:09.56 |
| paulgardiner: but maybe you are working on something else nowadays. | 10:10.09 |
paulgardiner | Still working on it, although mostly on annotation creation at the moment. | 10:11.32 |
| smalller.mp4 and original.mp4? | 10:13.21 |
sebras | paulgardiner: yes. in ~sebras/tmp | 10:13.39 |
| tor8: good evening! | 10:23.13 |
chrisl | plauto: ping | 10:25.00 |
plauto | chrisl: hi | 10:25.20 |
| i've seen the commit | 10:25.26 |
| trying it right now | 10:25.31 |
chrisl | plauto: Oh, good. It works for me - note there are still compiler warnings from the gtk stuff...... | 10:25.57 |
paulgardiner | Do I need to look at original.mp4 or is smaller.mp4 a lower quality version of it? | 10:26.52 |
sebras | paulgardiner: smaller.mp4 is enough to start with. | 10:28.54 |
paulgardiner | Drag select. Don't let go. And tap at the other side of the selection with second finger? | 10:30.34 |
plauto | chrisl: works good on both centos 6 and fedora 17 | 10:43.07 |
| thanks | 10:43.09 |
chrisl | plauto: cool - congratulations on finding the biggest build headache I've had in quite a while ;-) | 10:43.41 |
plauto | :-D | 10:44.26 |
| my pleasure | 10:44.29 |
chrisl | Feel free not to find any more :-) | 10:44.47 |
sebras | paulgardiner: or in rjw's terms "tap like a maniac at the side of the screen". | 11:41.17 |
| paulgardiner: it seems to be correlated to the bottom of the last line of the selection. | 11:41.42 |
paulgardiner | sebras: So far unable to repeat it. | 11:44.01 |
sebras | paulgardiner: :-/ | 11:44.09 |
paulgardiner | In the middle of some annotations-related coding at the moment. I'll try again when I've finsihed | 11:44.51 |
sebras | paulgardiner: np. | 11:46.03 |
Robin_Watts | My car is being a bastard. I have a garage coming out to try to get it started, but I don't know when they will arrive. | 14:27.06 |
| If I'm not here for one of the meetings, that'll be why. | 14:27.14 |
kens | I'll mention iti if required | 14:27.51 |
Robin_Watts | Thanks. | 14:28.01 |
| I think I've also caught my wifes stinking cold :( | 14:28.13 |
kens | I got lucky on that one and avoided my wife's stinking cold | 14:28.34 |
| chrisl, new buld enhancement bug.... | 14:47.11 |
chrisl | kens: I see it, I didn't realise the lcms2 API had changed.... | 14:47.52 |
kens | Me neither, but what do I know.... | 14:48.38 |
Robin_Watts | mupdf cluster tests are nearing 5 minutes. Need to optimise it a bit :) | 14:59.11 |
henrys | chrisl:why the forced push? | 15:19.13 |
chrisl | henrys: I forgot the bug # | 15:19.25 |
henrys | oh right | 15:19.35 |
Robin_Watts | car towed down the road - car started. Phew. | 15:30.56 |
| cluster is unhappy. | 15:31.45 |
chrisl | Maybe my forced push confused it? | 15:32.00 |
Robin_Watts | no, this doesn't look like that sort of failure. | 15:32.21 |
| Looks like either network problems or client timeouts. | 15:32.39 |
chrisl | Robin_Watts: I have a huge battery booster for getting recalcitrant cars started - although, I think I went overboard, 'cause I reckon it could start a tank engine (and weighs a ton!). | 15:37.27 |
Robin_Watts | chrisl: I have one of those too. | 15:37.39 |
| And since yesterday I have a battery charger. | 15:37.49 |
| The problem is with the rotary engine in the mazda, if you get unburnt fuel in the engine, it sits there, and the seals don't work, and so you never get compression enough to burn the fuel off. | 15:38.36 |
| You need to tow the car to turn the engine over a bit, and then it'll start. | 15:38.51 |
henrys | don't you just jump from other cars? | 15:38.53 |
Robin_Watts | henrys: I tried that too. I have tyre marks on my lawn from where I had to put the van. | 15:39.17 |
chrisl | With the MX-5 battery being about the size of a PP3, I've had a charger for years - I got the booster pack when I joined Artifex, and am leaving the car at the airport... | 15:39.27 |
Robin_Watts | chrisl: sensible. | 15:39.46 |
| Most airport car parks have a battery pack on a trolley for just this reason. | 15:40.06 |
chrisl | henrys: if you've already flattened one battery trying to start a car, using jump leads is just likely to flatten the other (or burn out the alternator!) | 15:40.21 |
Robin_Watts | chrisl: Best plan is use the other car to charge the flat one via the jump leads. But none of this helps when the problem is a flooded engine :( | 15:41.18 |
henrys | well yeah - at the airport the maintenance folks are driving around with some heavy duty stuff - they probably jump several cars a day. No wonder the parking raters are so high. | 15:41.31 |
Robin_Watts | In general the mazda is fine as long as we remember to always drive it for at least a couple of minutes (never just move it off the driveway). | 15:41.57 |
| And, given the propensity of the RX8 to need towing, guess where Mazda put the towing hook? Behind the numberplate! | 15:42.54 |
chrisl | Robin_Watts: At least tne of the aftermarket ECUs for the RX-7 had a feature where, if you held the accelerator flat to the floor when the ECU powered up and kept it there, and turned the motor over, it wouldn't fire the injectors, effectively just pumping air through - so that would evaporate condensed fuel, if the battery could handle it. | 15:43.25 |
Robin_Watts | chrisl: I believe the RX8 has that as standard (certainly the forums seem to suggest it). | 15:43.47 |
| It's never worked for me though. | 15:43.52 |
chrisl | My old ECU for the MX-5 had the same thing, but for a different reason..... | 15:44.15 |
henrys | tor8:here's another argument you can give Miles to get out of a meeting: Artifex is a bad citizen: http://www.nytimes.com/2013/01/27/sunday-review/the-biggest-carbon-sin-air-travel.html | 15:45.03 |
| I was surprised how significant air travel is. | 15:45.49 |
tor8 | oh, air travel is huge in terms of pollution | 15:46.09 |
Robin_Watts | On the otherhand, air travel is WAY better than if we all drove to the meetings. | 15:48.12 |
chrisl | My car would sink....... | 15:48.31 |
kens | SOmeon owuld need to gtive me a lift..... | 15:49.01 |
henrys | nuclear submarine is a possibility | 15:49.44 |
tor8 | can't we just upload our brains to some cloud server? ;) | 15:50.34 |
Robin_Watts | A small one should suffice. | 15:50.45 |
| </hhgttg> | 15:50.52 |
chrisl | tor8: interesting time to suggest that, just the cluster appears to having issues! | 15:51.13 |
henrys | if we set up a good video conferencing meeting once a week dragged miles to it once it was working well - after he came every week for a while his needs for a quarterly might lessen. Something to think about. | 15:57.12 |
Robin_Watts | henrys: I have severe doubts that those of us on ADSL could realistically cope with a video conference with 12 participants. | 15:58.05 |
sebras | Robin_Watts: isn't that an argument for stepping up to a notch in ADSL speed..? | 15:58.59 |
Robin_Watts | sebras: Unless Artifex wants to pay to upgrade the local exchange, I'm maxed out :) | 15:59.20 |
| 8Mbps down, 448kbps up. | 15:59.39 |
henrys | I think I'd be close I have about 20 mbs. I know torn would be okay | 15:59.46 |
| s/torn/tor8 | 15:59.57 |
| but I can up my speed to business class. | 16:00.37 |
sebras | Robin_Watts: I guess it all comes down to what size/quality you want too... most of the time you probably only want one of those videos in high-res anyway. | 16:00.59 |
Robin_Watts | 6.61/0.37 measured, which is about right. | 16:01.04 |
henrys | okay so meeting time - but it looks like paulgardiner is driving on with annotations and there isn't much to discuss. | 16:01.15 |
Robin_Watts | henrys: The first update to mupdf has made it to google play. | 16:01.42 |
paulgardiner | I'm over most of the problems I mentioned in the report, but I've run into a new one. | 16:01.50 |
henrys | Robin_Watts: awesome | 16:02.08 |
Robin_Watts | And we're at about 118 active installs :) | 16:02.31 |
paulgardiner | Nothing serious though. Just need to make the partial-update mechanism handle new and deleted annotations. Currently it handles only altered ones. | 16:02.53 |
henrys | I was very interested in translation - in particular asian languages. | 16:03.08 |
Robin_Watts | paulgardiner: Ah. So when you delete an annotation the system doesn't know to redraw that bit of the screen? | 16:03.24 |
henrys | I don't know if paulgardiner is on the everybody list and received Robin_Watts mail | 16:03.43 |
Robin_Watts | henrys: You mean the email I forwarded... right. | 16:03.51 |
| I will send to Paul. | 16:03.58 |
paulgardiner | Robin_Watts: nor to some degree when I create one. | 16:04.19 |
| Robin_Watts: It's all to do with the fact that we cache annotations in the page object. | 16:04.41 |
| henrys: other-language versions of the android app may not be too hard because all strings are already indirected through a strings file. | 16:06.00 |
Robin_Watts | I think the email was primarily for the listing on google play. | 16:06.40 |
paulgardiner | ah right | 16:06.56 |
Robin_Watts | but actually, you may be right. | 16:07.11 |
sebras | paulgardiner: how does an android app choose what language to use? | 16:07.54 |
henrys | you select your language when you set up your device - well I did so I assume there must be some environment setting to query right? | 16:08.47 |
tor8 | do we really want the burden of supporting non-english speakers? | 16:08.48 |
paulgardiner | sebras: don't actually know, but I imagine you provide multiple string files and it's fairly automatic. | 16:09.02 |
Robin_Watts | paulgardiner: I suspect I'd need to build and upload multiple apks. | 16:09.25 |
henrys | tor8:do I really want to burden the largest potential markets in the world - yea I think so. | 16:09.35 |
paulgardiner | Robin_Watts: I'd hope not, but you may well be correct | 16:09.50 |
Robin_Watts | henrys: What is the aim of us having MuPDF on the app store? | 16:10.17 |
henrys | tor8:seriously I don't think it would that much maintenance - famous last words | 16:10.25 |
sebras | paulgardiner: I was thinking of run-time. I assume the all translations are embedded in the same apk. | 16:10.52 |
Robin_Watts | My understanding was that it was there so we could point potential customers at it and say "look, here is how well it works" | 16:10.56 |
henrys | Robin_Watts:advertisement and mind share of pdf technology | 16:11.01 |
tor8 | henrys: I'm more worried about people coming to us and asking support questions in a language we're not prepared to handle. but I've heard rumors kens knows chinese, so we should be covered ;) | 16:11.08 |
kens | ROFL | 16:11.28 |
Robin_Watts | Right, so do we care that we are limiting ourselves to customers that can speak at least pigeon english ? | 16:11.33 |
sebras | tor8: now I have more reasons to study chinese. ;) | 16:11.37 |
tor8 | adding translations for the major languages to the apps should be easy enough, but we'll probably need to translate the info/advertisement thingy on the market pages too | 16:11.54 |
Robin_Watts | tor8: The email includes that. | 16:12.15 |
| It's a pain that presumably we'll have to pay again every time we update the text. | 16:12.30 |
paulgardiner | Is that email from someone who offers translation services? | 16:12.31 |
Robin_Watts | yes. | 16:12.35 |
henrys | I think any marketing person is going to tell you this a no brainer ... | 16:13.07 |
Robin_Watts | Well, if Miles is happy to pay, we can certainly pursue it. | 16:13.23 |
tor8 | Robin_Watts: maybe we ought to perfect the text and messages we'll need translating first though. | 16:13.43 |
Robin_Watts | tor8: indeed. | 16:14.01 |
paulgardiner | just one concern: doing so early might create an overhead on each new string we add in the future. | 16:14.16 |
Robin_Watts | supposedly they only charge for "new" or changed strings. | 16:14.38 |
henrys | we should also shop around right? It may even be possible to get a volunteer. | 16:15.05 |
paulgardiner | We might be able to just add English words in all languages each time and then have them translated in batches | 16:15.07 |
henrys | what are we saying kens can do it ;-) | 16:16.09 |
kens | I only write Pinyin | 16:16.23 |
| Not simplified or traditional | 16:16.31 |
Robin_Watts | We can probably avoid paying for UK english and swedish. :) | 16:16.57 |
henrys | I speak some texan | 16:17.19 |
paulgardiner | :-) | 16:17.40 |
| I could do Somerset | 16:17.52 |
Robin_Watts | Git yer MuPDF here! Faster'n a rattlesnake up a drainpipe. | 16:17.59 |
| Population of Somerset that own smartphones? :) | 16:18.11 |
| Sorry, population of Somerset that own a smartphone and can use it enough to download an app? >8*) | 16:18.39 |
henrys | does anyone know the most common route for open source apps to add this kind of stuff? | 16:19.06 |
paulgardiner | Yeah, that's a point. But it's so simple s/er/eerr/ | 16:19.11 |
henrys | I've never really looked into it. | 16:19.44 |
Robin_Watts | henrys: They wait for volunteers, I think. | 16:19.56 |
sebras | henrys: there is a translators project for the GNU apps. volunteer-based. | 16:20.09 |
| henrys: http://translationproject.org/html/welcome.html | 16:20.41 |
henrys | so we would have to get a volunteer who would sign our cla | 16:20.43 |
| | 16:20.48 |
Robin_Watts | For those of you with Android devices, you might want to download the new apk and experiment with tapping the edges of the screen to move around. In particular try zooming in and doing it (if you have a file with columns of text, so much the better). | 16:21.22 |
henrys | Robin_Watts: will do | 16:21.47 |
Robin_Watts | If people like the way it behaves, maybe we should look at implementing that on iOS and/or the desktop versions before we ship the next release. | 16:22.00 |
henrys | sebras:I like the translation project | 16:22.31 |
Robin_Watts | tor8 may now proceed to scream at me. I have no idea how hard it would be to do with the way the iOS app is coded. | 16:22.48 |
henrys | I wonder how active it is? | 16:22.49 |
sebras | henrys: I'm not sure. however I guess one could find the necessary keywords for "cancel" "open" "copy" and so on there... | 16:23.28 |
tor8 | Robin_Watts: aaaarrrrrhhhh! actually, I hope it shouldn't be too tricky to reimplement. | 16:23.43 |
| I believe the android and ios both use the same nested scroll views approach? | 16:24.04 |
henrys | wait the translation page has a robot the TP-robot | 16:24.17 |
Robin_Watts | tor8: The android app doesn't have a nested scroll view. | 16:24.20 |
tor8 | Robin_Watts: ah. right. | 16:24.31 |
Robin_Watts | It has a single 'ReaderView', where ReaderView is a custom class what paul wrote. | 16:24.41 |
tor8 | Robin_Watts: so the android zoom view zooms *all* pages in the big canvas when zooming, whereas the ios just zooms the currently visible one | 16:25.09 |
sebras | Robin_Watts: could it be this new smart scroll that I accidentally triggered somehow in ~sebras/smaller.mp4? | 16:25.17 |
Robin_Watts | tor8: Yes, where "all" is "all 3". | 16:25.37 |
| sebras: It could be. | 16:25.42 |
kens | coffees | 16:25.47 |
tor8 | nested scroll views is the "recommended" way to do paged views with zooming on iOS. anything else is ridiculously difficult to implement. | 16:25.52 |
paulgardiner | sebras: that would explain why I couldn't repeat it. I'm working further down master. | 16:26.06 |
Robin_Watts | sebras: If your crashing involves clicking on the edge of the screen, they quite possibly. | 16:26.11 |
| s/they/then/ | 16:26.17 |
henrys | 5 minutes until the half hour - paulgardiner are you all good for another week of work, anything you need? | 16:26.19 |
tor8 | even getting that nested view thing to work and re-render the images at the correct resolution was a lot of research and guess work... | 16:26.24 |
sebras | Robin_Watts: it does. | 16:26.26 |
paulgardiner | Nope. Fine thanks Henrys | 16:26.45 |
henrys | zebras would you be interested in studying the translation issues on a consulting basis? | 16:26.56 |
Robin_Watts | paulgardiner: So maybe we just disable edge clicking in selection mode? | 16:27.04 |
henrys | sebras your not a zebra | 16:27.14 |
| I must stop colloquy from changing my words | 16:27.32 |
paulgardiner | Robin_Watts: assuming that's what's causing it. May well not be. | 16:27.45 |
sebras | henrys: no, I'm not, I'm a goat. :) I'll take a quick look at how it works in android. | 16:28.27 |
paulgardiner | sebras: waa tapping with one finger while another was already down. Not sure how that comes through the gesture recognsiers | 16:28.41 |
sebras | Robin_Watts: what about zooming in selection mode? | 16:29.13 |
| Robin_Watts: the problem is that once i clicked with the second finger then mupdf stayed in this whacky mode even while I was only scrolling with a single finger... | 16:29.50 |
paulgardiner | sebras: any exceptions thrown? | 16:31.07 |
sebras | paulgardiner: not that I see. | 16:31.19 |
| paulgardiner: logcat was clean. | 16:31.31 |
henrys | the nexus 7 needs a little charging then I'll download... | 16:31.59 |
Robin_Watts | tor8: Can I call mudraw on an exploded xps file ? | 16:36.09 |
tor8 | Robin_Watts: point it at the _rels/.rels file | 16:36.21 |
Robin_Watts | Fab, thanks. | 16:36.43 |
tor8 | that works in gxps too | 16:37.15 |
kens | My network keeps falling over... :-( | 16:53.27 |
henrys | so our release schedule was sort of set up to better serve ubuntu - now ubuntu is changing: http://www.networkworld.com/news/2012/103112-ubuntu-263863.html | 16:56.19 |
marcosw | chrisl: can I ask you about release candidate regression testing? | 16:57.30 |
kens | I'm not sure that really affects us, except that their freeeze takes place later | 16:57.40 |
chrisl | marcosw: ask away | 16:57.41 |
marcosw | so what do I want to do to the ghostpdl directories so that the weekly regression tests run the gs907 branch? | 16:58.41 |
chrisl | marcosw: "git checkout gs907" | 16:59.10 |
| henrys: my understanding is that their release schedule isn't changing, just the process building up to the release is being "streamlined" - or have I got that wrong? | 16:59.51 |
kens | chrisl I believe you are right | 17:00.15 |
| So their freeze will be later | 17:00.28 |
chrisl | So unless we *want* to change our release schedule....... | 17:00.51 |
henrys | yes that is my read too. | 17:02.15 |
| I assume tkamppeter will say something if we could do something easily that would be useful for the new setup | 17:03.11 |
| that was the beginning of the meeting I suppose | 17:03.28 |
| so we have 1 blocker | 17:03.43 |
| ray_laptop:do you have a schedule? | 17:06.02 |
| we could probably ship with that problem. | 17:06.19 |
| I'm all released focused and didn't have other topics for the meeting? Anybody else? | 17:07.02 |
kens | Bot right now | 17:07.14 |
alexcher | gs still has a bunch of SEGV-causing bugs. | 17:07.59 |
mvrhel_laptop | I have a question not released releated | 17:08.22 |
henrys | alexcher:regressions? | 17:08.33 |
ray_laptop | henrys: I see what's going on with the psdcmyk ones, Looking for 'why' so I can do a clean fix (i.e. low risk) | 17:08.41 |
henrys | mvrhel_laptop: shoot | 17:09.07 |
chrisl | alexcher: where are these segv's reported? I'm not seeing any on the cluster tests | 17:09.41 |
alexcher | henrys: it's hard to say. Some look like old regressions. | 17:09.43 |
marcosw | henrys: if we want to regression test the gs907 candidate with all of weekly tests I'll need 7 days, starting from tomorrow. | 17:10.00 |
henrys | a segv that is a regression should be marked blocker - everyone should do that right? | 17:10.22 |
alexcher | 693365, 693398 | 17:10.39 |
henrys | marcosw:are we going to count from when ray commits or are we going to test and assume ray_laptop's fix is clean after review? | 17:11.07 |
mvrhel_laptop | so the customer with no number asked me "if it is possible to switch ICC profiles in C code" as he would like to change the ICC profiles in a job page by page. In the context of this, he is talking about having a system parameter change (in this case SourceObjectICC) | 17:11.14 |
henrys | we got a number yesterday | 17:11.35 |
mvrhel_laptop | It is not clear to me how they are invoking gs | 17:11.36 |
alexcher | 693561 is a regression | 17:12.12 |
mvrhel_laptop | so I am not sure how best to reply to this question | 17:12.19 |
marcosw | I'm going to test the gs907 branch as it exists today. If any changes occur to that branch the clock will reset. | 17:12.27 |
henrys | mvrhel_laptop: they can set user and device parameters in c code right? | 17:12.31 |
ray_laptop | mvrhel_laptop: they can either send PS "setpagedevice" param lists, or call gs_putdeviceparams | 17:12.32 |
mvrhel_laptop | right to both of these | 17:12.50 |
| ok. | 17:12.59 |
ray_laptop | mvrhel_laptop: but we also have a 'C' call interface, right ? (for PCL to use) | 17:13.09 |
henrys | marcosw:fair enough | 17:13.18 |
mvrhel_laptop | ray_laptop: let me check that I did this one. | 17:13.44 |
henrys | ray_laptop: mvrhel_laptop did that I added the calls to pcl. | 17:14.29 |
mvrhel_laptop | yes. it is set up like the others | 17:14.30 |
ray_laptop | mvrhel_laptop: calls to gsicc_set_profile with the profile type identified by the enum | 17:15.11 |
tkamppeter | henrys, Feature Freeze for Ubuntu Raring (13.04) is March 7. See https://wiki.ubuntu.com/RaringRingtail/ReleaseSchedule. | 17:15.30 |
marcosw | chrisl: as a test I did a 'git checkout gs907' and then a git fetch and now see this when I do a git status: | 17:15.36 |
| # On branch gs907 | 17:15.37 |
| # Your branch and 'origin/gs907' have diverged, | 17:15.38 |
| # and have 4 and 3 different commits each, respectively. | 17:15.38 |
henrys | fwiw something I did this week is run coverage on my pieces this found a bunch of stuff I should remove and stuff that needs testing. Probably a worthwhile exercise for all of us. | 17:15.38 |
chrisl | marcosw: you probably wanted to fetch first...... git pull --rebase ? | 17:16.50 |
marcosw | I did. | 17:17.03 |
henrys | I thought of that because my coverage testing revealed we don't test the icc calls - although I did test them once. | 17:17.22 |
marcosw | git fetch ; git pull --rebase ; git checkout gs907 ; git fetch ; git reset --hard -q master ; git reset --hard -q origin/master ; git status | 17:18.32 |
henrys | tkamppeter: I'm sure we'll have something out before that. | 17:18.45 |
marcosw | trying again, just to make sure I didn't screw anything up. | 17:18.58 |
chrisl | Erm, why are you resetting to master? | 17:19.25 |
marcosw | that's what the regression test script does. can't recall why that's in there, probably added a long time ago because of some git weirdness. | 17:20.02 |
chrisl | So that's the problem - you've checked out the gs907 and then done a reset to the master branch | 17:20.41 |
marcosw | and why have the gs907 and master branches diverged by 4 and 3 different commits? | 17:21.30 |
chrisl | Because there are different commits on them | 17:21.52 |
marcosw | already? When did you tag 907? | 17:22.13 |
chrisl | This morning | 17:22.20 |
| my time | 17:22.25 |
henrys | kens:alexcher blocker is your bug. | 17:23.05 |
chrisl | Not least is the fact the master now has version 9.08 in it, but a few others have happened, too. | 17:23.17 |
| marcosw: ^^ | 17:23.23 |
kens | henrys, which one ? | 17:24.04 |
henrys | http://bugs.ghostscript.com/show_bug.cgi?id=693561 | 17:24.22 |
marcosw | so I will modify the scripts to not do the git reset command until the gs907 release. Anything else? | 17:24.26 |
| chrisl: ^^ | 17:24.31 |
kens | henrys I would not class that as a blocker | 17:24.48 |
| Its not new | 17:24.52 |
henrys | okay but a crash does deserve some attention | 17:25.12 |
kens | In fact its nearly 2 years old | 17:25.13 |
chrisl | marcosw: you can still do the reset, but you need to reset to the branch you want to use, not always to master | 17:25.14 |
kens | henrys yes, agree totally, but I want to finish what I'm doing now | 17:25.29 |
henrys | okay can we up it to p2 so it doesn't fall through? | 17:25.53 |
kens | Sure, let me do that | 17:26.01 |
henrys | thanks alexcher | 17:26.14 |
kens | Don't know why it was at P4 | 17:26.15 |
henrys | almost to the 1/2 hour. Anything else? | 17:27.04 |
chrisl | So, what are we doing with the segv's? | 17:27.30 |
kens | For now, ignore the #693561, I will fix it but I can't see a 2 year old change as a blocker | 17:28.02 |
henrys | chrisl:any sego should be P2 or more do you think any should be blockers? | 17:28.41 |
| s/sego/segv | 17:28.49 |
alexcher | SEGVs are different but, IMHO, should be at least P3. | 17:29.48 |
henrys | IMHO a sego regression wrt 9.07 should block the release segv regression wrt 8.71 is differen | 17:29.55 |
chrisl | The two alexcher mentioned above are new to me, I haven't looked at them at all | 17:30.25 |
henrys | I don't know colloquy insists it's a sego not a segv | 17:30.27 |
chrisl | maybe a "sego" isn't a blocker ;-) | 17:30.47 |
henrys | there you go - ship it! | 17:31.12 |
mvrhel_laptop | damn it | 17:31.49 |
| email sent before I was done typing! | 17:31.57 |
| i hit some wacky keyboard shortcut | 17:32.08 |
| sigh | 17:32.13 |
chrisl | Ugh, days before a release is not a good time to bring up a segv in the garbage collector...... :-( | 17:33.19 |
marcosw | mvrhel_laptop: don't you have "send undo" enabled in gmail? saved me many a time (along with "did you forget the attachment?"). | 17:33.38 |
mvrhel_laptop | marcosw: I don't know about that. I have all my email getting popped to the windoze email tool on my machine | 17:34.36 |
ray_laptop | alexcher: bug 693304 doesn't say how to reproduce it. It is OK if I just run with -sDEVICE=ppmraw -o nul: -Z: -q Bug688528.pdf | 17:35.12 |
henrys | mvrhel_laptop: easy to fix. | 17:35.55 |
| it's when that first draft expresses how you really feel⦠that's when it gets ugly ;-( | 17:36.38 |
alexcher | ray_lapop: probably, you can just make this change anr run regression. | 17:36.44 |
ray_laptop | my email tool (Thunderbird) reminds me about attachments | 17:36.52 |
mvrhel_laptop | henrys: yes :) | 17:36.57 |
| Luckily it was just a 1/2 of a technical response | 17:37.13 |
chrisl | ray_laptop: I don't think 693304 is a concern for the release, though | 17:37.21 |
ray_laptop | alexcher: OK. let me try... | 17:37.26 |
| chrisl: correct. It isn't in the code. | 17:38.01 |
henrys | so marcosw you need 7 days starting at any day right? | 17:38.15 |
kens | Hmm do we have the SMask fix in the code ? I don't recall seeing that commit | 17:38.20 |
chrisl | I don't believe we do | 17:38.34 |
alexcher | kens: not yet. | 17:38.41 |
kens | Oh, I thought we wanted that for the release, oh well | 17:38.55 |
Robin_Watts | kens: You mean 693115 ? | 17:39.03 |
marcosw | henrys: yes. tomorrow being the earliest possible, since today's regression tests have already started. | 17:39.20 |
kens | Robin_Watts : yes | 17:39.22 |
henrys | yes can't always get what you want | 17:39.26 |
Robin_Watts | So spake the poet Jagger. | 17:39.42 |
chrisl | alexcher: shouldn't 693365 go to mvrhel_laptop or possibly ray_laptop? | 17:40.18 |
marcosw | ironically he probably does alway get what he wants. | 17:40.21 |
kens | He didn't die before he got old though | 17:40.44 |
alexcher | chrisl: probably, | 17:41.05 |
henrys | I think that was a different poet | 17:41.08 |
chrisl | Roger Daltry | 17:41.27 |
kens | Indeed | 17:41.31 |
Robin_Watts | That was Townsend or Daltry. | 17:41.32 |
kens | Daltry | 17:41.36 |
Robin_Watts | chrisl can type faster :) | 17:41.46 |
chrisl | Typing was okay, took me a bit to remember the name! | 17:42.12 |
mvrhel_laptop | 693365 should probably be dropped on me | 17:42.45 |
kens | OK I'm off, goodnight all | 17:42.49 |
henrys | okay so if I have this correct marcosw will set the 7 day clock in motion asap when ray fixes his blocker we reconvene and decide if we want to sneak it in at which point we restart the 7 day clock. | 17:42.56 |
| we are 45 minutes in so best to finish up. | 17:43.26 |
kens | I htought we had.... | 17:43.36 |
chrisl | This could make for a **long** release process :-( | 17:43.37 |
mvrhel_laptop | depends on how fast ray is | 17:44.03 |
henrys | maybe something to bring up at the meeting - marcosw where does 7 days come from? is that sped up with more CPU cycles? | 17:44.30 |
chrisl | And how "just one more fix"s we get..... | 17:44.31 |
mvrhel_laptop | yes. there is always that | 17:44.44 |
Robin_Watts | henrys: 7 days is presumably "we do 7 different overnight tests". | 17:45.02 |
| and marcosw wants to run the branch through all those overnight tests and get clean results before releasing. | 17:45.41 |
marcosw | it would be really nice if we didn't diverge master too much from the gs907 branch, makes sending out patches to customers much harder... | 17:45.45 |
henrys | Robin_Watts:if an overnight test takes 4 hours though - presumably we could do them all in 7 * 4 = 28 hours a bit more than a day. | 17:46.45 |
| just for the release, right? | 17:46.52 |
Robin_Watts | henrys: Assuming no one wanted to use the cluster in that time, sure. | 17:47.06 |
kens | That would hurt me | 17:47.16 |
chrisl | marcosw: the way to do that would be to cherry-pick from master onto the release branch, and then diff on the branch. The discard the cherry-picked commit from the branch | 17:47.20 |
ray_laptop | Are the nightly tests on the cluster, or marcos' machine | 17:47.47 |
kens | is impressed chrisl sounds like he knows what he's talking about | 17:47.53 |
henrys | well it could be lowest priority and we could still get things done a lot faster than the 7 days. | 17:48.10 |
| gitanese | 17:48.41 |
chrisl | kens: I speak from experience - I spend plenty of time pretending I know what I'm talking about ;-) | 17:48.41 |
Robin_Watts | chrisl: The smartest way to work might be for marcosw to create a new branch based on gs907 for each customer he has to release to. | 17:48.42 |
| Then he can cherry-pick onto that branch, and diff. Then he has a record of what patches he's sent to what customer. | 17:49.03 |
ray_laptop | Robin_Watts: yuck | 17:49.03 |
marcosw | Robin_Watts: yeah, I'm not doing that. | 17:49.20 |
chrisl | Robin_Watts: yes, but personally, I want to discourage the sending of random patches to customers, anyway! | 17:49.21 |
ray_laptop | so we create a branch for each customer ??? | 17:49.25 |
Robin_Watts | ray_laptop: Why yuck? It's no more work than what marcosw has to do already. | 17:49.34 |
kens | Yeay, *lots* of branches :-) | 17:49.37 |
ray_laptop | (at least those we send patched to) | 17:49.39 |
henrys | let's veto that early | 17:49.46 |
Robin_Watts | ray_laptop: He doesn't publish those to the main repo. | 17:49.46 |
| His local repo just has a few extra tags in it. I don't see the problem. | 17:50.11 |
henrys | I tank marcosw already did veto it. | 17:50.22 |
| s/tank/think | 17:50.35 |
ray_laptop | Robin_Watts: he could, right ? That way any of us working on a bug for a particular customer can checkout that customer's branch | 17:50.47 |
Robin_Watts | ray_laptop: Absolutely he could. | 17:51.01 |
marcosw | henrys: it takes more than 4 hours for the nightly/weekly regression tests. It's about 12 hours for the run and then another 0 to 30 hours for the fuzzy step (yes, if there are many bitmap changes the fuzzy step takes multiple weeks to run, which is why sometimes it takes me a while to find regressions). | 17:51.10 |
ray_laptop | We just have a tag for those customers that get patches (probably a couple dozen) tags like cust531 or cust330 | 17:51.32 |
Robin_Watts | but the smart way to do it would be for him to push those branches to his personal repo on casper, so we can pull from there if we want, but the main golden repo doesn't get polluted. | 17:51.36 |
| ray_laptop: Exactly. | 17:51.43 |
| I mean, suppose Gemma finds 5 different problems in a release. Does marcosw have to keep remembering what patches he's previously applied before he rebuilds for her? | 17:52.24 |
| git makes it trivial. | 17:52.36 |
henrys | marcosw:wow I din't know it was that intensive | 17:52.37 |
Robin_Watts | Anyway, I've suggested the idea, and if marcosw doesn't want to do it, then so be it. | 17:52.58 |
kens | OK I really do have to go, goodnight all | 17:53.04 |
Robin_Watts | night kens | 17:53.09 |
henrys | I have a quick errand to do and will be back shortly. | 17:53.51 |
marcosw | Robin_Watts: it's not necessary, it's very rare for a customer to need more than one patch between releases (Gemma being the only exception). But even they've slowed down; I think mvrhel_laptop has finally found all of the transparency/spot color/gradient/etc. combination bugs. | 17:54.01 |
Robin_Watts | marcosw: Oh, now you've done it... | 17:54.24 |
marcosw | henrys: it takes the cluster 30 minutes to run on 17 machines so that means 8 hours on one machine. But the cluster code don't save bitmaps, so that's faster. And the regression tests pause for cluster runs. | 17:56.38 |
alexcher | marcosw: you can run regressions on i7a and i7b boxes. | 17:58.36 |
marcosw | alexcher: doesn't help, I can't view the bitmaps at any reasonable speed. That's also the reason I don't run on the amazon cloud. | 17:59.19 |
henrys | marcosw:you need more computers in your house ;-) | 17:59.55 |
marcosw | In theory I could run the regressions on i7a and i7b or the amazon cloud and then re-run locally only those files that have changed⦠| 17:59.59 |
| henrys: I actually have 5 cluster nodes at the moment, but two of them are headless so don't work for the same reason. | 18:00.32 |
ray_laptop | mvrhel_laptop: are you there ?? | 18:01.22 |
mvrhel_laptop | ray_laptop: yes I am | 18:01.45 |
marcosw | I intended the extra two cluster nodes to move to miles' office, but his internet is so slow we already waste a minute or so on each cluster run waiting for his nodes to upload the results. | 18:03.16 |
| I could reduce this by having each node pre-analyze the log files and send less data, but occasionally having the full logs available on casper is useful; the real solution, which I keep putting off implementing, is to pre-analzye the data and send the complete logs later. | 18:03.38 |
| oops, it's 10:00, time to run to Uni. I'll be lurking the rest of the day, but mostly in meetings, so don't expect a prompt reply. | 18:04.33 |
ray_laptop | mvrhel_laptop: sorry. | 18:07.50 |
| mvrhel_laptop: so what I am seeing is that when setting up the thread's device, the device_profile->supports_devn is false. | 18:08.45 |
mvrhel_laptop_ | ok that is what you suspected yesterday ray_laptop | 18:09.06 |
ray_laptop | mvrhel_laptop: OK, so I can just copy that from the underlying device's icc_struct->supports_devn (into the rendering thread's device) | 18:09.55 |
mvrhel_laptop | I would think that would do it | 18:10.14 |
ray_laptop | mvrhel_laptop: OK. trying that... | 18:10.31 |
henrys | well talk to marcosw later but it seems we should be able to leverage more of the cluster to do this testing. | 18:13.00 |
ray_laptop | mvrhel_laptop: regression testing it. | 18:23.37 |
| bbiab... | 18:24.33 |
henrys | btw:certainly miles should get faster internet. Hard for me to believe San Rafael doesn't have fast internet options | 19:39.39 |
| marcosw ^^^ | 19:39.52 |
Robin_Watts | tor8: Urm... Can you explain: | 19:41.16 |
| fz_intersect_rect(s, &fz_infinite_rect); // FIXME: Eh? | 19:41.25 |
| in fz_bound_shade ? | 19:41.33 |
marcosw | henrys: I suggested that and Miles looked into it several months ago.There is a problem with upgrading that I can't quite remember, something to do with routers or NAT or something. I'll see if I can find the email and forward it to you. | 19:41.36 |
Robin_Watts | oh, no tor8. | 19:41.49 |
marcosw | henrys: btw, don't bet on San Rafael having fast internet. At least not where Miles' office is. E.g. he doesn't have cell phone reception. | 19:43.10 |
henrys | I thought there would be plenty of high tech there demanding such but maybe not | 19:43.56 |
| no fiber? | 19:44.13 |
marcosw | henrys: can't find an email from Miles with the details. Just me asking him about it in late september. I suspect we must have discussed it by phone | 19:44.32 |
| As I recall the only option was DSL. | 19:44.43 |
| at least at a reasonable price. | 19:45.04 |
henrys | maybe we need to rent space or something⦠I don't know about your situation but sabrina is fine with the current number of cpu's at my house. | 19:46.11 |
| ;-) | 19:46.22 |
Robin_Watts | henrys: Well, you have the worlds warmest basement now, right? | 19:47.11 |
henrys | well there is that | 19:48.00 |
marcosw | I keep putting off doing the math as to what a cluster node costs to run in terms of electricity (offset by reduced heating costs in winter but increased air-conditioning costs in summer). My electric bill is ~$300/month, but that includes electric car charging. | 20:05.07 |
| If I recall neither Miles nor rayjj pay extra for electricity at their offices :-) | 20:06.04 |
| never mind, I did the math back in 2010. At the time I calculated a cluster node at $31.95/month (assuming 53 commits and 149 clusterpushes per month, which was apparently the average back then). This is at the very high california electricity costs of $0.40/Kwh but doesn't includes running night/weekly regression tests. | 20:08.44 |
| Does this really mean the $300/month I'm paying for electricity is more than half for cluster nodes?!?! | 20:10.01 |
| Actually, I can check that. PGE (our power company) allows me to download detailed energy usage⦠(This is why I don't have my PhD yet, too easily distracted). | 20:11.08 |
| Oops, my electricity bill was is actually $472 last month, so that change the math a bit. | 20:12.33 |
alexcher | A node consumes about 100W on average, which corresponds to $7.5/month in PA. | 20:14.43 |
marcosw | alexcher: so you pay ~$0.10/Kwh? | 20:16.02 |
alexcher | yes | 20:16.11 |
marcosw | apparently oregon is the cheapest place for electricity, which is why so many data centers have opened/moved there. | 20:16.35 |
ray_laptop | anybody else having trouble getting to ghostscript.com ? | 20:17.03 |
marcosw | not me. | 20:17.18 |
alexcher | not me | 20:17.51 |
marcosw | yup, oregon rates are as low as $0.047/kWh. | 20:18.47 |
ray_laptop | what's the IP so I can try pinging it ? | 20:19.16 |
| pelase | 20:19.22 |
marcosw | 184.73.189.105 | 20:19.44 |
| is dns not working for you either? | 20:19.51 |
ray_laptop | apparently it's DNS | 20:20.15 |
henrys | .40/Kwh sounds very high | 20:20.18 |
ray_laptop | I can ping that fine | 20:20.29 |
marcosw | highest in the country, I believe. | 20:20.36 |
| while I was looking for the oregon rates I found a site that compared various energy costs, northern california was the highest of all the ones they showed (only slightly higher than southern california though). the numbers did not take into account the differences in climate, presumably southern california residents run their air conditioning more hours on average, so presumably they pay even more per month than we do, at least in the summ | 20:23.02 |
| I should disclaim that we have a strange rate, since we have an electric car our rate is based on the time of day (to encourage charging at night when there is extra capacity). from midnight to 8:00am we only pay ~$0.20/kWh. the $0.40/kWh number is during the afternoon in the summer. | 20:24.39 |
henrys | how many miles per kWh do you get? | 20:26.48 |
marcosw | I need to check the car for the actual number, but on average Leaf's get about 3.5 miles/kWh. | 20:30.04 |
henrys | I was thinking about an electric but until more of my power comes from something other than coal I think I'd more of a menace than benefit. | 20:31.56 |
marcosw | It's still supposed to be better for the planet than gasoline. | 20:32.42 |
ray_laptop | darn. dnsr1.sbcglobal.net AND a.resolvers.level3.net CAN'T resolve ghostscript.com :-( | 20:32.54 |
| what's your DNS (anybody!) ? | 20:33.12 |
| henrys: I would think your area would be getting wind farms going | 20:34.07 |
marcosw | henry: "The UCS finds that electric vehicles charged from the grid produce lower emissions than a petrol-powered car that gets 27mpg. And they did so, the report claims, even in places where the electricity is produced primarily from coal." | 20:34.08 |
| this is from the Economist: http://www.economist.com/blogs/babbage/2012/04/electric-cars | 20:34.22 |
Robin_Watts | ray_laptop: Set your resolver to use 8.8.8.8 | 20:34.41 |
marcosw | "Places that are best for charging electric vehiclesâCalifornia and parts of New Yorkâhave the lowest proportion of coal-fired capacity and the highest proportion of hydro and renewable sources in their generating mix." | 20:34.48 |
henrys | we do have some non trivial fraction coming from wind but still lots of coal. | 20:34.49 |
ray_laptop | Robin_Watts: I'll try... Thanks | 20:34.52 |
marcosw | sorry,didn't mean to bold that. | 20:34.53 |
Robin_Watts | So, electric cars beat hummers. and my car. but not much else. | 20:35.08 |
ray_laptop | Robin_Watts: that works | 20:35.13 |
Robin_Watts | ray_laptop: Google DNS :) | 20:35.26 |
henrys | you shouldn't need that though report a bug with your isp if they gave you a bad server | 20:36.22 |
Robin_Watts | My dads car (Citroen Zsara picasso thing, seats 7 or 8) gets upwards of 50mpg. | 20:36.38 |
marcosw_ | UCS == Union of Concerned Scientists | 20:36.38 |
ray_laptop | henrys: well, I tried 2 totally different servers (see above) 68.94.156.1 and 4.2.2.1 | 20:36.59 |
marcosw_ | Robin_Watts: and how often does he put 7 people in it? The Leaf gets the equiv of 104mp and seats 5 (and we often have 4 people in it). | 20:37.39 |
ray_laptop | Robin_Watts: WOW! I thought I was doing good with 28 mpg with my Lexus 400h SUV | 20:37.42 |
Robin_Watts | marcosw_: normally he's just driving himself, but sometimes he gets my sisters family in there. | 20:38.10 |
ray_laptop | I wish they would make a hybrid that uses LNG | 20:38.18 |
Robin_Watts | marcosw_: But to get that mileage, you probably need to drive like my dad :) | 20:38.43 |
ray_laptop | it should be a minor tweak | 20:38.51 |
marcosw_ | ray_laptop: honda makes a civic that uses CNG (but it's not a hybrid). | 20:39.16 |
ray_laptop | Robin_Watts: true. My neighbor has the same model, same year and gets 23 mpg | 20:39.17 |
| marcosw_: right. But a hybrid that runs CNG should be almost as good as electric, with better range | 20:40.06 |
marcosw_ | Robin_Watts: true, we only get about the equiv of 70 mpg out of the leaf. Since the car has such good regenerative breaking average speed is the biggest issue. | 20:40.32 |
Robin_Watts | I want to see the Williams Hybrid Power flywheel storage system make it into some road cars. | 20:41.24 |
marcosw_ | didn't some busses use flywheel storage? Helped with the turning as well :-) | 20:41.53 |
Robin_Watts | Currently it's in some porsche racers, so I think it's tough enough. | 20:41.54 |
| marcosw_: Presumably only when turning left? :) | 20:42.07 |
henrys | I thought there were reports of the leaf degrading mileage wise after only a few years. | 20:42.20 |
marcosw_ | http://en.wikipedia.org/wiki/Gyrobus | 20:42.53 |
| in 1955~ | 20:43.02 |
Robin_Watts | marcosw_: It's been done in busses, yes. The williams unit is way smaller though. Got to be better than a battery; charging/discharging batteries is never going to be good for their life. | 20:43.13 |
henrys | I want a fission powered car | 20:44.00 |
Robin_Watts | Top Gear reviewed the leaf, and didn't like it (predictably enough). Their biggest criticism was the lifespan of the batteries, and the fact that Nissan wouldn't say how much they'd cost to replace them. | 20:44.33 |
| henrys: fusion, surely? | 20:44.38 |
marcosw_ | henrys: presumably the batteries degrade with use. We've put ~20,000 miles on our car and haven't noticed anything yet. The batteries in our civic hybrid needed to be replaced after 6 years and 110,000 miles, which was covered under warranty. | 20:44.48 |
| Robin_Watts: I don't know why Nissan wouldn't say. It's $15,000 (plus labor). | 20:46.26 |
| Nissan has said that a capacity reduction of more than 20% is a warranty item. But only covered for 5 years or 60,000 miles. | 20:47.26 |
Robin_Watts | ah, maybe they've published those figures since the Top Gear review (which was last year) | 20:47.51 |
marcosw_ | otoh, we leased or leaf, so in another year and a half it won't be our problem. At that point we'll be worried about the full discharge bricking the batteries Tesla issue. | 20:48.25 |
| http://jalopnik.com/5887265/tesla-motors-devastating-design-problem | 20:48.46 |
henrys | our local tesla shop closed down, I miss window shopping | 20:49.10 |
Robin_Watts | marcosw_: It's OK, they'll be using the dreamliners cast offs by then :) | 20:49.13 |
marcosw_ | Robin_Watts: which have the bonus of keeping you toasty warm on even the coldest of days :-) | 20:49.39 |
henrys | cool you can brick your car - like a router or something. | 20:50.12 |
marcosw_ | I figure we have to buy a Tesla, we drive past the factory pretty much every day. | 20:50.12 |
| henrys: except more expensive to fix... | 20:50.25 |
henrys | so one good blackout and you could take out a fleet of them. | 20:51.38 |
marcosw_ | apple has announced that they are re-opening a manufacturing facility in Fremont. Most buy local to support the local economy. | 20:52.12 |
Robin_Watts | marcosw_: jeez. That's astounding. | 20:52.14 |
| marcosw_: Gotta get those apple key rings made somewhere. Cos you just know it won't be high value items. | 20:52.50 |
marcosw_ | Robin_Watts: http://9to5mac.com/2012/12/05/assembled-in-usa-imacs-are-originating-in-fremont-ca/ | 20:54.00 |
Robin_Watts | "assembled in usa". i.e. they stick the apple logo on in the USA :) | 20:54.31 |
marcosw_ | Robin_Watts: I think the US has rules about that (I know there are for cars, where they have to tell you how many of the subassemblies are built outside of the country and where). | 20:56.25 |
ray_laptop | my Lexus has an 10 yr / 150kmi warranty on the batteries | 20:56.26 |
henrys | How do they pay your crazy kWh charges - must get a discount | 20:57.08 |
| ? | 20:57.10 |
ryanakca | Any ideas why mupdf.com is failing to resolve? (I'm asking here seeing that ns{1,2}.ghostscript.com are listed as nameservers for the domain) dig mupdf.com @ns1.ghostscript.com times out. | 20:57.59 |
| Similarly withhttp://www.mupdf.com | 20:58.16 |
marcosw_ | ray_laptop: I believe that's required in California, since the batteries are part of the emission control system. | 20:58.20 |
| mupdf.com is down for me as well. I wonder if I screwed this up :-( | 20:58.43 |
Robin_Watts | marcosw_: and for me. | 20:59.06 |
marcosw_ | yeah, all of the none ghostscript.com sites are down (ghostxps.com, etc). I guess I better stop talking about cars and figure out why. | 20:59.48 |
ray_laptop | there must be something going on with the network. I can't believe that we really have 5 cluster machines down. | 21:00.07 |
| ryanakca: try resolving using 8.8.8.8 That fixed it for me | 21:00.42 |
Robin_Watts | ray_laptop: no, even with that mupdf.com is knackered. | 21:01.15 |
henrys | henrysx6 says down but I'm using it right now - all systems go on my end | 21:02.06 |
ryanakca | ray_laptop: I don't get an answer section back, it won't resolve. http://paste.debian.net/230235/ | 21:02.17 |
ray_laptop | henrys: yep. Some major network glitch that we'll all hear about on the news | 21:02.31 |
| Robin_Watts: is mupdf.com hosted at the same IP (184.73.189.105) ? | 21:03.56 |
marcosw_ | ray_laptop: yes. | 21:05.18 |
| you can get to the other domains via: http://184.73.189.105/svn.ghostscript.com/ | 21:05.53 |
| , for example. | 21:06.02 |
| but I can't get to mupdf.com that way... | 21:06.46 |
ray_laptop | marcosw_: so http://184.73.189.105/mupdf.com should work | 21:06.49 |
marcosw_ | http://184.73.189.105/ghostxps.com/ works | 21:07.01 |
| I wonder why mupdf.com is special... | 21:07.17 |
| right, mupdf.com isn't stored in /var/www on casper | 21:07.52 |
| it's in /home/tor/public_html/mupdf, so http://184.73.189.105/~tor/mupdf/ | 21:08.50 |
ray_laptop | marcosw_: yep. That's what is different | 21:08.57 |
marcosw_ | Maybe I'll suggest to tor that mupdf.com shouldn't be run out his personal account anymore :-) | 21:09.32 |
ray_laptop | we should really move it to the same place (/var/www/) | 21:09.34 |
marcosw_ | what's darcs.ghostscript.com? | 21:10.26 |
Robin_Watts | darcs is akin to git. | 21:10.55 |
| Something was kept in it for a while. possibly mupdf. | 21:11.15 |
marcosw_ | why no hg.ghostscript.com ? | 21:11.16 |
Robin_Watts | because not even tor8 is that mad :) | 21:11.28 |
marcosw_ | http://www.outageanalyzer.com/ has some data, but I'm not sure it's the problem we are seeing. | 21:13.52 |
ray_laptop | marcosw_: can't get to outageanalyzer.com (JK) | 21:15.09 |
| only two OPEN outages | 21:16.48 |
marcosw | alexcher's nodes are showing the same dns issues: Could not resolve hostname `svn.ghostscript.com': Host not found (http://svn.ghostscript.com) | 21:17.41 |
ray_laptop | 30% reduction in our cluster nodes :-( | 21:20.47 |
marcosw | so how is dns implemented for our domains? | 21:21.20 |
ray_laptop | marcosw: alex_i3770 was just lucky ? | 21:21.23 |
Robin_Watts | marcosw: Using bind on casper. | 21:21.35 |
| with the magic files in /var/cache/bind | 21:21.46 |
ray_laptop | marcosw: DNS ? Implemented ? It just happens dude | 21:21.49 |
Robin_Watts | I have no idea why it's the files in /var/cache/bind that are used rather than the ones in /etc/bind/ but it is. | 21:22.10 |
marcosw | ray_laptop: I'm not sure why iMac and alex_i3770 didn't have problems, dns seems pretty firmly not working. | 21:22.35 |
ray_laptop | whew... My test is almose finished | 21:22.46 |
Robin_Watts | marcosw: They may have a cache somewhere. | 21:22.50 |
henrys | I guess we're in a bind ⦠sorry can't help it | 21:23.00 |
ray_laptop | henrys: UGHHH | 21:23.12 |
marcosw | Robin_Watts: yes, but all the cluster nodes should have a similar cache. Unless they are configured differently. | 21:23.27 |
Robin_Watts | no, I mean the ISP may have a DNS cache. | 21:24.02 |
marcosw | Robin_Watts: so why are there so many entries in /var/cache/bind? Are we providing dns for all those domains? | 21:24.06 |
| Robin_Watts: but in that case the co-located nodes should all be using the same cache. E.g. I believe my router does DNS caching, but that means all of my cluster nodes get the same answer. | 21:25.08 |
| so the /var/cache/bind directory doesn't contain any information for mupdf.com (or ghostxps.com, etc.). | 21:26.10 |
ray_laptop | OK. Down to 26 regressions with NumRenderingThreads=4 and my latest patch. (better than 1,113). 17 of the 26 are 99-01-fixed.PS | 21:28.11 |
| henrys: so I should be able to wrap up 693590 today (or at least make a case for why it's no longer a blocker) | 21:28.55 |
henrys | that's good let's hope nothing else comes up. | 21:29.45 |
| I do see something I don't like in PCL but I don't think we should hold up the release for it. | 21:30.06 |
ray_laptop | henrys: are we going to cherry-pick alex's 19dfb490 fix ? | 21:31.48 |
henrys | I would doubt we have a commercial customer using @FILE but I'm not sure, if not I'd rather just leave it for next time. | 21:34.05 |
| what do you think? | 21:34.18 |
marcosw | Robin_Watts: I'm still wondering if the dns issues are something we have control over. But my dns mojo is poor. Any suggestion? | 21:34.28 |
henrys | ray_laptop:we can also just leave it to chrisl's discretion (my vote) | 21:35.21 |
marcosw | yeah, so ns1.ghostscript.com is not responding (and that's casper) and the seccondary name server is listed as 64.183.45.162 and that's peeves and it's not working either. | 21:37.09 |
| I'm going to reboot casper. | 21:37.50 |
| well, that didn't help. | 21:41.17 |
| rayjj: have you changed anything on peeves recently? | 21:42.46 |
henrys | failures to both nameservers for me: http://www.simpledns.com/lookup-dg.aspx | 21:44.25 |
marcosw | marcosw: I think I'm on to something, named has library issues on casper. I should have it back up as a name server shortlely. | 21:44.58 |
| okay, that should have fixed verything | 21:48.21 |
henrys | what was it? | 21:49.02 |
marcosw | the problem was that casper's named broke,presumably weeks ago when I upgraded to 12.04 but we didn't notice since peeves was providing secondary name service. | 21:49.39 |
henrys | ah | 21:49.59 |
marcosw | I don't know what changed on peeves so that it stopped working, but I'll work with rayjj to figure out what's going on. | 21:50.12 |
henrys | I seem to connect now but with peeves not ns1.ghostscript.com as I would've predicted | 21:51.31 |
marcosw | Sending request to "e.gtld-servers.net" (192.12.94.30) | 21:52.46 |
| Received referral response - DNS servers for "ghostscript.com": | 21:52.47 |
| -> ns1.ghostscript.com (184.73.189.105) | 21:52.48 |
| -> peeves.ghostscript.com (64.183.45.162) | 21:52.50 |
| Sending request to "ns1.ghostscript.com" (184.73.189.105) | 21:52.51 |
| Received authoritative (AA) response: | 21:52.53 |
| .. | 21:52.54 |
| . | 21:52.54 |
henrys | okay well mine sent the request to peeves and it worked. | 21:53.29 |
marcosw | apparently the name servers are chosen randomly, I just tried again and got ns2.ghostsript.com | 21:54.06 |
| I wonder if the cache on peeves expired. It's been ~30 days since I did the 12.04 upgrade :-) | 21:54.34 |
| btw, the reason that named broke on casper is that someone (Ralph?) installed a copy without using the ubuntu package system, so when the update to 12.04 happened named didn't get upgraded but a library it needed did and the old library version was no longer there. | 21:55.25 |
henrys | sinful | 21:57.52 |
| is there some way to check what's installed that isn't part of an official package. I think there is. | 21:58.22 |
| good question for tkamppeter | 21:58.58 |
marcosw | henrys: I found ways to do that for perl modules, which was the big problem with the bugzilla upgrade. | 21:59.03 |
henrys | marcosw:new customer needs access to gs, just sent the email to you | 22:01.48 |
marcosw | alexcher: did you notice that 19dfb490f5cd6a81d8a4482160adf15439d74b5b failed it's cluster run? Compile errors in ghostpcl. | 22:01.52 |
alexcher | marcosw: yes | 22:02.22 |
marcosw | at first I thought the problem was related to the dns issue, that one or more of the cluster nodes failed badly during a git pull or something. | 22:03.17 |
| brb, going to grab a late lunch. | 22:03.30 |
henrys | alexcher why don't you just revert that we are supposed to be frozen right? | 22:13.18 |
alexcher | henrys: I've committed to the trunk, the branch gs907 is frozen. | 22:23.25 |
henrys | alexcher oh sorry | 22:26.19 |
marcosw | getting back to my earlier question about energy cost to run a cluster node it looks like I'm paying $0.40/hour during times when there is nobody home and the heater is off, so the only thing consuming appreciable power should be the cluster nodes, the refrigerator, and the TiVos. | 22:41.21 |
henrys | how many tivo's does one need? | 22:42.52 |
marcosw | As I mentioned earlier 3 of my nodes are running nightly/weekly regression tests and so are consuming 220 watts, so the math works out nicely 220*3+100*2=860 which leaves 140 watts for misc. | 22:42.59 |
| henrys: two, one for the living room and one for the bedroom. | 22:43.11 |
| assuming I've done the math correctly I'm spending ~$150/month for power on behalf of artifex. | 22:44.41 |
| which is what I guessed much earlier, before the dns fiasco. | 22:45.20 |
| henrys: BTW, I couldn't find a tool that lists programs installed outside of packages, so I wrote one :-) Unfortunately it uses "apt-file search" so is unbelievably slow. | 22:46.13 |
henrys | ah I can imagine how you did that. | 22:47.08 |
marcosw | yeah, just feed the output to cut, sort, uniq, dpkg, and grep and there's your answer. | 22:49.14 |
| henrys: still churning away, but so far the list of programs that aren't installed via packages isn't too bad: autoconf, bison, imagemagick, valgrind, and a bunch of others that I don't recognize (i.e. xfont-utils). | 23:04.40 |
| what the heck, apt-search uses zgrep, no wonder it's slow. How big can the the repositories contents be uncompressed? | 23:08.43 |
| Forward 1 day (to 2013/01/30)>>> | |