| | <<<Back 1 day (to 2011/11/21) | 2011/11/22 |
| tharkun | Robin_Watts: Does that work? Don't ps file have a speciallized structure where it groups headers and . . .? | 00:00.10 |
| Robin_Watts | It depends what you want to do. | 00:00.45 |
| | postscript files as fed to a postscript interpreter just read all the way through. Catting 2 of them together will work just fine. | 00:01.06 |
| | Some postscript files follow a 'style' with specified headers etc in comments that allow other tools to operate on them in a consistent way. | 00:01.46 |
| | but those headers are ignored by interpreters themselves. | 00:02.00 |
| henrys | psmerge ... there is a discussion here http://stackoverflow.com/questions/3444931/how-to-merge-two-postscript-files-together | 00:03.38 |
| tharkun | henrys: thanks i will read that while conmuting later on. | 00:10.17 |
| | Robin_Watts: file12.ps will later be converted to a pdf file. let me tias and i'll report back later. | 00:10.57 |
| Robin_Watts | converted by gs? Then catting will be fine. | 00:11.22 |
| tharkun | Robin_Watts: Thanks | 00:13.40 |
| henrys | thanks for responding to that Robin_Watts, I was going to bring up at the meeting if this customer is getting to be too much of a burden, we can request fees for service. | 01:07.43 |
| mvrhel_laptop | henrys: I will look into the joint cache construction | 01:10.58 |
| | issue | 01:11.01 |
| henrys | I think it would help. | 01:19.24 |
| mvrhel_laptop | uhoh. kids are getting suspicious about santa. studying the melting ice caps | 01:46.25 |
| Robin_Watts | Awesome c=xkcd today. | 10:28.53 |
| tkamppeter | kens, thank you very much for the quick fixing yesterday, I have uploaded the fixed Ubuntu Ghostscript packages for testing now. | 11:46.13 |
| kens | Great news tkamppeter, let me know if there are any more problems | 12:26.53 |
| Robin_Watts | Only 19000 differences :( | 13:51.24 |
| | Marcos is going to kill me... | 13:51.32 |
| kens | Yikes! | 13:51.38 |
| chrisl | Robin_Watts: rather a lot of seg faults, too :-( | 14:05.29 |
| Robin_Watts | hadn't spotted them :( | 14:05.47 |
| | I was expecting lots of 'small' changes to images, but I'm seeing changes I don't understand at all. | 14:06.13 |
| chrisl | Oh, sorry, looks like they seg faults that have stopped - long output, I missed the heading...... | 14:07.05 |
| | s/they/they're | 14:07.17 |
| Robin_Watts | Ah, lots of SEGVs should have stopped since the last clusterpush, yeah. | 14:08.44 |
| chrisl | Yeh, I'll keep my nose out of your regression runs :-) | 14:09.43 |
| Robin_Watts | found a typo that explains the wierdnesses. | 15:37.34 |
| kens | Sounds positive | 15:37.49 |
| Robin_Watts | mvrhel: Morning. | 16:23.23 |
| mvrhel | good morning | 16:23.29 |
| Robin_Watts | I'm wondering if I should pass bug 692692 to you. | 16:23.46 |
| | It's a problem with antialiasing and transparency compositors. | 16:24.08 |
| mvrhel | that is fine. I thought this was fixed since I had added the copy alpha in the pdf14 device | 16:24.53 |
| | but apparently there is still an issue | 16:25.07 |
| Robin_Watts | I'll have a go if you want, but I suspect the learning curve time for me may dwarf the time it'd take you to fix it :) | 16:25.43 |
| mvrhel | no feel free to pass if off to me | 16:26.00 |
| Robin_Watts | thanks. | 16:26.04 |
| henrys | Robin_Watts:it doesn't sound like the file is wrong spec-wise. vectors are any part of pixel and images are center of pixel - are we doing that and adobe isn't? | 16:29.34 |
| Robin_Watts | henrys: We are, and I guess adobe isn't. | 16:30.20 |
| | OR... there is some tiny vertical shift involved, which means that the rectangles smear onto 2 lines for us. | 16:31.03 |
| chrisl | Robin_Watts: Might it be worth finding out what CPSI does? Although having Acrobat export a TIFF is *better* than its display, I've still found cases where it was at odds with a "real" rip. | 16:33.29 |
| henrys | another idea is that adobe is converting a pure white or black image to a rectangle and inheriting the vector rendering rules. | 16:36.08 |
| Robin_Watts | The images aren't pure black and white. | 16:36.38 |
| | (they are monochrome, but a mixture of black and white pixels) | 16:36.51 |
| henrys | oh okay | 16:36.52 |
| Robin_Watts | It would probably be possible to have this a as an option. | 16:37.49 |
| | (i.e. have -dGRIDFIT) | 16:37.57 |
| chrisl | Or maybe image masks use line art scan conversion rules - I seem to remember discussion that in my previous job........ can't remember if we reached a conclusion | 16:38.35 |
| | s/discussion/discussing | 16:38.52 |
| Robin_Watts | chrisl: That would make sense, in a way. | 16:39.20 |
| chrisl | that's what I was thinking - I could see it being a defensible decision | 16:39.45 |
| mvrhel | Robin_Watts: thanks for closing 692512 | 16:40.37 |
| | and for fixing it :) | 16:40.44 |
| | back to 692674 for me... | 16:41.43 |
| Robin_Watts | mvrhel: You're welcome. I think the issue (white lines) still exists though for some files, but the bug itself as reported by the customer is fixed. | 16:42.56 |
| henrys | there is an "adjust" in the image structures, easy enough to check. | 16:43.54 |
| | gs_image_t_init_adjust() ... | 16:45.59 |
| ray_laptop | morning, all | 16:46.10 |
| henrys | hi | 16:46.31 |
| ghostbot | salut | 16:46.31 |
| ray_laptop | kens: I was going to mention that on your opdfread change, the setpagedevice has a 'within +/- 5 points' tolerance when media matching -- what do yoy think about making the 'window' of detecting a changed PageSize larger ? | 16:48.15 |
| Robin_Watts | Hmm. I only have a gs_pixel_image_t, not a gs_image_t at the point at which I need it. | 16:48.31 |
| kens | ray_laptop : yes could do that. It sthe action of actually calling setpagedevice that flushes the Duplex though | 16:49.08 |
| henrys | well I think chrisl was saying we'd simply do all masks with adjust == .5 or that was my interpretation. | 16:49.22 |
| kens | But its easy enough to make it bigger | 16:49.27 |
| | I was more keen to make sure it got into Till's patch | 16:50.06 |
| ray_laptop | kens: right. That's the important part -- so he doesn't shift away from us (again) | 16:50.41 |
| Robin_Watts | henrys: yes. | 16:51.38 |
| kens | ray_laptop : that was the idea. | 16:51.52 |
| Robin_Watts | That would be enough for this file, certainly. | 16:51.56 |
| kens | In fairness, at least two of those bugs were really rather bad. | 16:52.03 |
| henrys | phil used his email in a bug report should we (I) remind him it is a public forum or is he okay with it? | 16:52.11 |
| | anybody know? | 16:52.14 |
| kens | I don't know, and unless omseone does, we should remind him | 16:52.42 |
| ray_laptop | Phil knows that it's a public area, but he might like to have his email removed. Marcos should probably contact him. | 16:54.43 |
| | and even though we treat all customers as if they prefer the anonymity gained by us using customer #'s, I don't think they care. We have several customers that put their own bugs in with their company email address. | 16:56.28 |
| kens | Yes, agreed, many people do not mind | 16:56.48 |
| henrys | Right I just want to make sure his actions were purposeful. | 16:57.30 |
| Robin_Watts | meeting time? | 16:59.39 |
| henrys | oh I have 2 minutes | 16:59.46 |
| kens | 10 seconds for me | 17:00.01 |
| | 5PM | 17:00.14 |
| henrys | funny my laptop is slow | 17:00.20 |
| | server is right | 17:00.26 |
| ray_laptop | agrees with my laptop | 17:00.49 |
| kens | and marcosw is here | 17:00.53 |
| marcosw | oh uh, what did I do now? | 17:01.03 |
| kens | arrived on time | 17:01.22 |
| henrys | anyway we had a bad incident this week where an important bug didn't get worked on because it was assigned properly. I've updated the who owns what document. | 17:01.31 |
| | But please don't sit on bugs if you know somebody else can make progress with them. | 17:01.49 |
| Robin_Watts | s/properly/improperly/ ? | 17:02.09 |
| kens | I generally re-assign if I think that, did I miss one ? | 17:02.14 |
| henrys | improperly | 17:02.24 |
| | no alexcher missed this one | 17:03.00 |
| | the blame isn't important - let's all review our bugs and make sure we don't have problems assigned that other can work on - use the ownership guide or ask. | 17:03.41 |
| alexcher | henrys: yes. I'm lurking | 17:04.21 |
| henrys | my other topic was high maintenance custoemrs. If you think you have one - I think chrisl is working with now it doesn't hurt to go to Miles check the revenues and see if it is worth charging them extra. | 17:05.53 |
| | not so interested in the money but providing incentives for them to do their own work. | 17:07.13 |
| kens | That same customer is also badgering robin and tor | 17:07.21 |
| henrys | right | 17:07.34 |
| chrisl | rather wishes he'd ignored that forwarded mail from marcosw! | 17:07.48 |
| kens | Should send the 'canned response' like I do ;-) | 17:08.25 |
| henrys | so I am going to talk to Miles about him if there is somebody else let me know. | 17:08.35 |
| | is tor8 around? | 17:08.54 |
| tor8 | henrys: ugh. crypto export compliance and cruft. | 17:09.04 |
| ray_laptop | I think Raed's company is paying us fairly well | 17:09.25 |
| chrisl | kens: it didn't start with the normal "it doesn't work" complaint..... :-( | 17:09.32 |
| henrys | are you talking about the "feds"? | 17:09.33 |
| | tor8? | 17:09.39 |
| tor8 | henrys: app store upload | 17:09.44 |
| kens | chrisl close neough ;-) | 17:09.54 |
| chrisl | kens: it was actually a build question that got me involved | 17:10.17 |
| henrys | ray_laptop:well maybe they can pay us more. | 17:10.46 |
| ray_laptop | henrys: that's always welcome :-) | 17:11.01 |
| kens | chrisl I think that was a mis-understnading | 17:11.11 |
| | due ot English as a second language | 17:11.19 |
| | Just like mine it seems | 17:11.30 |
| henrys | tor8:actually I wanted to ask if anyone is working on making mupdf a firefox plugin. | 17:11.34 |
| kens | I thought it was | 17:11.48 |
| tor8 | henrys: I think someone worked on a fork once, but I think he's lost interest. it's on github somewhere. | 17:12.11 |
| henrys | going further - I wonder if we shouldn't be trying to get mozilla to pick up mupdf as a built in to compete with chrome. | 17:12.41 |
| chrisl | kens: no, he really wants to include a CIDFont fallback in the romfs, which isn't currently supported, really....... | 17:12.59 |
| tor8 | firefox plugins are very platform specific, you have to do win32 or gtk+ calls to draw stuff and get input. having them pick it up would be preferable to their pdf.js stuff :) | 17:13.21 |
| kens | chrisl then he's madder than I thought | 17:13.35 |
| Robin_Watts | henrys: mozilla is actively trying to slim down firefox. | 17:13.39 |
| tor8 | I think having an official MuPDF firefox plugin for windows and linux, and make sure it's listed in their extension browser, would be a very good thing. | 17:14.37 |
| Robin_Watts | tor8: unless they can hook it in as a javascript object to produce an image/images that they can then display in the browser. | 17:14.57 |
| chrisl | kens: in some ways. But it seems that the usual kanji TTF suspects aren't on every Windows system, so it kind of makes to include one - we're going to, after all! | 17:14.57 |
| henrys | one of the reasons I prefer chrome is builtin pdf viewing. | 17:15.25 |
| kens | chrisl, fine, but then on disk, not in ROM.... | 17:15.25 |
| tor8 | Robin_Watts: chrome proposed the cross platform Pepper API for browser plugins -- what they use for their bulit-in flash, but mozilla said no thanks to that | 17:15.45 |
| Robin_Watts | I know someone working at mozilla. He's on the video side, but he might give me a contact... | 17:16.21 |
| henrys | I have one connection at mozilla, I'll ask too. | 17:16.31 |
| mvrhel | would a mupdf plugin do xps to? my old employer has a plugin for xps in firefox and safari | 17:16.36 |
| ray_laptop | they may have NIH-itis | 17:16.43 |
| chrisl | kens: I'm going to include it in the romfs - it adds under 2Mb, and will save a lot of hassle. | 17:16.46 |
| kens | From our POV trht ameks sense, but he is presuambly doign an installer, | 17:17.09 |
| ray_laptop | chrisl: include the droid asian font ? | 17:17.17 |
| chrisl | ray_laptop: yes | 17:17.23 |
| henrys | anybody else have meeting topics? | 17:17.29 |
| tor8 | mvrhel: yes. | 17:17.36 |
| ray_laptop | yes | 17:17.36 |
| | ken: henry: yesterday we were discussing having pdfwrite (and ps2write) be able to close files. I had thought about exposing 'gs_closedevice' but was concerned about parts of gs not being happy if the current device was closed, so I thought a more cautious approach would be to have a '.reopendevice' that does gs_closedevice(dev); gs_opendevice(dev); | 17:17.38 |
| henrys | ray_laptop:ah yes I forgot | 17:17.50 |
| ray_laptop | I tried this and it seems to work, and doesn't crash. | 17:17.56 |
| kens | ray_laptop : and how do we trigger it ? | 17:18.03 |
| ray_laptop | I did it with: -sDEVICE=pdfwrite -sOutputFile=x.pdf examples/tiger.eps -c ".reopendevice << /OutputFile (y.pdf) >> setpagedevice (examples/tiger.eps) run quit" | 17:18.44 |
| tor8 | henrys: do you have Xcode set up for development? | 17:18.51 |
| | ...for iOS development | 17:18.57 |
| kens | So a custom operator | 17:18.57 |
| henrys | tor8:no just regular xcode. | 17:19.10 |
| tor8 | right. so I'll just hope I haven't got any embarassing bugs left :) | 17:19.28 |
| ray_laptop | kens: right. but it lets a server force PDF to close and re-open which created the x.pdf _and_ the y.pdf | 17:19.34 |
| kens | Sure not a problem, and I don't have to dop anything, so even better from my POV | 17:20.02 |
| Robin_Watts | ray_laptop: Random idea... | 17:20.08 |
| kens | I don't know how you will change filenames though ? | 17:20.13 |
| Robin_Watts | could we make gs automatically do that when it meets a -sOutputFile= ? | 17:20.25 |
| henrys | tor8:I will set up for future stuff can you put the code somewhere? | 17:20.45 |
| ray_laptop | kens: the other approach we had discussed was to close when OutputFile changed IFF we had accumulated anything, then re-open with the new OutputFile | 17:20.56 |
| | Robin_Watts: right | 17:21.08 |
| tor8 | henrys: the code is in git | 17:21.09 |
| Robin_Watts | so gs -sDEVICE=pdfwrite -sOutputFile=animals.pdf tiger.eps owl.pdf -sOutputFile=trees.pdf birch.pdf oak.pdf | 17:21.15 |
| kens | That was what I recalled. | 17:21.30 |
| henrys | tor8:okay no extra stuff needed | 17:21.46 |
| kens | Robin_Watts : will that command line work ? | 17:22.04 |
| | I think not | 17:22.06 |
| tor8 | just the usual apple nightmare with certificates and provisioning profiles... | 17:22.14 |
| Robin_Watts | kens: Why not ? | 17:22.20 |
| tor8 | but the source is all in the git | 17:22.20 |
| henrys | tor8:I have the ipad but I guess there is no way to get it there... I hate apple. | 17:22.27 |
| | sometimes | 17:22.33 |
| ray_laptop | Robin_Watts: I'm not sure that the second -sOutputFile does anything -- we pass it to the device from gs_init when we start, but after that I think it only just changes the value of OutputFile in systemdict | 17:23.01 |
| kens | Robin_Watts : I am not familiar with teh 'server mode' but would it process the saecond -s properly ? | 17:23.03 |
| tor8 | this iOS stuff is the most openly developer hostile environment I've ever had the misfortune to use | 17:23.03 |
| Robin_Watts | henrys: There *used* to be a way to subscribe to the apple dev stuff as a company rather than an individual. | 17:23.04 |
| ray_laptop | which is why I used 'setpagedevice' | 17:23.18 |
| Robin_Watts | kens: Right. My command line was "what I'd like to see" rather than necessarily "what currently works" | 17:23.39 |
| ray_laptop | Robin_Watts: if the goal is to allow an app using gs as a server, then setpagedevice is just as convenient | 17:23.53 |
| tor8 | Robin_Watts: yeah, but that needs miles to create the account... | 17:24.02 |
| kens | I'm more interested in getting the server stuff to work than the command line | 17:24.19 |
| Robin_Watts | tor8: No, it needs one of us with a company credit card, I think :) | 17:24.21 |
| | kens: One without the other would seem suboptimal. | 17:24.33 |
| tor8 | they want a legal representative of the company to sign off on it | 17:24.38 |
| kens | Robin_Watts : I disagree | 17:24.51 |
| ray_laptop | tor8: I can forge Miles' signature pretty well ;-) | 17:25.02 |
| Robin_Watts | tor8: Right. But that might be a smarter way to go long term. If we are submitting to the app store, presumably it *ought* to come from a company rather than from an individual ? | 17:25.32 |
| | s/a company/the company/ | 17:25.44 |
| ray_laptop | seriously, Miles probably would not object to signing | 17:25.45 |
| Robin_Watts | kens: I can't see how having both work would be bad :) | 17:26.04 |
| kens | Its not bad, but its extra work | 17:26.31 |
| | for no real benefit IMO | 17:26.39 |
| henrys | tor8:so you want miles to subscribe as artifex? | 17:26.42 |
| ray_laptop | Robin_Watts: the gs_init processing of options during start-up is pretty messy | 17:26.55 |
| kens | The server needs to work, if it falls out that the command line does too, then fine, but I don't want to expend extra effort for the command line | 17:27.15 |
| ray_laptop | Robin_Watts: many options only work BEFORE the first file is processed | 17:27.30 |
| kens | I don't see people calling for that ability from the command line though | 17:27.38 |
| Robin_Watts | ray_laptop: Right. Shame though. | 17:28.06 |
| ray_laptop | Robin_Watts: besides, you _can_ do it from the command line (using -c "..." as I did) | 17:28.10 |
| Robin_Watts | yes, but it's not pretty. | 17:28.22 |
| henrys | the command line is not important - it seems being able to use the api to create a server is very important - I am very surprised this hasn't come up. | 17:28.46 |
| tor8 | henrys: I have signed up with my 'artifex' personality, which should be good enough for our limited use. I won't object to us getting a company account though! | 17:28.50 |
| | it'll certainly make internal testing easier | 17:29.03 |
| kens | henrys it has come up, more than once | 17:29.05 |
| | Its on my 'TODO' list | 17:29.12 |
| Robin_Watts | It feels like I've ended up fighting for something that isn't really important. I can see that the server mode is way more important, and that the cli mode would just be 'nice'. | 17:29.35 |
| ray_laptop | Unless someone has a strong objection, I'll do a patch with the change and documentation and send it to tech for comment | 17:29.47 |
| kens | Like I said, if it turns out that the LCI works, then that's a bonus | 17:30.01 |
| | yes please ray_laptop | 17:30.09 |
| ray_laptop | OK. | 17:30.13 |
| Robin_Watts | New topic? Windows cluster node. | 17:30.16 |
| ray_laptop | I thought that was old news ;-) | 17:30.29 |
| | thanks for doing that Robin_Watts | 17:30.35 |
| henrys | Robin_Watts:you don't have that working yet ;-) | 17:30.41 |
| Robin_Watts | I have a windows cluster node working. I'd like Marcos to review my changes to the cluster code. | 17:30.49 |
| | And the differences between unix and windows are as expected. (not that many, and banded mode only) | 17:31.15 |
| henrys | what happened with the difference - bandheight etc. | 17:31.21 |
| | ? | 17:31.26 |
| | meeting officially is over at 10:30 | 17:31.49 |
| chrisl | Just quickly before everyone moves on: Working on Bug 692681: Distiller consumes the file happily, but CPSI (thanks ray_laptop!) errors. My inclination is to default to handling the input (as Distiller does), and throw an error if we're in CPSI mode - any other opinions? | 17:32.05 |
| henrys | so can we filter the windows box to just run non banding? | 17:32.09 |
| kens | Waiting for a phone call to tell me to put dinner on | 17:32.11 |
| Robin_Watts | like I say, that's an outstanding issue that causes some differences (not a huge number, but more than we'd want on a daily basis) | 17:32.12 |
| | henrys: I could try to add that. | 17:32.26 |
| ray_laptop | running banding mode on Windows isn't that important, IMHO -- and if there is only a single node, running less tests makes sense | 17:33.25 |
| Robin_Watts | ok. so that's that topic done, thanks. | 17:33.51 |
| kens | Anything we can get it to do is a plus | 17:33.59 |
| henrys | now to get chrisl's sparc online. | 17:34.08 |
| kens | :-O | 17:34.14 |
| mvrhel | :) | 17:34.24 |
| chrisl | Hmm, so you *might* get one job per run out of it! | 17:34.32 |
| Robin_Watts | The windows cluster node calls nmake rather than make. nmake can't run in parallel, so compiling is much slower. | 17:34.46 |
| ray_laptop | Robin_Watts: how do we keep the Windows node from delaying cluster runs ? | 17:35.07 |
| Robin_Watts | With the sparc, I doubt you'd get compiles done before the rest of the job had finished. | 17:35.10 |
| henrys | ray_laptop:I do wonder if you are going to get a great performance improvement over running from start if you do close the device. | 17:35.44 |
| Robin_Watts | ray_laptop: The windows node generally finishes compiling by the time the others have done 400 tests or so. | 17:35.57 |
| | so as long as you're testing enough jobs there is no problem. | 17:36.19 |
| kens | henrys the biggest erpformance is going to be writing the PDF file. | 17:36.40 |
| | performance hit | 17:36.44 |
| ray_laptop | henrys: I'll test that with the PS file from the bug | 17:36.50 |
| marcosw | doorbell, back in a bit | 17:37.00 |
| chrisl | Robin_Watts: I don't pay much attention generally, but I think my SPARC/Solaris box takes ~15 minutes for a clean build. | 17:37.21 |
| ray_laptop | Robin_Watts: so all we do is compile on the Windows machine ? | 17:37.24 |
| kens | will be interested in test result | 17:37.25 |
| henrys | I still think mvrhel's projects are worthwhile - he is working on lowering startup costs on a couple fronts. | 17:37.35 |
| Robin_Watts | ray_laptop: No, we compile, then run tests on it. | 17:37.45 |
| ray_laptop | henrys: yes, I agree that is good too | 17:37.55 |
| Robin_Watts | All the nodes start compiling, then ask for jobs to be served. If a node takes longer to compile, it just doesn't start testing jobs as fast as the others. One taking ages to compile doesn't stop the others getting on with it. | 17:38.29 |
| ray_laptop | Robin_Watts: so it just requests (non-banded) jobs and we compare those with the linux results ???? | 17:38.37 |
| Robin_Watts | Yes, I'll try and make it so when it asks for jobs it asks for non-banded ones. | 17:39.01 |
| chrisl | henrys: if we want a big-endian cluster machine, we would be better served trying to track down a PPC Mac Pro, I reckon. | 17:39.18 |
| Robin_Watts | If we can get the sparc to compile within 15 mins, then that could be helpful too. | 17:39.33 |
| | (Even if it doesn't test anything!) | 17:39.40 |
| marcosw | chrisl: I have a PPC iMac, but it takes 30 minutes to compile ghostscript and ghostpdl. | 17:39.54 |
| henrys | I do wish we had big endian coverage. | 17:39.58 |
| Robin_Watts | Would a powerpc mac mini do? | 17:40.15 |
| ray_laptop | not very much -- what the SPARC usually trips on are alignment issues during execution | 17:40.16 |
| chrisl | ray_laptop: good point. | 17:40.32 |
| ray_laptop | I've also got a PPC mini -- but I think I', back at 10.2 | 17:40.42 |
| alexcher | I have 2 PPC32 boxes, one is ready to use. | 17:40.49 |
| Robin_Watts | OK, so we could make it do a token few jobs. | 17:40.51 |
| henrys | is the PPC's alignment strict like the sparc? | 17:41.14 |
| chrisl | In truth, my SPARC machines are just too noisy to keep on all the time - I'd have to find somewhere suitable to house them. | 17:41.40 |
| Robin_Watts | ray_laptop: Why is 10.2 a problem? | 17:42.21 |
| henrys | well alexcher could buy one for the basement for 50 cents or whatever they go for? | 17:42.30 |
| marcosw | I think it makes more sense to have big-endian and/or sparc to run as nightly or weekly regressions, having them run a small number of cluster jobs at random isn't going to help much. | 17:43.11 |
| henrys | I would like a machine that segv on an unaligned access big endian | 17:43.12 |
| Robin_Watts | marcosw: They don't need to run 'random' cluster jobs. | 17:43.35 |
| marcosw | so which cluster jobs would you have them run? | 17:43.51 |
| alexcher | henrys: AFAIK, PPC has no alignment problems, like x86 | 17:43.59 |
| Robin_Watts | IF we're changing the cluster so that nodes report their abilities, then we can tune them to run specific jobs. | 17:44.21 |
| ray_laptop | Robin_Watts: I thought 10.2 was really old | 17:44.37 |
| Robin_Watts | ray_laptop: Why do we care? | 17:44.47 |
| | We're just using the basic posix layers. | 17:44.59 |
| ray_laptop | alexcher: I think you are correct about alignment on PPC | 17:45.02 |
| chrisl | robin_watts: rather defeats the point of running the sparc - we want to check none of our test cases cause unaligned memory accesses | 17:45.34 |
| marcosw | Robin_Watts: I guess I don't see how to pick the "specific jobs". | 17:45.36 |
| Robin_Watts | marcosw: I dunno - can we not pick 5 jobs that use images/text/bitmaps ? | 17:46.26 |
| ray_laptop | Running the SPARC on a weekly basis (even if it takes all week) would let us 'spot check', then we can find the problem before a customer does | 17:46.36 |
| henrys | I think we'd catch most unaligned problems with just a small mix of jobs - fonts, images, vectors etc. | 17:47.02 |
| alexcher | Can Valgrind test for unaligned access ? | 17:47.15 |
| Robin_Watts | ARMs can enforce alignment. | 17:47.53 |
| henrys | not sure obviously valgrind is extremely slow. | 17:47.53 |
| Robin_Watts | but I don't want to think about setting the beagleboard up as a cluster node :) | 17:48.14 |
| henrys | I can just make this an agenda item and we can talk about it later, let's get some use out of the windows box first. | 17:48.36 |
| ray_laptop | alexcher: good idea! http://stackoverflow.com/questions/1929588/x86-howto-catch-data-alignment-faults-aka-sigbus-on-sparc | 17:48.40 |
| chrisl | Robin_Watts: that's not a run-time option, is it? | 17:48.44 |
| Robin_Watts | chrisl: It's a software switch, yes./ | 17:48.57 |
| marcosw | Robin_Watts: if we are only going to run 5 jobs that doesn't seem like it would require a cluster. Which gets back to my earlier point of it making more sense to run as a nightly/weekly task. | 17:49.20 |
| chrisl | Robin_Watts: Oh, I thought it had to be set during initialisation....... | 17:50.00 |
| Robin_Watts | marcosw: It'd be a smoke test. | 17:50.03 |
| chrisl | ray_laptop: I never had much luck getting qemu-sparc and it's variants to work well enough to be useful. | 17:51.39 |
| ray_laptop | chrisl: Just the first thing out of professor google's mouth. | 17:52.36 |
| marcosw | Robin_Watts: I agree a sparc/g5 smoke test is a good idea, but it shouldn't be part of the cluster. If it's important to run it on every commit it can be done the same way that the clang warnings runs. | 17:53.12 |
| henrys | chrisl:when is the last time you ran gs on the sparc - just curious? | 17:53.15 |
| chrisl | henrys: when I fixed the aligned access bug - whenever that was..... | 17:53.41 |
| | henrys: early October, it seems | 17:54.36 |
| alexcher | marcosw: we can have several clusters for different platforms. | 17:55.16 |
| henrys | so I'll put it on the agenda anyway. | 17:55.19 |
| chrisl | I try to run some smoke tests while the release candidates ferment in the outside world | 17:55.36 |
| | I wonder if my garage would get too cold in the winter for the two SPARC machines...... | 17:57.29 |
| henrys | Robin_Watts, ray_laptop:on another topic, you don't think we are taking a performance hit treating bitmaps with 32 bit reads and write sizeof(long) on windows 64? | 17:58.07 |
| Robin_Watts | I don't see how we can be. | 17:58.39 |
| | Indeed, I'd suspect that if anything linux is losing out. | 17:58.51 |
| | (more padding means less likely to fix in cache etc) | 17:59.08 |
| | If we were writing 64bit things to 32bit aligned data, then yes, we might be losing out, but AIUI, we only ever write at int alignment. | 17:59.57 |
| henrys | do you mean long alignment? | 18:00.58 |
| | ints are 32 on 64 bit linux and mac | 18:01.16 |
| ray_laptop | henrys: there aren't that many "big" fills and the CPU is usually NOT on critical path in those loops (memory is) | 18:01.37 |
| kens | Time for me to go, goodnight all | 18:01.54 |
| mvrhel | bye kens | 18:02.01 |
| ray_laptop | bye, kens | 18:02.09 |
| Robin_Watts | henrys: no. | 18:02.15 |
| | I don't mean long alignment. | 18:02.37 |
| ray_laptop | longs are 64-bits on 64-bit linux | 18:02.52 |
| Robin_Watts | I am aware that ints are 32bit on all 32/64 bit windows/linux systems. | 18:02.57 |
| ray_laptop | but ints are 32 | 18:03.02 |
| Robin_Watts | AIUI we only ever write ints. | 18:03.08 |
| | hence insisting on long alignment is overkill. | 18:03.17 |
| ray_laptop | iirc, only the DEC Alpha had a 64-bit int | 18:03.30 |
| | which caused problems when porting code to it. alexcher probably remembers that one better | 18:04.05 |
| henrys | Robin_Watts:oh I see what you mean now. | 18:04.39 |
| alexcher | ray_laptop: Perhaps, GCC can be configured to have sizeof(int)==8 | 18:05.49 |
| ray_laptop | henrys: I was concerned that gdevm64 was writing 64-bit gx_color_index values, but it actually writes 2 32-bit parts | 18:06.03 |
| | alexcher: I hope not ! | 18:06.35 |
| henrys | but the bitmap code could be modified to operate on the largest chunks possible - arguably that would be the right way to do things. | 18:09.21 |
| Robin_Watts | henrys: Yes, we could try and change code to make use of longs if they were available. | 18:10.20 |
| | but just because a platform offers an int64_t doesn't mean we should use it. | 18:10.37 |
| | ARMs offer an int64_t, but code will be faster that uses int32_t's. | 18:11.14 |
| henrys | Robin_Watts:I thought the heuristic should be the native word size, does that fail to produce the fastest code in some circumstance? | 18:11.56 |
| Robin_Watts | henrys: OK. And how do you detect "native word size" ? | 18:12.21 |
| henrys | does it not correspond to pointer alignment? | 18:13.16 |
| Robin_Watts | henrys: Urk. | 18:14.32 |
| henrys | ARCH_ALIGN_PTR_MOD more specifically. | 18:15.22 |
| Robin_Watts | We have fast efficient 64bit registers on 32bit x86 AIUI. | 18:15.41 |
| henrys | but anyway maybe a low priority agenda item. | 18:16.12 |
| Robin_Watts | I'd rather stick with 'long' by default and override it in 'known' cases. | 18:16.40 |
| henrys | fair enough. | 18:17.13 |
| Robin_Watts | (and make our code make use of 'long's) | 18:17.40 |
| ray_laptop | henrys: alexcher: I was thinking about deprecating the .setpdfwrite extension (change to doc to deprecated, leave it in as a { } proc for clients that use it). With my GC reclaim changes, it is ineffective anyway | 18:22.47 |
| | and of course, remove it from our ps2pdfxx and dvipdf scripts | 18:23.54 |
| henrys | Robin_Watts:what is stdint.h:SIZE_MAX on this windows box? | 18:24.40 |
| Robin_Watts | let me look. | 18:25.50 |
| alexcher | ray_laptop: it's fine for me. | 18:26.16 |
| henrys | just curious | 18:26.19 |
| Robin_Watts | No stdint.h ? | 18:27.29 |
| tor8 | henrys, Robin_Watts: app uploaded. now we wait... | 18:29.24 |
| henrys | I forgot was it 7 to 10 days? | 18:31.12 |
| tor8 | hopefully in time for the meeting. | 18:32.48 |
| Robin_Watts | henrys: crt/limits.h: SIZE_MAX is _UI64_MAX on win64, UINT_MAX on win32. | 18:33.42 |
| henrys | I was just thinking that could be a candidate for word size... | 18:34.59 |
| | ray_laptop:okay by me. | 18:39.48 |
| ray_laptop | hmm... my .reopendevice worked ok on a simple test, but then I ran it with annots.pdf as input to create individual PDF pages and I get " .\base\gsicc_lcms.c:33: gscms_error(): cmm error : Corrupted memory profile" errors on _some_ of the pages :-( | 18:50.42 |
| | processing 5002_EPSF_Spec.pdf works OK (except for the expected "ERROR: A pdfmark destination page 33 points beyond the last page 1." messages | 18:56.24 |
| mvrhel | meeting at the school. bbiaw | 19:18.37 |
| ray_laptop | have to go pick up the twins (they get out early today). bbiaw. | 19:23.01 |
| AlecTaylor | My patch is ready :) | 19:25.12 |
| | I'll submit it tomorrow along with an explanation of the various techniques I considered, the algorithms I attempted, and the algorithms I ended up using, along with some initial results (accuracy %). :) | 19:27.02 |
| Robin_Watts | AlecTaylor: Cool. Open a bug on bugs.ghostscript.com and attach it there ? | 19:27.10 |
| AlecTaylor | Robin_Watts: Once I've finished writing about it, sure. I would also like to post at least the explanation on the mailing-list. | 19:35.04 |
| Robin_Watts | AlecTaylor: Sure. From our point of view, having it on the bugs database is the best place; anyone can then comment etc, and we have a permanent, easy to access record. | 19:36.02 |
| AlecTaylor | will do both then :) | 19:36.38 |
| henrys | marcosw:still about? | 19:36.43 |
| Robin_Watts | That's strange... | 19:37.24 |
| | Can someone run this for me please? | 19:37.39 |
| | gs/debugbin/gswin32c.exe -dFirstPage=25 -dLastPage=25 -o out%d.ppm -r72 -dMaxBitmap=100000000 -sDEVICE=ppmraw ../ghostpcl | 19:37.51 |
| | /tests/pdf/Bug6901014_Additional_testcase.pdf | 19:37.53 |
| | That appears broken for me, even when I back out my changes... | 19:38.07 |
| | Oh. Ignore me. | 19:39.22 |
| AlecTaylor | :P | 19:39.27 |
| Robin_Watts | -dFirstPage=25 means that the first page I get out is out1.ppm :( | 19:39.44 |
| | I'd like to buy a kindle touch from amazon.com - they aren't out in the UK. Could I order one and get it sent to one of you, and then get you to bring it to Miami, please? | 19:51.53 |
| henrys | sending it to me is fine. | 20:21.57 |
| Robin_Watts | henrys: Thanks. | 20:39.03 |
| | henrys: Will be with you on Friday. | 20:43.30 |
| henrys | okay | 20:43.43 |
| Robin_Watts | Many thanks for this - I don't see why they don't release it in the UK... | 20:43.54 |
| | Down to 8000 diffs. | 20:45.23 |
| henrys | actually it looks like firefox is going with PDF.js it can't imagiine mupdf wouldn't be a better solution than that. | 21:12.18 |
| | s/it/I | 21:14.52 |
| Robin_Watts | bonkers | 21:38.35 |
| sebras | PDF.js wasn't that exceedingly slow and didn't render PDF properly? | 22:17.53 |
| | and by properly I mean that it failed to render a big chunk of the spec. | 22:18.10 |
| | Forward 1 day (to 2011/11/23)>>> | |
IRC Logs