# IRC Logs

## Log of #ghostscript at irc.freenode.net.

Search:
 <<> -i2 surely? 10:35.49 For what value of d can: !isnan(d) && CLAMP(d, -FLT_MAX, FLT_MAX) != d ? 10:38.05 Oh, d is a doulbe. 10:38.28 double. 10:38.31 OK, all that looks plausible except the shift thing. 10:38.44 sebras Robin_Watts: well, you found a bug I believe. I never actually changed the shifting, just the test... 10:44.46 but yes, it should be -i2. 10:45.09 thanks for the review. 10:47.27 Robin_Watts Ah :) 10:55.40 You're welcome. 10:55.44 Robin_Watts runs. back soon. 10:56.26 kens2 paulgardiner : Robin has gone for a run and tor isn't here right now 11:07.21 If you were looking for them 11:07.34 paulgardiner kens2: Thanks. I was hoping to talk to Tor, but I wasn't necessarily expecting to find him yet. I just thought I'd hang around - catch him later. 11:14.07 kens2 No problem :-) 11:17.13 paulgardiner There's a lot more people here than I was expecting at this time, given where I thought most people were based. Must be a lot of early risers. :-) 11:23.54 kens2 A lot of them aren't really here as such, just logged in and recording 11:24.24 paulgardiner Ah right. 11:24.35 kens2 I see we do have a troll present too :-) 11:24.46 paulgardiner :-) 11:25.00 Robin_Watts troll ? 12:10.42 kens2 Chrysoprase 12:10.53 kens2 suspects Robin does not read Discworld 12:11.26 Robin_Watts Not for a while. 12:11.40 kens2 Chrysoprae the troll runs the mafia equivalent, the 'Breccia' 12:12.07 tor8 morning all 12:12.38 kens2 Hi tor8 12:13.26 paulgardiner Hi Tor 12:15.16 sebras tor8: yo! 12:16.47 paulgardiner tor8: I've just made the two changes you requested: clamping vertical movement, and tap to previous/next. I'm just sticking patches in an email. 12:20.36 tor8 great. I'll take a look at that today then. 12:22.48 sebras I noticed a small UI-omission when I tested the android app. 12:29.47 If I search for a word that I know exists then it is highlighted, if I then start to edit the search text it remains highlighted, even if I ultimately search for a word that does not exist. 12:31.08 I think the ios app clears the highlights as soon as the edit field is changed. but tor8 would know the details. 12:31.42 paulgardiner tor8: I also have an idea for improving scroll performance. Not something quick I'd be keen to work into the original quote, and not something I'm certain will work, but at least worth mentioning. 12:32.51 The fact that your removing the textured background doubled the speed, suggests that Android isn't special casing opaque views, and is drawing below them. 12:33.49 tor8 okay, I don't know what sort of agreement you have with henrys but I'm sure he might be open to negotiating 12:34.05 paulgardiner Even with the background taken away, that behaviour of android would slow things down, because the hq patch overlays the lq cache of the page. 12:35.24 If we could get android to draw just what is visible, then it may be very much faster. Views allow their draw method to be overridden, so I believe there is nothing stopping us using our own more efficient algorithm. 12:37.39 tor8 paulgardiner: you mean to override the container/background view draw method? 12:39.22 paulgardiner Yes. Well, to change the draw method for all the views involved in the app. I think there's only two levels at which we'd have to get in. ReaderView and PageView. 12:40.30 All the container views I guess I mean 12:41.08 Against doing that... there may be some way I'm not currently aware of to just tell Android to use a better algorithm. Also I wonder if just building for a higher API level might change behaviour. 12:42.12 Robin_Watts I bet building for a higher API won't solve it. 12:42.41 paulgardiner Probably not. Just a randome thought. 12:42.59 I was just thinking it would be bad to do all that work, only later to find there was a simpler approach. 12:44.16 tor8 paulgardiner: what unit is ReaderView.SCROLL_SPEED given in? 12:45.26 it's a bit too slow 12:45.34 or, well, the deceleration is a bit slow 12:45.50 paulgardiner Can't find SCROLL_SPEED at the moment. The Scroller does have a friction setting though. 12:47.17 Just looking now. 12:47.25 tor8 at the top, it's set to 2. I think you use it as a multiplier to keep the number of pixels to travel proportional with the duration 12:48.16 paulgardiner Helps if I look at the right file :-) 12:48.21 Yeah that's exactly it 12:48.45 And it was determined by trial an error. 12:49.00 tor8 doesn't work well on a 10" tablet :) 12:49.09 grr. damn ant build system is broken :( 12:49.34 doesn't rebuild when things change 12:49.40 paulgardiner :-) No. I suppose it wouldn't 12:49.52 It might be better to just use a constant duration, or perhaps logarithmic on distance 12:50.38 tor8 500ms seems to work rather well 12:53.20 oh, in ios I use a 400ms duration 12:54.16 paulgardiner Ah! I should have looked. 12:54.38 tor8 paulgardiner: I had to go through some contortions there though, since if you tapped again while it was animating the animation layer would go haywire 12:57.50 it'd start a new animation that played at the same time as the old one! 12:58.08 400ms feels a bit too fast though, but that may just be because of the lower framerate 12:59.12 paulgardiner Oh yeah. I didn't give it much testing. 12:59.16 Did you put in a test for whether the scroller was active? 12:59.58 tor8 it just seems to drop some of the taps on the android, which isn't a big deal if the animation is fast enough 12:59.58 I reimplemented the animation at a lower level (create my own animation instead of calling the setContentOffset:animated: method) which solved it 13:01.21 the problem was that the default scroller animation didn't use the real animation layer, so it had all sorts of weirdness going on that the real animation framework would have solved 13:01.57 paulgardiner Oh I see. I thought you meant you had to alter the stuff I sent you. 13:02.33 tor8 no, the stuff I have here from you works quite well at a fixed 400ms duration for the scrolling 13:02.57 paulgardiner Actually the thing I was just worrying about cannot happen because I have just the one scroller instance. 13:04.08 tor8 you mentioned earlier that you suspect it doesn't respect the isOpaque thing and redraws everything, I think you're right 13:04.15 Robin_Watts paulgardiner: Sell unidisp to google! 13:04.40 tor8 when zoomed in a bit so we get the HQ patch on top, the framerate halves again 13:05.08 paulgardiner Robin_Watts: :-) I was just thinking that actually it was your nice recursive painting algorithm we need here. 13:05.26 tor8: Right! I think that proves it. 13:05.48 Robin_Watts tor8: Are you still intending to move fz_obj to be pdf_obj at some point ? 13:06.11 tor8 which should be solved by using opengl and drawing front-to-back with the z-buffer ;) 13:06.25 or just being smart about what regions to blit :) 13:06.35 Robin_Watts: yeah, why not? 13:06.47 paulgardiner tor8: on the downside, I think the speed you get when zoomed right out, and with no background is as fast as we'll manage. 13:07.03 Robin_Watts tor8: I think it's the right thing to do. It will have some knock on effects in the hash table code (which currently relies on fz_indirect and fz_objcmp). 13:07.22 I think I need to rejig that bit now though... 13:07.32 tor8 that looks to be around 30fps (when it's not doing the button layer fade) so should be adequate 13:07.32 paulgardiner Ok. Worth considering then. 13:07.58 tor8 to get it to 60 I'm sure we have to hook into some HW acceleration 13:08.00 paulgardiner I'm surprised Android isn't using HW 13:08.22 tor8 maybe there are some flags to the bitmap class to make it load better as a texture? 13:08.33 they *say* they use HW for the regular ui stuff but given the performance I'm seeing here 13:08.48 Robin_Watts One of the criticisms raised at android was that they couldn't use hardware for the UI stuff. 13:09.10 tor8 ....my old software rasteriser pumping bitmaps to x11 on a 144mhz sparc performed better :( 13:09.13 paulgardiner Strangely, I don't think the HTC Desire I have here varies in speed with the amount of zoom and the presence of the background 13:09.14 tor8 paulgardiner: I think it's probably more than fast enough at that display resolution 13:09.35 Robin_Watts That may have been fixed in the tablet fork (v3) 13:09.44 tor8 quadruple it it starts bogging down 13:09.47 the pixel doubled mode probably uses HW to scale up 13:09.57 Robin_Watts or maybe they fixed it in Froyo and NOT the tablet fork, and the fixes will come to tablets in v4 (ICS). 13:10.15 tor8 because in pixel doubled mode it's smooth as butter, even with the textured back 13:10.35 paulgardiner Yeah, sounds smoother than I see on the Desire from how you describe it. 13:11.16 tor8 huh, switching from zoomed to high res mode makes the tiled background black on the repaint 13:11.54 there's something weird going on with that background! 13:12.02 paulgardiner On an earlier subjet, Scroller has a setFriction method. 13:12.53 tor8 paulgardiner: another thing I noticed, the scroll indicator in the outline view overlaps the last digit on the page numbers 13:18.42 paulgardiner: hm, is there a problem with spaces in filenames when trying to open files from the android file viewer? 13:22.49 I get "Unable to open document" there, but if I drop the file in Downloads I can open it from within MuPDF 13:23.16 paulgardiner Oh. That would explain a file that I was unable to open. Hadn't linked it to spaces in the name. 13:24.22 That's really strange. The mechanisms for triggering the viewer are almost identical in those two situations. 13:25.19 Could be a bug in the file explorer. Would be worth seeing if the same happens from the browser. 13:26.15 Which file explorer are you using? 13:26.42 tor8 "My files" ... something that came pre-installed with the tablet 13:27.23 other apps can open the same file though 13:27.38 paulgardiner Unlikely a bug there then. I think the same is happening with OI File Manager 13:27.50 Looks like getEncodedPath is the wrong thing to call. I'm seeing %20 in place of spaces 13:36.28 Fixed that. I'll send another patch. 13:41.00 Robin_Watts Need to reboot. brb. 13:55.02 tor8 paulgardiner: when you hit "Done" in the search field, shouldn't it start the search? 14:32.29 paulgardiner Does it not? 14:35.45 tor8 nope, it just hides the keyboard but doesn't launch a search 14:36.13 paulgardiner "Done" button on the keyboard? 14:36.49 tor8 yeah, the on-screen keyboard button that says you're finished typing 14:37.52 maybe it's only on tablet devices? 14:37.59 let me dig out my phone 14:38.04 paulgardiner You're right. It doesn't. 14:38.32 I think I looked at that, and it wasn't hard to get working. I'll look again now. 14:39.22 tor8: Another patch in the mail 14:48.55 tor8 is it just me or is the enabled/disabled state of android buttons nigh impossible to tell visually? 15:01.51 paulgardiner: just a thought, could the fact that the buttons view covers the full screen cause the slowdown? one would hope that the transparent area in the middle doesn't get composited 15:11.06 in case it draws each view to a texture with an alpha plane for combining 15:11.25 paulgardiner tor8: I wondered that too. I don't think anything daft would be done when the buttons view is completely visible, but possibly during the fade. 15:13.29 Robin_Watts Rather than fading on/off, could we slide on/off? 15:13.51 IF we think the alpha is causing the problem? (i.e. keep everything solid as much as possible) 15:14.20 paulgardiner Didn't the dade speed up when I changed back to using a direct animation, rather than ViewAnimator? 15:15.12 Is it the fade that's slow? Or is it scolling that slows down when beneath the buttons? 15:15.45 tor8 the fade wreaks havoc with everything on the tablet 15:16.49 let me double check the scrolling speed with the buttons on/off 15:17.10 no visible diff, so it's probably just during the cross fade 15:17.52 paulgardiner Could try just setVisible calls in place of the animation creation.. Shall I make a patch to try it? 15:17.59 tor8 how about robin's suggestion, with a slide instead of fade, or would that be difficult with the android layout things? 15:18.30 paulgardiner Hmmm, would look odd if the whole lot slid on from the side, so I'm guessing you'd want the button bar slide in from the top, and the seek bar/page number from the bottom? 15:21.12 tor8 yeah. I had that from the start with iOS but it messed up their very fragile scroll view setup 15:21.41 panning it in from the top would automatically add offsets to the scroll view underneath it! 15:22.03 paulgardiner Ok. That shouldn't be difficult. I'll have a go. 15:23.52 Did the last two patches seem to work ok? 15:24.35 tor8 paulgardiner: yes, they work great! 15:25.38 I'm fiddling with the icons 15:25.50 the iOS app store automatically adds rounded corners and drop shadows to the launcher, but no such luck for android... 15:26.20 paulgardiner: I tried a heavier file with big images and that impacts the fade a lot. is it triggering relayouts that cause you to re-render the page perhaps? 15:48.04 paulgardiner Sorry, lost me there. I cannot think how big images would make a difference. You mean a PDF file containing big images? 15:50.43 Unless it is shortage of memory as Robin suggested. 15:51.06 tor8 I tried a CBZ file, and there's a very noticeable delay between tapping and the fade starting to happen 15:51.25 Robin_Watts I think tor8 meant he tried a PDF file with large images in. 15:51.27 tor8 and the same happens on pdf:s that take a long time to parse 15:51.39 like ones with large images in them 15:51.49 paulgardiner Must be the background processing impacting on the UI thread 15:52.30 tor8 could be, or something gets hung up on a lock waiting for a page to finish processing 15:53.44 just guessing wildly here, it looks like it may be the rendering of the next/previous pages in the background that impacts it 15:54.06 because it's only after I recently flip to a page that it happens 15:54.24 yeah. If I flip forward so I see the loading spinner, tap in the middle it locks up until the page is ready 15:55.33 paulgardiner It shouldn't be to do with locking: the UI doesn't ever wait on the core lock. 15:55.33 ASyncTask could be running out of worker threads, I guess, but still that shouldn't stall the UI 15:56.08 tor8 the UI is definitely stalled here :( 15:56.25 but it only happens on the animation fade, never anything else 15:56.50 paulgardiner Maybe android uses the same worker thread set for ASyncTask and animations. 16:00.32 henrys tor8, paulgardiner:is there more to do? paulgardiner can just bill us hourly until things are wrapped up. 16:05.40 tor8 henrys: nothing major, but there are a few performance issues on the tablet which don't really show up on a phone device 16:06.16 henrys and we know the performance problems are in the ui vs mupdf. 16:07.52 sorry that was a question 16:09.32 Robin_Watts henrys: Yes. 16:10.13 It's to do with fading controls on/off etc, or scrolling while said controls are on the screen. 16:10.37 It's looking (AIUI) like a flaw in androids underlying redraw behaviour. 16:11.01 henrys well if it makes sense to get paulgardiner a tablet, assuming he wants to work on it, we can pay for the work. 16:11.19 we are saying this does not have a solution in the current incantation of the gui code? 16:12.07 Robin_Watts I believe at the moment there are sufficient unknowns for that not to be clear. 16:12.55 It may be we can fix it by just changing things slightly (say my moving to scrolling things on rather than fading them on). 16:13.22 s/my/by/ 16:13.31 paulgardiner We might get the one to do with fading sorted today. 16:13.33 henrys paulgardiner:well don't work for free just send an invoice to miles and I'll explain the overage 16:14.35 paulgardiner henrys; Thanks. That's brilliant. The little tweeks I've been doing today haven't been a problem, but the idea we have for improving the scroll speed would be a fairly significant change. 16:15.58 henrys which table do you have tor8? 16:17.03 tablet 16:17.08 tor8 samsung galaxy 10.1" 16:17.55 henrys did I mention that I hate doing the meeting agenda? 16:21.56 Robin_Watts Should I buy a samsung galaxy 10.1 and get it to Paul ? 16:23.24 (or should Paul just buy one and add it to the invoice?) 16:24.03 tor8 Robin_Watts: it may make sense to buy an android 4 tablet for testing though 16:24.24 Robin_Watts tor8: I've been looking, believe me :) 16:24.37 I was going to buy an asus transformer prime, but they have been beset with problems. 16:24.56 henrys I agree with tor8 - I think Robin_Watts should buy one and lend it to Paul. 16:25.02 Robin_Watts OK. Currys apparently have Transformer Primes available for next day delivery at 500 quid. 16:27.47 It ships as 3.2 (which is probably the same as tor8's device?) but is promised an ICS update any day. 16:28.39 oh. apparently ICS is out for it already. 16:29.11 tor8 yeah. the galaxy tab is (as usual with samsung) not planned for any updates 16:29.27 Robin_Watts OK, I'm buying it now... speak quickly otherwise... 16:31.08 sebras tor8 Robin_Watts: http://goo.gl/4iYdJ 16:37.58 so yes, the have stated that there will be an ICS upgrade...sometime. 16:38.18 s/the/they/ 16:38.40 tor8 they've said similar things before and not honored them 16:39.35 but we'll just have to wait and see 16:39.41 Robin_Watts, paulgardiner: I've pushed some icon and layout tweaks to casper's pg_android branch 16:40.43 sebras I guess that's the vendors trying to encourage you to buy their new models... 16:41.25 paulgardiner tor8: great. Could you push the patches too, assuming they're all good. 16:42.04 tor8 I added the patches I got from your mail as well 16:42.18 paulgardiner tor8: magic. 16:42.26 I have another I'm about to send - sliding instead of fading 16:42.48 kens marcosw_ ping 16:43.42 Too slow :-( 16:43.48 paulgardiner tor8: on the Desire, the sliding in of the buttons is very smooth. Smother than the doc scroll. 16:47.22 henrys meeting time 16:59.35 kens OK 16:59.41 is marcosw_ really here ? 16:59.48 marcosw_ no 16:59.55 kens Aha. 17:00.01 henrys kens:I think I simplified the raster pattern file all I can. 17:00.01 kens This delightful bug you've wished on me... 17:00.08 marcosw_ kens: I saw you're email regarding the pdfwrite issue. 17:00.11 kens henrys, that's fine thanks, I've been looking at somethign else alll day 17:00.26 henrys np 17:00.35 kens marcosw_ yes that's the one, is that a sufficient work-around, does it work for you ? 17:00.39 marcosw_ the customer is in the process of upgrading to a newer release, but needs a patch for 8.71 as well :-( 17:00.42 kens Can they just use -dUseCIEColor ? 17:00.54 Thyeir current settings don't work properly without it anyway 17:01.10 henrys ray_laptop:I assume you are not going to sf right? 17:01.11 marcosw_ You ask a valid question, I presume they have some reason for using it even though it doesn't make sense. I'll ask them. 17:01.29 kens Best to check if the problem goes away for them too that way. 17:01.44 ray_laptop henrys: right. I discussed it with Miles yesterday. I'll be on IRC during the meeting if anything comes up. 17:02.05 kens THe problem 'appears' to be the colour space (Indexed colour) in the form gets corrupted, I have no idea why and executing 500 times takes an age. 17:02.18 henrys okay new hotel wonder how the internet will be? 17:02.29 Robin_Watts ray_laptop: If you want to skype in at any point, I can share skype details with you. 17:02.32 chrisl marcosw_: 8.71 is unsupported now, isn't it? 17:02.40 ray_laptop If someone can set up a mic and skype me in (no need for visuals) I can hear the discussion 17:02.43 kens 9.0 uses the CMS and so the problem vanishes. Also there are significant changes to colour handling and the way pdfwrite idetifies identical images. 17:02.48 chrisl its not quite 2 years yet :-( 17:02.57 marcosw_ kens: and it only fails under Windows, so none of the tools that I usually use under Linux are applicable. 17:03.01 mvrhel_laptop really? 8.71 is less than 2 years still 17:03.13 kens marcosw_ yes, its a bugger :-( 17:03.17 henrys for this meeting I had questions about a few bug on alexcher_ plate. 17:03.22 marcosw_ chrisl: I think ken is correct, we are supporting 8.71 until the next release in August. 17:03.40 Robin_Watts ray_laptop: See private message. 17:03.49 henrys the portfolio thing seems stalled and the disappearing text patch. 17:03.52 chrisl kens, marcosw_ : ah - still going to hard to back-port the ICC workflow! 17:03.54 alexcher_ henrys: which ones? 17:03.59 kens Its 2 years are up in March I think, or April 17:04.02 chrisl, I'm not sure that's the change that 'fixed' it. 17:04.20 chrisl I thought it was this month..... 17:04.24 kens Seems we released later 2 years ago 17:04.36 mvrhel_laptop can a product go unsupported in the middle of a bug fix 17:05.00 henrys alexcher_:692832 17:05.04 kens I can continue to work on it, but its going to be very slow, for an obselete version of GS. I'd rather spend the time on more producrtive issues if possible. And there is a possible work-around 17:05.39 henrys tor8:anything for the meeting? 17:05.42 tor8 henrys: nope, not really 17:06.46 henrys kens I thought I saw a february month on an 8.71 fileset. 17:06.48 kens henrys, well I could be mistaken, I did a web search and it said it was reeased in March or April as I recall. 17:07.13 alexcher_ henrys: re 692832, the attached patch is the best I can do quickly. 17:07.29 chrisl Version 8.71 (2010-02-10) - From History8.htm 17:07.49 kens http://pages.cs.wisc.edu/~ghost/doc/GPL/gpl871.htm 17:07.54 marcosw_ kens: I was hoping you had a valgrind equiv. for windows that might show something. I agree that debugging by hand would be time consuming. Let me get back to the customer before you spend more time on it. 17:07.59 henrys alexcher:the other one was the portfolio issue, which I thought was easy - adding an option. 17:08.11 kens Which is probably why. THe date at the bottom is 2010-04-13 17:08.15 marcosw_ no, no valgrind for Windows 17:08.23 marcosw_ we released 8.71 on 2010-02-10, so it's been two years. 17:08.25 kens Well, its a paying customer.... 17:08.36 henrys alexcher_:chrisl can you weigh in on alexcher's patch and anyone else? 17:09.02 Robin_Watts kens, marcosw_: I feel I should say "memento" at this point on principle... 17:09.03 kens But like I said, if they can use the work-around of actually setting -dUseCIEColor (like the Gjhostscript message tells them to) then it would be better 17:09.05 marcosw_ Does memento exist in 8.71? 17:09.21 kens Robin_Watts : Not on 8.71 17:09.22 Robin_Watts It can easily be backported. 17:09.29 kens And will almost certainly destory the bug 17:09.39 Its *very* sensitive to the source I find 17:09.47 mvrhel_laptop sounds like -sdUseCIEColor is the answer 17:09.56 -dUseCIEColor 17:10.05 kens Well, it works for me, and it makes more sense than not setting it 17:10.10 henrys marcosw_:I don't think enough effort has been put into telling the customer to upgrade. 17:10.18 kens Since they are setting UseDeviceIndependentColor 17:10.24 mvrhel_laptop then certainly it makes sense 17:10.37 alexcher henrys: the portfolio is simple. I'm adding a loop that will dump the data for all files. 17:10.51 kens OK so I will back off on this until marcosw_ approaches the customer with the work-around at least 17:11.18 marcosw_ henrys: they are in the process of upgrading to 9.04 (or perhaps 9.05), but want a fix for 8.71 as well. I'll push back with leaving off the -dUseCIEColor option. 17:11.32 chrisl henrys: alexcher and I discussed it originally - give that we're dealing with a broken file, and we don't know for sure what CPSI does, I think Alex's is a reasonable heuristic to work around the issue. 17:11.35 henrys alexcher great I didn't know if it had been dropped since we discussed it quite some time ago. 17:11.48 kens marcosw_ If I do have to debug this, it will be a long while before they get a fix, that may be information affecting the customer's decision about the work-around.... 17:11.50 marcosw_ THey need to *add* -dUseCIEColor, not leave it off 17:12.13 marcosw_ kens: right. 17:12.28 henrys chrisl:sounds like we should commit the fix. 17:12.28 chrisl henrys: the other option would be to spend a indeterminate amount of time experimenting with various broken type 3 fonts in CPSI, which I really don't think it worth our time 17:13.29 alexcher marcosw_: Purify is a commercial memory checker for Windows. It's now distributed by IBM. 17:13.45 kens alexcher, will it work with our garbage collector and memory management ? 17:14.05 marcosw_ Yes, I've used Purify in the past. Do we own a copy? 17:14.07 henrys chrisl:I don't think that is warranted. 17:14.08 marcosw_ It used to be very expensive but worked really well. 17:14.22 kens I'm not convinced Purify will woork in this case 17:14.27 Robin_Watts It's still very expensive. 17:14.35 kens Its memory being managed by GS 17:14.36 ray_laptop usually the GC 'reloc' totally confuses most memory tools 17:14.47 but if you can get it to fail with every object in its own GC chunk, then the reloc is out of the picture 17:15.23 henrys marcosw_:any customer issues you think need attention? 17:15.37 beyond what we are talking about? 17:15.48 kens ray_laptop : I'm willing to try that, but have no idea how. My guess is it will not fail 17:15.50 alexcher marcosw_: Last time I checked it was below $1000. I tested it on Ghostscript but prefer Valgrind. 17:16.07 henrys and mvrhel_laptop:are you setup with alexcher and output intent or should we discuss? 17:16.34 mvrhel_laptop I have not heard from him about it 17:16.44 chrisl http://code.google.com/p/drmemory/ might be worth a look 17:16.47 mvrhel_laptop I was going to ask at the end of the meeting 17:16.52 marcosw_ henrys: it would be nice if mvrhel_laptop could take a look at http://bugs.ghostscript.com/show_bug.cgi?id=692865, it's a regression. 17:17.00 henrys mvrhel_laptop:also you were going to move the user param code into the graphics library. 17:17.01 mvrhel_laptop henrys: yes. I started that and then last week got cut short for me. I will get it done before the meeting 17:17.28 ray_laptop is sucking me into customer 532 stuff.... 17:17.42 henrys great. 17:17.46 more 532 stuff? 17:17.55 mvrhel_laptop screen creation.... 17:18.19 henrys I thought you made all these great tools that lets the user make those. 17:19.01 marcosw_ there is a partially functional purify trial that I might tryâ¦ The retail version is$951 but I believe that only includes a 1 year license (it might be an indefinite license with one year of support, I didn't read the small print). 17:19.42 henrys so alexcher it seems like we have a go ahead for committing your patch for missing type 3 text. 17:20.03 Robin_Watts marcosw_: It's probably 1 seat, non-floating too though :( 17:20.04 mvrhel_laptop yes. well I will let ray_laptop explain it all but apparently there is some political issue between japan and the US engineers about a particular issue which is likely due to them using screens with a transfer function which is causing the loss of levels in the light regsions 17:20.37 ray_laptop kens: setting -M1 will set the chunk size to 1K. Other than that you can zap psi/imainarg.c to make the -M set it smaller 17:21.03 mvrhel_laptop so it may be easy enough for us to go ahead and create some screens that demonstrate that this is the issue 17:21.15 i.e. which have the same embedded transfer function but more levels 17:21.42 henrys mvrhel_laptop:okay 17:21.49 marcosw_ Robin_Watts: true 17:21.55 ray_laptop henrys: I asked mvrhel_laptop to use his color screen generation tool to make screens that are similar to theirs but that have more levels 17:21.56 kens ray_laptop : these are pretty small string, indexed RGB colour spaces, but that might do it. 17:21.57 mvrhel_laptop alexcher: so did you get a chance to add in the interpreter parts for the output intent? 17:22.43 ray_laptop kens: is this the thing where the palette is getting polluted ? 17:22.56 alexcher mvrhel_laptop: not yet. 17:23.35 kens ray_laptop : yes, the colour space being written is from an /Indexed space, and the space is incorrect. 17:23.41 ray_laptop kens: it looks to me like someone is keeping a pointer to the palette data that doesn't track when the GC reloc's the object 17:23.43 kens It has been corrupted 17:23.46 henrys I'm concerned 532 is getting too much priority at the expense of other customers. It just seems non ending. 17:23.55 kens ray_laptop : We are comparing an old colour space to the new one. We write a new space if they don't match, using the current colour space. 17:24.11 So its the current space which is corrupted 17:24.18 ray_laptop kens: when I looked at the palette data at one point it contains string data 17:24.24 kens I get somewhere between 6 sand 12 warnings in gs_vmrecalim 17:24.41 But I don't know what they are telling me 17:24.55 marcosw_ I realized that I hadn't tried valgrind on the 32 bit build under Linux (I constantly forget that the default for linux is 64 bit and for windows it's 32 bit). There are some warnings produced that don't come up with the 64 bit build that may be related (i.e. in pdf_indexed_color_space()). I'll reassign the problem to me to continue this line of inquiry. 17:25.20 kens FWIW the problem persists with -M1 17:25.21 ray_laptop kens: what kind of warnings ? 17:25.28 henrys it does seem to me alexcher is a bit swamped again and him teaming up with kens or possibly chrisl again would be worthwhile. Of course that would stall what kens or chrisl are working on. Thoughts? 17:26.04 kens ray_laptop : String pointer 0x....[765] outside [0x...0x...) 17:26.18 And yes, the brackets are of different types 17:26.40 The warnigns are on line (15) of psi\igcstr.c 17:26.58 (153) sorry 17:27.07 ray_laptop kens: ahh -- that definitely sounds like a hit. So is the string the palette ? 17:27.07 kens It looks like it 17:27.15 chrisl henrys: I'm happy to help out with Alex's queue, but probably won't have much time for a bit 17:28.20 ray_laptop kens: so you need to see what structure has that pointer to the string. If you set a breakpoint, you can see if it is in an 'enum' or 'reloc' stage 17:28.29 kens One minute ray I had to rebuild for my breakpoint to work 17:28.57 henrys Robin_Watts:could you take a look at the xps crash for mvrhel_laptop if that makes sense to both of you? 17:29.17 kens Its a gs_gc_reclaim, does that help ? 17:29.28 ray_laptop kens: if you go up in the call stack, then you get to a point where the 'pre' object is visible and you can look at its type 17:29.29 mvrhel_laptop I was just digging into that one 17:29.31 Robin_Watts henrys: I could, but I'm right in the middle of some mupdf stuff at the moment. 17:29.43 henrys mvrhel_laptop:okay nvm 17:29.45 Robin_Watts Can it wait til after the meeting ? 17:29.50 oh, I'll leave it to mvrhel. 17:30.01 henrys Robin_Watts:sounds like mvrhel_laptop is on it. 17:30.15 Robin_Watts The MuPDF customer has 2 problems at the moment. 17:30.20 kens ray_laptop : Can't see a 'pre' object. 17:30.30 ray_laptop mvrhel_laptop: BTW, thanks for responding to the customer about the FastColor option 17:30.44 Robin_Watts 1) They are concerned about speed (something is up with the profiles of the new code on at least one particular test page) 17:30.48 mvrhel_laptop no problem 17:30.51 Robin_Watts 2) They need image decompression on demand, so I've done most of that. 17:31.05 ray_laptop kens: what's the (top of the) call stack ?? 17:31.13 henrys Robin_Watts:this is the company you recently had the conference with? 17:31.34 Robin_Watts Yes. 17:31.42 kens interp->interp_reclaim->ireclaim->gs_vmreclaim->context_reclaim->gs_gc_reclaim->gc_trace->gc_string_mark 17:32.10 henrys okay 2 minutes past meeting over... I'll be around most of the day. 17:33.09 kens It appears to be handling local VM ray 17:33.14 Hmm, interestingly, unlike last time, this is now pointing at uninitialised memory. 17:34.51 ray_laptop kens: sometimes it helps to run with -Z@$? 17:36.51 marcosw_ tor8: are you flying through Frankfurt? The ground controllers are striking and my brother's flight this morning from Frankfurt to the US was cancelled, he had to re-route via LHR. 17:36.59 kens ray_laptop : I can try thast but I doubt I will understadn the reams of output. Give me a minute 17:37.17 ray_laptop kens: -Z@$? doesn't (usually) generate any output 17:37.41 kens Well it made the GC problem go away 17:38.18 ray_laptop the '?' causes pointers to be validated and sometimes generates messages when an object isn't defined in any chunk 17:38.29 kens File is still broken 17:38.35 Image comes out flat gray now 17:38.50 Goes wrong much earlier though, o page 2 that's helpful 17:39.54 ray_laptop kens: try it without the -Z@ the '@' fills memory when allocated/freed 17:39.59 kens so -Z$17:40.37 ray_laptop kens: but if it is failing that quickly and well defined with -Z@ it should be much easier to track down. 17:40.38 kens: or -Z$? 17:40.50 kens -z$produces more gc errors 17:41.00 ray_laptop kens: this is with 8.71 ? 17:41.01 kens Yes, all with 8.71 17:41.10 HEAD works fine 17:41.18 Taking away the '@' gives more errors, and the images are back to corrupted again (instead of the same all the time) 17:41.54 ray_laptop kens: I'll take a look at it. I've been in memory hell before and know some tricks. It's probably getting late for you. 17:42.04 kens I think I'll use -Z@$ because it goes wrong much faster 17:42.10 ray_laptop builds 8.71 17:42.28 chrisl kens: I fixed a garbage collector problem, IIRC, strings in indexed color spaces 17:42.29 kens ray_laptop : I'd be grateful, but its hard to get it to fail, you need a pretty exact set of conditions 17:42.32 chrisl that could be it. 17:42.40 Do you remember the bug/commit hash ? 17:42.51 chrisl give me a minute, I'll try to find the commit 17:42.52 kens is still updating the laptop.... 17:43.09 ray_laptop chrisl: I looked in the log and couldn't find it, but I also (vaguely) recalled something with palettes 17:43.57 chrisl kens: http://git.ghostscript.com/?p=ghostpdl.git;a=commit;h=0bd3e960b37258e0dd1a99f5110783a37d479e70 17:43.58 kens chrisl, thanks I 'll try that. 17:44.20 ray_laptop chrisl: thanks -- I bet that's it. 17:46.21 kens Just rebuilding now 17:46.29 No GC errors 17:47.20 And output is OK. Looks like that was it 17:47.38 chrisl Yay! 17:47.43 kens Thanks chrisl 17:47.49 marcosw_ you can send that patch to the customer too. 17:48.00 chrisl I can't remember how big that patch was 17:48.06 kens Though you might like to try it first as well 17:48.09 Its only a few lines 17:48.16 add about 12, take 1 away 17:48.28 chrisl Excellent! 17:48.33 So, how come I could remember that, but I can't remember today's date....? ;-) 17:49.37 kens Specialised memory 17:49.46 Anyway, thanks for that, I can go back to Henrys raster pattersn tomorrow and on the plane now. 17:50.03 Dinner time, goodnight all 17:50.12 henrys nice chrisl 17:50.18 marcosw_ kens: I was working on my paper to "Predict drug function via cytological profiling" (one of the aims of my dissertation), so have no idea what patch I need to send to which customer. 17:50.19 chrisl marcosw_: http://git.ghostscript.com/?p=ghostpdl.git;a=commit;h=0bd3e960b37258e0dd1a99f5110783a37d479e70 17:50.42 kens The customer iwht the corrup[ted images 17:50.52 #692868 17:51.03 THe patch above fixes it. 17:51.10 marcosw_ kens: I thought we had given up on that? 17:51.13 kens marcsw_ 17:51.21 chrisl remembered fixing the actual problem 17:51.28 marcosw_ that darn chrisl :-) 17:51.40 kens I have no idea whay this does not show up with bisect, except that being memory, it shifts around when you change the code. 17:51.54 I still say the customer should use -dUseCIEColor though 17:52.13 Because of the other parameters they have set 17:52.24 marcosw_ Making me look foolish, I just emailed the customer saying that a fix would take weeks and now I need to mail them, never mind, here it is? 17:52.27 kens But its good to give them a real fix for the memory problem too. 17:52.39 You can say that we were lucky and had already fixed it, it just took time to find the fix. 17:52.57 Anyway, sorry ;-) 17:53.04 And again, night all. 17:53.30 chrisl marcosw_: to be fair, it took me several days to track down the problem originally, so you weren't entirely wrong in saying it could take some time to fix! 17:53.46 ray_laptop marcosw_: just tell them that you were estimating as if you had to fix it instead of the engineer that handled it ;-) 17:53.58 marcosw_ ray_laptop: I would be more insulted if you weren't correct ;-) 17:54.35 chrisl marcosw_: your other option is to wait a couple of weeks before sending them the patch :-) 17:55.13 marcosw_ Or I could take a page from Adobe Acrobat support and make them reinstall Ghostscript after wiping their drive and reinstalling windows. 17:55.54 chrisl If you really want to follow Adobe's policy, tell the customer 8.71 is working as intended, and change the documentation to describe what 8.71 is doing..... 17:57.07 mvrhel_laptop hehe 17:57.40 I like the uninstall reinstall suggestiong 17:58.04 suggestion 17:58.10 marcosw_ or tell them to upgrade to 9.05, since Adobe only appears to support the latest version of anything (which is doubly annoying since they charge for upgrades). 17:58.29 mvrhel_laptop alexcher: so do you have any idea when you are going to be able to get to the output intent 18:01.53 we can talk about it some at the meeting if you don't understand what needs to be done 18:02.20 tor8 marcosw_: munich 18:02.29 alexcher mvrhel_laptop: I'll try to finish before the meeting. 18:03.23 mvrhel_laptop alexcher: ok great! 18:03.50 alexcher Currently I'm puzzled by the bug 692852. The workaround introduced by 690786 covers almost everything, but 1bpc indexed color space is handled by AR9 differently. Should I add a special case? 18:04.24 marcosw_ Lufthansa MUC -> SFO? That should be a pretty nice flight, I took SFO->MUC 10 days ago and it was the emptiest flight I had been on in years. The flight attendant said that SFO<->MUC is usually only half full this time of year (Lufthansa used to only fly this route 3 days a week in the winter, I don't know why the changed to daily since the demand clearly isn't there). 18:05.34 henrys alexcher:we assigned that to Robin_Watts, why are you working on it? 18:07.33 alexcher henrys: because my previous patch attempts to fix a similar problem and I've got some insight into the problem during initial analysis. 18:10.36 ray_laptop mvrhel_laptop: I took a look at cust 532's xfer functions. While their input screens have all 256 levels, the TF _really_ compresses things. The 1200 1 bit starts with 0, 5, 10, 15, 20, 25, 30, 35, 37, 38 18:11.17 henrys alexcher:I thought it would be useful to have a second point of view and you are backed up on what is already on your plate. 18:13.54 if you insist on working the problem then please change the assignment back. 18:14.48 alexcher henrys: I'm posting a comment about AR9 properties. This bug is about reproducing how AR handles broken /Mask. 18:17.01 mvrhel_laptop ray_laptop: oh wow. sorry I have not gotten to this yet. 18:18.38 tor8 marcosw_: yup, that's the one 18:18.46 mvrhel_laptop I will start on it shortly 18:18.46 henrys alexcher:all I'm saying is if you are going to work on it assign it to yourself and if not assign it to someone else, everyone is very busy and duplicating work needs to be avoided. 18:20.17 chrisl alexcher: it looks like xpdf and poppler drop the Mask completely if the sample values in the mask don't make sense with the rest of the image properties - no idea if that tallies at all with AR behavior. 18:20.33 Robin_Watts So, Fujitsu are tightwads. Bug 692860. 18:24.36 alexcher chrisl: no AR uses the given number modulo the correct range. 18:24.57 ray_laptop mvrhel_laptop: but I don't see how that would cause levels to be mapped to the same ht pattern 18:24.59 alexcher chrisl: This is coded now, but 1 bpc indexed space seems to be an exception. 18:25.38 tor8 paulgardiner: moving the setBackgroundResource(R.drawable.tiled_background) to the layout rather than the mDocView (the layout that contains both the document view and the buttons view) solved the flickering glitches! 18:26.16 chrisl alexcher: okay, that does make some sense I suppose - lord, I do hate Acrobat :-( 18:26.26 henrys Robin_Watts:yes rather odd... I suspect he just wants a viewer to view pcl and they aren't using our product in any direct way. 18:27.12 chrisl henrys: if that's the case, he could at least be a bit nicer about it! 18:27.55 Robin_Watts Is it worth following up with a reply saying "Well, your bug will be dealt with at the free user (low priority) level then. I assume therefore that you are using GhostPCL in a GPL conformant manner." 18:28.02 alexcher chrisl: In the latest file AR drops the mask it the first element is invalid. If the 2nd element is invalid, it'd replaced by 1. 18:28.41 Robin_Watts tor8: Excellent! 18:29.19 henrys Robin_Watts:I'd like to let him sit for a bit... Miles has already read him the riot act. 18:30.01 Robin_Watts henrys: fair enough. 18:30.12 chrisl alexcher: like I said, hate Acrobat! I find it hard to believe that's "intended behavior"...... Sounds like you've got a good idea how to proceed, though. 18:31.13 ray_laptop mvrhel_laptop: the 1200 dpi screen is really screwy. Even though it is 24x24 they have LOTS of missing values (58 out of 256) pretty crappy screen if you ask me 18:31.23 henrys Robin_Watts:I do think of sending an email to corporate headquarters - this yahoo is waving your company name around and asking for free support, I would assume somebody at the company would frown upon that. 18:31.35 or maybe they'd just think him enterprising. 18:32.05 ray_laptop marcosw_: BTW, there is a utility too scan a FONTPATH once and generate a Fontmap that directly maps the fonts found. This is recommended on machines with lots of fonts (like I have) when you want to use -sFONTPATH. See toolbin/genfontmap.ps 18:56.31 marcosw_ ray_laptop: thx. I recalled something like that, but I wanted to see if this was problem before looking around. I'll tell the customer. 18:57.33 ray_laptop marcosw_: if you use that fontmap, you can still run with -sFONTPATH and it will only go there when it can't find a font in the Fontmap, but if the font is not found at all, then it will still scan, fail and substitute. I recommend leaving it off once the fontmap is generated. 18:58.28 marcosw_: if they run without the -q then the stdout will have something like "Scanning ..." 18:59.39 marcosw_ they aren't using -q. 19:00.43 ray_laptop did they send their stdout log ? 19:01.16 marcosw_: also -Z: would be interesting since it shows which page is causing the delay (if any) 19:02.10 marcosw_ no. 19:02.10 ray_laptop hmm... debugbin/gswin32c -r10 -dJOBSERVER - < comparefiles/Bug688308.ps gives a BUNCH of GPL Ghostscript GIT PRERELEASE 9.06: .\psi\igcstr.c(153): String pointer 0x2595e42[243] outside [0x2595e52..0x2596028) messages 19:07.19 mvrhel_laptop I am surprised bug 692865 was not caught by the regression testing 19:09.17 ray_laptop mvrhel_laptop: there isn't a command line -- does it fail when generating ppmraw or just to the 'display' device ? 19:11.42 mvrhel_laptop I am just running to tiff24nc. it fails then. I have found the issue and should have a fix in a bit 19:12.13 it has to do with some confusion on the color spaces that the xps interpreter sets up for defaults 19:12.30 ray_laptop mvrhel_laptop: I guess we have a new file for the regression testing :-) 19:12.39 mvrhel_laptop icc vs. devicegray etc 19:12.41 Robin_Watts Hey scott-san 19:23.52 scott-san Hi Robin. Was going to check with Tor or one of you guys to get the latest release number for MuPDF. 19:24.26 Robin_Watts 0.9 19:24.32 1.0 is due within a month. 19:24.38 scott-san Great just what I need. I'm sending out an OEM Agreement and neede the proper info. Thanks, see you Friday! 19:25.07 Robin_Watts cu! 19:25.21 ray_laptop mvrhel_laptop: I was reading the transfer function backwards -- the transfer functions they have are DEFINITELY compressing levels in the lighter shades. The 600 dpi 1-bit at input level 182 and above it maps tonly 38 values -- roughly every other value 19:27.02 mvrhel_laptop ray_laptop: ok. hopefully our screens can address this. I am getting ready to clusterpush the fix for this xps bug then I will get to work generating a screen. may grab some lunch first though 19:28.02 ray_laptop mvrhel_laptop: no problem. I'm going to send Len an update with what I see from my static analysis. 19:29.08 mvrhel_laptop ray_laptop: so which transfer function should I use of the ones that Len sent 19:49.11 TF_600_02LV.ps 19:50.29 that one goes with the 16x16 screen 19:52.36 ok I see 19:53.19 nevermind. 19:53.35 henrys ray_laptop:forgot to ask at the meeting should 692602 be reopened for a doc fix? 20:03.01 I guess chrisl_away is managing the man pages. 20:03.16 no pun intended. 20:03.28 ray_laptop henrys: sorry -- I was on the phone. Let me look at that one 20:06.41 mvrhel_laptop bbiab 20:08.49 ray_laptop henrys: it looks like the man pages just never got updated from the doc/Use.htm 20:09.26 henrys: so, if you want to re-open for chrisl to manually manage the man pages, that seems reasonable 20:10.10 henrys ;-) will do 20:10.27 ray_laptop IMHO we should just do away with the man pages and just have one that says "Use your browser and open doc/Readme.htm" 20:11.17 there's probably quite a bit of stuff in the man pages that is out of date 20:11.55 henrys I certainly don't like the duplication I imagine there would be a backlash to deal with though. Some people get all serious about man pages. 20:12.35 ray_laptop another "hot" issue from cust 532 :-( 20:44.58 rhce7320 Has anybody know of a print shop producing press plate bitmaps with ghostscript? 23:29.25 sebras Robin_Watts: hm.., concerning fz_bitmap_s is it ever the case that we have more than one component per bit? 23:45.19 Forward 1 day (to 2012/02/22)>>>
 ghostscript.com Search: