IRC Logs

Log of #ghostscript at irc.freenode.net.

Search:
 <<<Back 1 day (to 2011/07/20)2011/07/21 
mvrhel2 damn compile_inits = 0 doesnt seem to work00:14.56 
  sign00:15.00 
  I will try on windows00:17.41 
  ok. so it is yet again my fix that breaks compile_inits = 0 . This really should be tested on the clusterpush and put at the top of the report along with the new warnings04:02.29 
  I only happened to catch it to do some testing for ray's recent commit04:03.15 
henrys it seems like having the hardwired default of ROM0 causes problems I wonder if the search were the same in both configurations if you wouldn't catch the problems in the rom0 build.04:08.20 
  but yes we should test it.04:08.50 
  I'll work on the x11cmyk problem I just assigned to you I assume you won't object (unix and all) ;-)04:12.24 
mvrhel2 henrys: ok thanks.05:18.29 
  ok got the inits issue fixed. now I am going to do a clusterpush of the gray to K fix05:20.35 
ray_laptop interesting. Some new guy at HP (presumably in india) doesn't realize that the gs-bugs mailing list isn't private and sent a user ID and password :-/ Maybe one of us should tell them ??06:32.51 
kens chrisl I think I have an example fo the WMode problem you fixed yesterday, but in a PostScript file this time.07:49.47 
chrisl kens: It really shouldn't be the same problem.07:52.06 
kens Well, I think I'll let you decide :-)07:52.20 
chrisl Okay, punt it over.......07:52.47 
kens The file has text which is displayed in one direction, but when convrted to PDF with Distiller, the selection runs at 90 degrees, just like yesterdays PDF file.07:52.49 
  When I look in the file it has a CMap with WMode 107:53.01 
  Its the problem from Peter this morning, sent to support.07:53.15 
chrisl Okay, let me have a quick look07:54.03 
kens Ian sent me this link this morning:07:56.05 
  http://uk.reuters.com/video/2011/07/20/no-parking-especially-when-a-manhole-eru?videoId=217290140&videoChannel=7707:56.06 
chrisl That outs one of our burst water mains into perspective!08:08.32 
  s/outs/puts08:08.39 
kens There's also an amusing 'dog bites shark' video on there08:08.55 
chrisl kens: I don't understand what's going on with the file from Peter - we're *not* losing the WMode value from the Postscript, it is correctly set in the PDF, but the character spacing isn't correct.08:11.07 
kens You don't need to go to pdfwrite, you can see the sme result when rendering08:11.31 
chrisl I am. I've also just discovered that if I set the WModes in the PS files all to "0", the file renders "correctly".08:12.39 
kens That's bizarre.08:12.52 
  I didn't try that, I just assumed that it was the same problem08:13.06 
  When I saw the WMode 108:13.13 
chrisl I'm just about to check what Distiller makes of the file with WMode all zeros08:14.05 
kens Good idea!08:14.17 
chrisl But I'm having to reboot my VM.....08:14.25 
kens Hmm, the file isn't *that* comlicated.... ;-)08:14.41 
chrisl No, but Vista in a VM is *that* dumb!08:16.22 
  Strange, changing the WMode in the file makes *no* discernible difference rendering with Jaws08:17.52 
kens Now that is indeed interesting. Suggest that I'm barking up the wrong tree with the WMode08:18.18 
chrisl Hmm, I think we are using the *wrong* CMap - I think we end up using the "-V" CMap, instead of the "-H" one08:20.00 
kens Hmm, can't think how that happens.08:22.49 
chrisl No, I can't either08:23.45 
  Hmmm, Windows didn't reboot, it shutdown.....08:24.29 
kens I think I'm going to get cross with Rahul Sabnis soon.08:27.11 
chrisl Who be he, then?08:27.47 
kens The idiot complaining that they get a broken PDF file after they mail it, and so it must be a Ghostscript bug.08:28.08 
  #69236308:28.23 
  Actually, since he's said it publicly, he's from HP, probably in India.08:28.45 
  The Daziel buffoons I expect08:28.56 
  Assuming they are even a commercial customer at all08:29.12 
chrisl Ah, yes, not being too bright!08:29.36 
  kens: that document(s)PS.ps: changing the WModes makes no apparent difference in either Jaws or Distiller, but one way GS works, and the other it doesn't - colour me confused!08:36.26 
kens That's weird. Not least that changing the WMode makes no difference to other interpreters08:37.03 
  That suggests its not being used by Jaws/Distiller, but is (apparently incorrectly) by GS.08:37.24 
chrisl yeh, but that doesn't make much sense either :-(08:38.32 
kens No it certainly doesn't08:38.41 
  I'll open a bug report for it and get back to the customer08:45.08 
  OK chrisl #692365, I've assigned it to you. For somre reason I still seem tobe getting 'text' bugs.09:16.50 
chrisl Oh, gee thanks! Cutting that file down is going to be a pain :-(09:18.06 
kens I know, sorry about that, but I think it si your area.09:18.23 
chrisl I think I'm going to stay away from customer contact whenever possible........09:24.30 
kens I'd advise it I have to say.09:24.43 
  Curisouly the HP people haven't come back since I asked if they were a commercial customer.09:25.15 
chrisl I'm sure they will come back, and even more indignant than before!09:26.01 
  And, to make the day complete, my parents' car won't start, so I have to go over in a while and help get it running.09:26.14 
kens Oh joy :-(09:26.22 
chrisl I'm glad I bought that batter booster!09:26.48 
  s/batter/battery09:26.54 
kens "Mmm, batter booster" says Homer.09:27.15 
chrisl It must be that Scottish heritage coming to the surface....... :-)09:27.48 
  Right, I'm off to try to get an MX-5 to run - not sure when I'll be back.......10:18.57 
nitin Hi, i tried to compile the mupdf source but it is giving me this error.... LINK build/debug/mupdf build/debug/x11_main.o: In function `winopen': /cygdrive/d/mupdf/apps/x11_main.c:123: undefined reference to `_XOpenDisplay' /cygdrive/d/mupdf/apps/x11_main.c:127: undefined reference to `_XInternAtom'10:25.40 
  please help10:25.54 
kens Looks like you don't have X11 installed.10:26.20 
nitin oh... i am using cygwin10:26.52 
kens Does cygwin support X11 ?10:27.10 
nitin so in my case, do i need to install x11 for cygwin10:27.18 
kens I don't use cygwin so I can't really be sure.10:27.40 
  But the error is in X11_main, and is an undefined reference to _XInternAtoim so I would say it needs something to support X1110:28.11 
nitin ok..thanks for the clue10:28.49 
  i try to get x11 support in cygwin then will get back to this channel10:29.16 
kens tor or Robin might be able to help better than me10:29.29 
nitin since tor is not available right now... i am waiting... thanks for ur suggestion Kens10:30.21 
Robin_Watts nitin: Hi.10:48.22 
  This is presumably so you can get the generated directory?10:48.43 
nitin Hi Robin10:50.46 
  i do get the genarated directory but when i am running ndk-build them it is giving me some error...part of it was pasted above10:53.01 
Robin_Watts I can't see any errors pasted above that would have come from the ndk-build.10:53.41 
nitin i closed the window... i'll repost it10:54.26 
  $ make -f makefile LINK build/debug/mupdf build/debug/x11_main.o: In function `winopen': /cygdrive/d/mupdf/apps/x11_main.c:123: undefined reference to `_XOpenDisplay'11:16.08 
  when i am trying to compile the mupdf source, above error (it is just a part of the whole) comes... 11:16.56 
Robin_Watts I don't see how that can be part of the ndk-build errors.11:19.51 
  Are you following the steps in ReadMe.txt in the android directory ?11:20.28 
nitin yes...11:29.37 
  it is the step 9, where i am getting this error11:30.35 
  i tried to compile the mupdf source using cygwin11:30.59 
Robin_Watts 9) Change into the android directory, and edit local.properties into your11:32.28 
  favourite editor. Change the sdk path there as appropriate. This should be11:32.30 
  the only bit of localisation you need to do.11:32.32 
  ?11:32.35 
nitin i am trying to compile and then that error came up...though now, i have these headers availabe in generated directory - cmap_cns.h, cmap_gb.h, cmap_japan.h, cmap_korea.h, font_base14.h, font_cjk.h, font_droid.h11:34.04 
Robin_Watts nitin: I am concerned that your step 9 may not be the same as mine.11:34.30 
  And as such therefore you may be using an old version.11:34.40 
nitin it is same11:34.44 
Robin_Watts So how can you be getting that error in step 9? Do you mean step 8 or step 10 ?11:35.10 
nitin but as what tor8 suggested me to compile the mupdf source in order to get the generated directory files since it is not available for download11:35.50 
  this error came up only when i tried to compile the source11:36.20 
Robin_Watts Right. That should be step 8.11:36.39 
nitin yes...11:36.46 
Robin_Watts If you've got the generated directory (and it sounds like you have) then ignore the errors and move on to step 9.11:36.58 
nitin $ make -f makefile LINK build/debug/mupdf build/debug/x11_main.o: In function `winopen': /cygdrive/d/mupdf/apps/x11_main.c:123: undefined reference to `_XOpenDisplay' /cygdrive/d/mupdf/apps/x11_main.c:127: undefined reference to `_XInternAtom' /cygdrive/d/mupdf/apps/x11_main.c:128: undefined reference to `_XInternAtom' /cygdrive/d/mupdf/apps/x11_main.c:129: undefined reference to `_XInternAtom' /cygdrive/d/mupdf/apps/x11_main.11:37.16 
  am i doing anything wrong here?????11:37.45 
Robin_Watts 9) Change into the android directory, and edit local.properties into your11:38.03 
  favourite editor. Change the sdk path there as appropriate. This should be11:38.05 
  the only bit of localisation you need to do.11:38.07 
  ok, no possibility for errors there.11:38.19 
  10) Change into the android directory (note, the android directory, NOT11:38.31 
  the android/jni directory!), and execute (in a Cygwin window on Windows!):11:38.34 
  ndk-build11:38.35 
nitin ok.... so i don't need to worry about these errors..right11:38.36 
  ok...11:38.43 
  thanks11:38.46 
  ok.. i am moving on to step 9 now then....wish i donn find enough errors11:39.59 
  Compile thumb : mupdfthirdparty <= jbig2.c11:46.48 
  In file included from D:/mupdf/android/jni/../../thirdparty/jbig2dec/os_types.h:53,11:46.56 
  from D:/mupdf/android/jni/../../thirdparty/jbig2dec/jbig2.c:22:11:47.08 
  D:/android-ndk/platforms/android-8/arch-arm/usr/include/stdint.h:48: error: redefinition of typedef 'int8_t'11:47.23 
  D:/mupdf/android/jni/../../thirdparty/jbig2dec/os_types.h:47: note: previous declaration of 'int8_t' was here11:47.39 
  now when i am performing step 9... i am getting those errors11:49.16 
Robin_Watts step 10, but never mind :/11:49.37 
  Let me look.11:49.43 
nitin yes step 10... 11:50.21 
Robin_Watts What version of the ndk are you using ?11:52.09 
nitin Android NDK, Revision 5c (June 2011)11:53.46 
Robin_Watts ok, that's the same version I have here.11:53.59 
nitin good for me11:54.54 
Robin_Watts trying a clean build now.11:55.12 
  ok, that worked.12:00.00 
  I wonder if I have uncommitted changes in thirdparty.12:00.11 
nitin if it has worked for you....then what i need to do here to get it right????12:01.45 
Robin_Watts right. Load thirdparty/jbig2dec/os_types.h into an editor12:03.11 
  Then change line 43 from: #else to #elif !defined(HAVE_STDINT_H)12:03.51 
nitin great...12:05.53 
  i made the change12:06.03 
Robin_Watts That should solve it. I'll try and get that pushed back into the original repo. Sorry about that.12:06.06 
nitin no sorry... you helped me here...12:06.34 
Robin_Watts chrisl: We should look into photoshop group licenses.12:07.59 
nitin Robin_Watts: It worked for me.... THANKS A LOT12:10.37 
Robin_Watts nitin: Fab.12:10.46 
  actually, has anyone used photoshop elements? Is that going to be too crippled for what we want ?12:11.43 
  53 quid seems a much saner price to pay.12:12.05 
kens Hmm, SaGS has a bug report with the ICC profiles 'fix'.13:11.32 
  ray_laptop : ?13:27.43 
Robin_Watts Right. Unicode crap...13:51.42 
  ray made the interesting observation that the title bar on the Ghostscript Image window is OK. I might have a quick look at that...13:53.20 
kens Odd13:53.37 
henrys kens:I don't think we give customers the number - I might be mistaken but we always just ask if they are customers.14:32.35 
kens Oh, OK, do you know i they are ?14:32.49 
henrys no I do not14:33.02 
kens I was guessing they were the same Indian bunch as usual ;-)14:33.14 
  But maybe not....14:33.20 
  In which case, maybe we should check how htey are using GS >:-)14:33.45 
henrys I know there was a recent "cloud" customer that sounds like a possibility didn't seem like the Indian bunch.14:33.46 
kens I looked up the product, its a document converter thing.14:34.05 
  But that odesn't really narrow it down any.14:34.27 
  Its been a busy day on support :-(14:34.41 
  I think we have two new blockers for release.14:34.49 
henrys regressions?14:35.00 
kens Yes.14:35.05 
  SaGS's patch for the icc profiles fix.14:35.16 
  And 128-bit AES seems to have stopped working (decryption)14:35.29 
  reports from 2 different customers (but both referencing the file from one of our bug reports)14:35.54 
henrys haven't finished my mail yet just got through the knuckleheads' message14:36.14 
kens :-)14:36.26 
henrys kens those two customers work together.14:40.37 
kens Ah!14:40.44 
flemic hello14:51.54 
  somebody there?14:52.16 
kens Yes, severl somebodys14:52.30 
flemic hi, a have a problem. is it possible to read box informations of a pdf file with ghostscript?14:53.36 
  just dump the informations...!?14:53.52 
kens Yes.14:54.16 
flemic trimbox, mediabox etc14:54.20 
kens I thnk pdf_info.ps does htis14:54.23 
flemic ok, thank! I will check that 14:55.55 
kens Seems it dumps MEdiaBox and CropBox, I have no doubt it can be extended to do the tohers.14:56.37 
  others14:56.47 
henrys besides aes what is the other blocker?14:57.40 
kens The crash that SaGS has.14:57.50 
  I'll find the number14:57.55 
  #69236714:58.09 
  I htink a customer also has found this but reported it differntly14:58.21 
  Yes, Jasper. He says he gets a crash when repeatedly zooming in GSView. I expect its the same problem.14:59.30 
  It certainly sounds the same.15:00.08 
henrys I wonder is the memory pointed to should be in local vm.15:00.27 
kens Ummm, pass.....15:00.38 
  I haven't actuallu examined hte problems15:00.47 
  I di dlook at Peter's text problem because it was reported against pdfwrite15:02.04 
  Then wished it off to chris15:02.26 
henrys I can't imagine the customer is using gsview in a commercial capacity.15:10.01 
kens No I guess not :-)15:10.12 
  I did see another repot with an intermittent crash which might be related.15:10.23 
  Was from Thomas, cc'ed to Jasper.15:11.32 
  Got bounced the first tiem and they retried the mail to support without the images attached.15:11.50 
  No, that's a red herring, sorry.15:12.10 
  But if I understand SaGS correctly, and use of GS which doesn't cloe the app and restart between jobs oculd have the same problem.15:12.48 
henrys I thought Robin_Watts cleared lib_ctx15:18.17 
kens No idea.15:18.48 
Robin_Watts ?15:19.30 
kens SaGS's bug report and proposed patch15:19.53 
  #692367 I think15:20.21 
tkamppeter Robin_Watts, hi15:43.50 
Robin_Watts tkamppeter: Ah, hi.15:43.57 
  I was hoping to talk to you.15:44.02 
  I committed changes that should make stars.pdf a bit more bearable.15:44.23 
tkamppeter Robin_Watts, how important is interpolation for printing, as printing is always a rendering with high resolutions, at least 300 dpi.15:44.42 
Robin_Watts The cluster reported differences in various test files on the cups device, but when I run them locally, I get identical results before and after my change.15:45.19 
tkamppeter Can I use always -dNOINTERPOLATE or do I get a visibly lower print quality then?15:45.40 
Robin_Watts If you look at the stars.pdf file, the 'shadings' you see behind each of the stars (the shadows) are given by a bitmap.15:46.02 
  If you render the file twice, once with and once without interpolation, and compare them, you'll see a noticable stepping in the shading.15:46.35 
  For stars.pdf there is a visible degradation in quality.15:47.01 
tkamppeter Robin_Watts, so I will leave interpolation in place.15:47.21 
henrys tkamppeter:can you tell us how does the quality compare to poppler?15:47.47 
Robin_Watts I *may* be able to find a bit more speed by tweaking the pattern bboxes used when tiling, but I don't want to promise that a) there is an improvement still be to found, or b) that I will find it in time for release.15:48.17 
  Anyone around with access to a Windows 7 box ?15:50.34 
henrys if ray_laptop isn't around I'll start up my vm.15:51.40 
tkamppeter henrys, I will try.15:51.52 
Robin_Watts http://ghostscript.com/~robin/0001-Fix-Bug-692355.patch15:52.09 
  That should, I hope, fix the windows title bar problems.15:52.21 
  I've tested it out on windows XP Pro 32bit here, and I can enter unicode chars, copy/paste them, and it all seems to do the right thing (and the titles on the windows are right, including when you close the window and it changes the title to include " - closing" momentarily)15:53.17 
henrys w6bullfrogs16:00.32 
  well there's a password I won't use again.16:00.54 
mvrhel2 oh looks like it is going to be a busy day16:10.17 
  I wonder what broke with respect to writing out to CIELAB color space. 16:10.42 
kens Time for me to go, goodnight everyone, I may be back later.16:13.36 
henrys Robin_Watts:works for me but wasn't broken before the patch.16:25.49 
Robin_Watts henrys: You have a windows 7 machine and it worked ?16:26.09 
  I mean it used to work ?16:26.24 
henrys correct16:26.26 
Robin_Watts then that's very strange.16:26.34 
henrys both work16:26.37 
Robin_Watts I guess I'll wait for ray.16:27.02 
henrys is ray_laptop the onyl 7 machine to produce this?16:27.04 
Robin_Watts I thought it was ray + at least one other person.16:27.22 
marcosw_ henrys: do we need a support meeting this week?16:35.48 
henrys if you can just scrub support and your list I think that will be fine but I'm available.16:36.56 
marcosw_ I'll go through my bugs and the support list later today. 16:37.43 
henrys marcosw_:I would like to ask folks to leave support to you and focus on projects and bugs - but I'm not sure how you would feel about that. Do you think there is too much for one person?16:38.17 
mvrhel2 ok I see what the issue is with 692364. This is not a new issue but has existed for some time16:38.48 
  the problem is that the graphic state is initialized with DeviceGray. Then we initialize the icc manager. And then we start drawing. Unfortunately, the color spaces in the graphic state are still DeviceGray and not ICC due to the fact that no other color spaces were set (and hence installed and converted over to ICC). So when we try to go to CIELAB as the target, gray maps to something weird16:40.19 
Robin_Watts mvrhel2: I've never understood why we don't initialise the icc manager before anything else.16:41.08 
  (I'm sure there is a good reason)16:41.15 
mvrhel2 If I had control of the interpreter I would16:41.18 
  It would be the first thing I would do after getting the icc_dir set up16:41.37 
Robin_Watts What causes the icc manager to be setup ?16:41.44 
  is it a call in the PS initialisation files or something ?16:41.58 
marcosw_ henrys: The support load varies widely over time but usually I have no trouble keeping up. I'd rather folks left support to me, unless I find myself falling behind and ask for help. 16:42.03 
mvrhel2 set user params usually triggers it16:42.11 
henrys marcosw_:okay thanks16:42.25 
Robin_Watts Is there a reason it has to be done then? Can't we just hardwire a call in earlier?16:42.36 
mvrhel2 since the values that it needs can be set as a user param16:42.41 
Robin_Watts Ah! ok.16:42.49 
mvrhel2 unfortunately, I am at the mercy of order in which things are done in the interpreter. What I should do is this.16:43.24 
Robin_Watts Is it possible to start the icc manager up without those user params ?16:43.38 
mvrhel2 When the ICC manager IS set up, I should inspect the graphic state and fix the existing color spaces in it if they are DeviceGray16:43.55 
Robin_Watts shuts up and waits for mvrhel2 to talk16:43.55 
  mvrhel2: That sounds a bit nasty to me.16:44.14 
mvrhel2 I think that makes the most sense16:44.14 
Robin_Watts Because it relies on you being able to check all the extant color spaces.16:44.44 
mvrhel2 it is relatively minor. Simply populating a member variable int the colorspace16:44.45 
  ?16:44.54 
Robin_Watts Could I suggest an alternative?16:45.00 
mvrhel2 why is it nasty? the initialization occurs once16:45.12 
  part of the initialization is to make sure that the pgs start up color spaces are also set properly16:45.33 
  the DeviceGray color spaces in there are basically place holders16:45.50 
Robin_Watts Right, but when the initialisation happens you have to look at the graphics state and check to see if the color spaces are correctly set up, and if not, retrospectively fix them.16:46.01 
  What if there are other color spaces around that you can't get to through the state?16:46.23 
  (like say it's a nested state)16:46.31 
  (That may not be possible now, but who knows what will happen in future).16:46.54 
mvrhel2 ok. I suppose. I don't see how else to do this though16:47.27 
Robin_Watts Would it be possible to do something where the 'default' states are REALLY just placeholders?16:47.31 
  s/states/colorspaces/16:47.44 
mvrhel2 I suppose we could initialize to something different16:48.08 
Robin_Watts and when we meet them during execution fetch the defaults from the icc_manager?16:48.15 
mvrhel2 and then if we end up in the remap of that thing, install the proper object16:48.19 
  but I don't see really much difference16:48.54 
Robin_Watts What you just described sounds nicer than what I had in mind actually.16:49.23 
  So you initialise with a colorspace that when remapped replaces itself with the current default ?16:49.53 
  The difference in subtle, but it's there.16:51.01 
mvrhel2 This of course would involve the breaking of a const but since when has that stopped me16:51.05 
Robin_Watts s/in/is/16:51.08 
  Currently we intialise with something that's wrong, and rely on being able to "fix it up" later.16:51.40 
  (and if we ever fail to fix something up, we'll have problems).16:51.52 
mvrhel2 well technically here is the deal. DeviceGray is supposed to be the colorspace if nothing else is explicitly set in the document16:52.20 
Robin_Watts With the new scheme, we initialise with something that says "fix me", and whenever we hit such a thing, we fix it automatically using the same mechanisms that have been in gs for donkeys years.16:52.33 
mvrhel2 the issue is that we have not gotten our DeviceGray color space in the graphic state completely set up as we want it since we dont have everything available when we intialize16:53.01 
  this is why I was thinking of just doing it when we do have everything16:53.11 
  which occurs when we initialize the icc manager16:53.25 
Robin_Watts Right, but do you see the difference I'm trying to highlight ?16:53.40 
mvrhel2 yes. I see the difference. 16:54.12 
  The nice way is a bit more work since I need to introduce a new color space but that is not the end of the world16:54.55 
  I do like it in that it is clean and clear16:55.30 
  I do feel like the clock is ticking on this though16:55.55 
Robin_Watts Yes. I'd hope it would be easier for future generations of ghostscript coders to understand :)16:56.13 
mvrhel2 anyone else have an opinion before I embark on this? henrys?16:56.51 
Robin_Watts I've made my case, so I'll leave it to you, as you have a much better idea of how much work it is.16:56.55 
  mvrhel2: If the 'quick and nasty' fix is indeed quick, then maybe do that first?16:57.13 
mvrhel2 that is what I am thinking16:57.21 
  especially since it is a customer issue and they may want a patch for 9.0316:57.36 
  giving them a patch that adds in a new color space may be tricky16:57.49 
henrys well the graphics state is gray because the nulldevice is installed right?16:58.24 
mvrhel2 well gray is fine. but when we start doing fill page etc, it should be ICC based16:58.54 
  so that we can, as in this case of going to CIELAB as an output device color, map the fill page to CIELAB = [100,0 0]16:59.21 
Robin_Watts mvrhel2: I committed the dirty bbox changes.16:59.46 
mvrhel2 great16:59.51 
  DeviceGray is the default color space if nothing else has been set in the document.17:00.16 
  this is true for all devices17:00.21 
  default source color space17:00.35 
Robin_Watts And I have (I think) a fix for the unicode windows stuff. So until tkamppeter finds that I broke a load of stuff, I have nothing on the critical path for the release.17:00.45 
  Let me know if there is anything you'd like to offload to me.17:00.55 
henrys or right I'm thinking of the color model sorry17:00.58 
mvrhel2 np. so what happens now, is that we map the DeviceGray of the fill page to RGB using the old procs17:01.24 
  so it goes to 255, 255 25517:01.33 
  which is a very saturated RED in CIELAB17:01.42 
henrys okay so you need the device icc profile set?17:02.24 
mvrhel2 well it is set in when we do the fill page17:02.41 
  and the manager is all set too17:02.48 
  the issue is that the graphic state has DeviceGray color spaces set 17:03.01 
  not ICC color spaces17:03.05 
  they are left over from the init of the graphic state17:03.27 
  For a quick fix, I was going to "fix" the color spaces in the graphic state, when the ICC manager is initialized17:03.53 
  this assumes that we don't do any fills and drawings before the manager is initialized.17:04.13 
  I hope that never happens (I am wondering if it can happen)17:04.25 
  this is where the interpreter start up code becomes involved17:04.54 
  as there would be nothing that I could do in the C code to fix that one17:05.25 
  let me try that fix real quick 17:06.35 
henrys I think the output icc profile has to be set in gs_init.ps under /setdevice - for the same reason the halftone has to be set there.17:09.07 
  I guess I should be testing this stuff in the other languages.17:10.35 
mvrhel2 As I mentioned once, I would like to make a stress test for testing all these ICC options17:13.06 
  it could be done on just a handful of files17:13.19 
  and run during the clusterpush17:13.32 
henrys mvrhel2:see the erasepage in /setdevice - I think that is where the problem is - probably later it would be okay.17:14.01 
mvrhel2 well it wont be ok if the DeviceGray colorspace in the pgs stays the same17:14.46 
  if we never encountered another color space in the document, then those color spaces would remain17:15.10 
  and we would use the old procs, and basically ignore the ICC settings17:15.25 
henrys from what you are saying DeviceGray should never be put in in the first place if you want ICC to be the default.17:17.15 
tkamppeter Robin_Watts, I have built the current GS now and for me stars.pdf needs 15 minutes in standard mode.17:19.25 
Robin_Watts tkamppeter: And gives the right result at the end?17:19.48 
  That sounds like progress to me.17:19.57 
  How does that compare to poppler etc ?17:20.09 
  henrys: Right. That was my argument. Michael suggested a scheme whereby we install a "placeholder" colorspace (call it "DefaultGray" or something?). When that actually gets used, its remap function will be called, and it can replace itself with the current default from the icc manager.17:21.57 
tkamppeter Robin_Watts, 1:15 min with -dNOINTERPOLATE17:22.00 
mvrhel2 henrys: well yes. the problem is that when we initialize the graphic state, we have to put something in for the color spaces so we start out with DeviceGray, since that is the Default if nothing else is set in the document17:22.24 
  we can't put in an ICC color space yet, as we dont even have an ICC manager at this time17:22.44 
  that comes after, and then we still need to set it up, which relies upon the user parameters since it is configurable from the user params17:23.13 
henrys what does the manager depend on for initialization?17:24.00 
mvrhel2 it needs the default icc profiles to use for the source colors17:24.17 
henrys well it must need the memory manager17:25.14 
mvrhel2 someone on the command line could set a -sDefaultGrayProfile="my_gray_source_profile.icc"17:25.31 
henrys set it up in gs_lib_init117:25.33 
mvrhel2 it does need the memory manager17:25.44 
tkamppeter Robin_Watts, the output is correct (6 stars) and quality-wise (seeing the files highly magnified "pixel-peeping") I do not see a difference, which would make me tend more to introduce -dNOINTERPOLATE by default.17:26.26 
mvrhel2 but it is a member variable of the graphic state17:26.35 
  it has to be set up though through the user parameter settings17:26.58 
Robin_Watts tkampeter: seeing the output files 1:1 at 720dpi, I could clearly see the difference.17:27.14 
mvrhel2 that I hope is done before we actually do any fill pages etc17:27.20 
Robin_Watts tkamppeter: But this isn't a typical file. I could probably find you much smaller simpler files that would look much worse without interpolation.17:28.26 
henrys mvrhel2:well then gs_initgraphics right?17:29.03 
  oh there is an NB above about that.17:29.46 
mvrhel2 so normally in gs_setcolorspace_only we install the color space17:31.03 
  if there is a change in the color space id17:31.11 
  this is when we do the conversion of the DeviceRGB, DeviceGray DeviceCMYK to an ICC color space17:31.41 
  the thing that occurs is that we never really "install" the DeviceGray color space that we stuck in the graphic state17:32.17 
  of course we can't really do that until the icc manager is set up17:32.35 
  and that can't occur until we have the user parameters all set17:32.43 
henrys tell me again why you can't install icc gray on gsstate.c:256?17:33.10 
mvrhel2 ok. so in gs_imager_state_initialize the icc manager is allocate17:33.48 
  line 23417:33.56 
  but we dont have the user parameters yet to fill it in17:34.10 
tkamppeter Robin_Watts, can you give me a good example file to test this?17:34.19 
mvrhel2 all the member variables are NULL17:34.27 
Robin_Watts tkamppeter: I'll try and create one.17:34.39 
mvrhel2 so we can't assign an ICC profile to the Gray color space at line 25617:34.52 
  so, in this sense, DeviceGray is a placeholder17:35.17 
  I could actually take care of this problem when we go to the remap proc for DeviceGray17:35.42 
  Robin_Watts; that would not really be any different than adding a new color space17:36.06 
Robin_Watts mvrhel2: That sounds nice.17:36.21 
henrys mvrhel2:are we getting the erasepage before user settings?17:36.39 
  that in itself doesn't make sense for other reasons.17:36.56 
mvrhel2 henrys: that is a concern. 17:37.22 
henrys I wonder if all this goes away if we just put off that erasepage in setdevice until after user settings.17:38.01 
mvrhel2 well we still need to fix the DeviceGray color spaces in the graphic state17:38.18 
  those will still be wrong17:38.23 
  let me fix that in the remap of the DeviceGray17:38.41 
henrys I thought the user settiings would fix that?17:38.44 
mvrhel2 no. we still will have the issue that the graphic state was initialized by DeviceGray17:39.07 
  and we can't set it to ICC at that time17:39.14 
henrys the code just execute the default setting implicitly.17:39.15 
mvrhel2 yes.17:39.20 
  unless it encounters a setrgb17:39.29 
  or something17:39.31 
henrys we need alexcher here.17:39.57 
mvrhel2 there is the open question of are there any erase pages etc that occur before we set the user parameters17:40.27 
  in this particular file from the customer, the user parameters have all been set before the fillpage occurs17:42.33 
tkamppeter henrys, Robin_Watts, the output quality of Poppler and Ghostscript are more or less the same. Poppler is somewhat faster on stars.pdf than GS with -dNOINTERPOLATE. Poppler: 48 sec, GS: 1:15 min.17:42.42 
henrys so you are saying in the languages we should never see Device* color space in the graphics state when rendering is that correct?17:42.49 
mvrhel2 so it is possible for me to fix the problem in gs_remap_DeviceGray17:42.50 
  that is correct17:42.56 
Robin_Watts If the ICC manager isn't setup when you try to remap, then don't swap out and live with the gray ?17:43.10 
mvrhel2 can't do that17:43.22 
  if the destination is CIELAB 17:43.33 
  you get the nice red background as output17:43.44 
Robin_Watts That way if we never initialise the icc manager we get the old standard way of working? oh.17:43.45 
mvrhel2 if I was going to an RGB output color we would be OK with that approach17:44.12 
  let me fix this in gs_remap_DeviceGray17:44.57 
  gx_remap_DeviceGray I mean17:45.11 
alexcher here17:45.14 
mvrhel2 alexcher, is it possible that an erasepage can occur in the intitialization before the user parameters are set17:45.43 
  I am hoping the answer is no17:46.03 
alexcher mvrhel2: Yes, it occures quite early in the initialization process.17:46.26 
mvrhel2 how can you do a drawing operation before you know what color space you are drawing?17:46.55 
Robin_Watts tkamppeter: http://ghostscript.com/~robin/blob3.pdf17:47.05 
mvrhel2 There must be another one that occurs after the user parameters are set then17:47.12 
Robin_Watts render that with and without interpolation, and you'll see why interpolation is important.17:47.50 
mvrhel2 Robin_Watts: your link does not work17:48.13 
Robin_Watts ahem. Try again now.17:48.28 
  sorry.17:48.30 
alexcher mvrhel2: gs creates a device, sets device parameters, makes it a current device, and does erasepage.17:48.31 
  mvrhel2: user and system parameter change may happen at a wrong moment.17:49.47 
mvrhel2 and at that time the user parameters are not necessarily set?17:49.56 
alexcher mvrhel2: User and system parameters are Level 2 features but device is initialized on level 1.17:51.02 
tkamppeter Robin_Watts, get yourself a new camera 16 pixel is a little bit old-fashioned. Mine has 18MPixel.17:51.43 
henrys mvrhel2:why can't you simply simulate the ops second command line parameter in the code so it happens whenever ghostscript is run? At least for now.17:52.21 
alexcher mvrhel2: We, probably, have a dotted version. In need to check the code.17:52.21 
tkamppeter Robin_Watts, I think a 16-pixel-picture is a little unrealistic. Its output is too much dependent on the PDF viewer/print filter.17:52.43 
Robin_Watts Random thought, and please feel free to tell me to go away, but... wouldn't all this be sidestepped by NOT passing the icc profile dir in as a userparameter, but having it handled as a special case by the executable ?17:53.41 
  tkamppeter: Yes, it's an extreme example.17:54.07 
mvrhel2 henrys and alexcher: I don't quite understand what you last said17:54.09 
henrys well the bug reporter set the default profile and it works correct?17:54.46 
tkamppeter Robin_Watts, if you have a PDF file generated by an application and meant to be a print job, its bitmap parts are expected to have enough resolution, and if not it is probably intended to see the pixels.17:54.51 
Robin_Watts tkamppeter: not true.up 17:55.02 
mvrhel2 henrys: oh yes17:55.04 
  I wonder why that worked. 17:55.17 
alexcher mvrhel2: gs may have custom operators to achive the same effect on level 1.17:55.20 
mvrhel2 oh alexcher. now I understand17:55.35 
tkamppeter Robin_Watts, do you have an example?17:55.39 
henrys so I'd assume you could put that somewhere in our the vast abyss and get the same effect.17:55.45 
Robin_Watts tkamppeter: I can't believe you can't see the difference between stars.pdf with and without interpolation at 720dpi.17:55.55 
henrys preferrably in the C language.17:56.13 
mvrhel2 alexcher: so you have some PS code in gs_lev2.ps17:56.16 
  that sets up the icc manager17:56.22 
Robin_Watts Render to png16 and then view the resultant images at 1:1.17:56.25 
mvrhel2 is it not possible to have the occur earlier?17:56.33 
  s/the/that17:56.39 
henrys alexcher:when can we stop supporting level 1 I thought we agreed we could stop.17:56.49 
  ?17:56.51 
mvrhel2 alexcher line 63317:57.34 
  to line 72417:57.59 
alexcher henrys: yes, we should, but this requires significant changes in many places.17:58.20 
mvrhel2 or do an erase page following this17:58.44 
alexcher mvrhel2: Thank you, I'll check this.17:58.52 
tkamppeter Robin_Watts, I stepped through several magnifications and they looked equal for me all the time.17:59.51 
  Robin_Watts, how did you the test? What output format did you create and how did you view the result?18:00.20 
mvrhel2 alexcher: my line numbers might be off a bit as I have a few things in my version for support of Gray to K18:00.20 
Robin_Watts tkampeter: I ran: gs -sDEVICE=png16m -o out.png -r720 stars.pdf18:00.53 
  and then the same with -dNOINTERPOLATE18:01.01 
  Then I loaded the pngs into windows bitmap viewer and sent them to 1:1.18:01.33 
tkamppeter Trying this now ...18:02.23 
  Can I use -dNOINTERPOLATE with all input formats and all output devices? Or does it belong to special formats?18:07.47 
Robin_Watts It's to do with the internal graphics library, not the output device.18:08.09 
tkamppeter So I can use it always?18:08.26 
Robin_Watts yes.18:08.32 
tkamppeter Robin_Watts, on the PNG tests the computing times are of the same order and the generated images look also of equal quality for me.18:19.58 
Robin_Watts tkamppeter: what exact command lines did you use please?18:20.31 
tkamppeter Robin_Watts, the PNG files are even exactly identical (diff).18:20.59 
  Command lines:18:21.10 
Robin_Watts Right. Did you put -dNOINTERPOLATE before the filename ?18:21.20 
mvrhel2 oh there are 2 fill pages that occur in this file18:21.30 
tkamppeter time gs -sDEVICE=png16m -o stars-gs-standard.png -r720 stars.pdf18:21.33 
mvrhel2 oh 318:21.38 
  actually 1018:21.53 
tkamppeter real15m15.741s user15m8.510s sys0m2.690s18:21.57 
  time gs -sDEVICE=png16m -dNOINTERPOLATE -o stars-gs-nointerpolate.png -r720 stars.pdf18:22.10 
  real1m23.835s user1m19.260s sys0m1.430s18:22.34 
mvrhel2 henrys and Robin_Watts: so a few lines in gx_remap_DeviceGray fix the issue18:22.45 
Robin_Watts mvrhel2: Nice!18:23.01 
henrys great what if users want to use the device color spaces?18:23.25 
tkamppeter The times show that the second actually ran without interpolation, and only the second.18:23.38 
mvrhel2 henrys: that is handled with the SMALL_PROFILES option18:24.03 
  as those will do the simple mappings18:24.17 
henrys 10/418:24.25 
Robin_Watts tkamppeter: So -dNOINTERPOLATE clearly had an effect, but didn't change the output.18:24.31 
  I'm now confused, because I'm seeing the same files produced here, and I can't say why.18:24.50 
tkamppeter Robin_Watts, yes this is also how I see it.18:24.50 
Robin_Watts tkamppeter: OK, I'm going to have to eat humble pie here.18:25.40 
tkamppeter Robin_Watts, also the two CUPS raster files are identical, coming from command lines which differ only by -dNOINTERPOLATE.18:26.02 
Robin_Watts I've just looked at the two produced files, and the artifacts I could see without interpolation enabled are in fact visible in the non-interpolated ones too.18:26.18 
  s/non-/18:26.28 
mvrhel2 ok. so now, the question is do I make this part of the whole gray to K commit18:26.29 
Robin_Watts so I apologise for that red herring.18:26.42 
chrisl Robin_Watts: what's the memory use like now with stars.pdf?18:26.51 
mvrhel2 henrys: if I just commit this fix, it is going to end up mapping some gray spaces to composite CMY which is what we dont want18:27.01 
Robin_Watts Now I want to understand why we are using interpolation anyway...18:27.11 
mvrhel2 so, I am going to make this part of the whole fix 18:27.23 
tkamppeter Robin_Watts, so looks like that I will insert -dNOINTERPOLATE in the default GS options in gstoraster, at first at least in the Ubuntu package, and if no one complains later upstream.18:27.34 
Robin_Watts mvrhel2: Can you not do that fix first ?18:27.41 
  tkamppeter: Sure. If I turn up anything to change that decision,I'll let you know.18:28.14 
mvrhel2 if is going to generate thousands of regressions which the other commit is going to undo18:28.21 
tkamppeter Robin_Watts, seems that now we get all complaints of Linux users about too slow printing fixed with a little patch ...18:28.28 
mvrhel2 If that seems confusing I can explain more18:28.54 
Robin_Watts mvrhel2: no, I mean you have 2 commits, right, A and B.18:29.06 
mvrhel2 yes18:29.16 
Robin_Watts With git you can commit them as A and B, and then rebase -i to change the order of them.18:29.23 
mvrhel2 well I have not commited anything yet18:29.29 
  they are both uncommitted works18:29.37 
  in two different checkouts18:29.43 
Robin_Watts would that not let you keep them separate and yet cause no revertable diffs ?18:29.45 
  ah, ok. so either order would cause differences that the other would revert.18:30.12 
mvrhel2 well no. 18:30.34 
  actually, If I did the gray to K first18:30.43 
  then the fix that I have here18:30.48 
  that should not have double regressions18:30.56 
  hmmm18:31.29 
  just trying to reduce the number of files to look at18:31.39 
  and these fixes are somewhat related18:31.49 
  I had run into this issue in the gray to K but had not figured out quite what to do about it until now18:32.13 
  I guess a sep. commit is better for a patch though18:32.36 
  for the customer18:32.39 
  I will do that18:32.41 
  so they can fix 9,03 easily18:32.54 
chrisl mvrhel2: I don't know how the graphics library works, in this context, but I assume that all the languages have implicit or explicit "core" color spaces which must be supported?18:44.27 
mvrhel2 well Gray, RGB, CMYK18:44.45 
  PS has its pile of CIE ones18:44.55 
  PDF has its oddball Cal ones plus ICC18:45.09 
  then there are Sep and DeviceN18:45.16 
  of course XPS has ICC up to 8 colors18:45.26 
  everything is ICC in XPS18:45.34 
  that is about it I think18:45.47 
chrisl Okay, so is there an argument for including profiles for those "core" spaces (gray, rgb and cmyk) in the executable as static data? So they are always ready for use?18:45.55 
mvrhel2 chrisl: yes ray_laptop is going to bake in the ps_cmyk, ps_gray, ps_rgb icc profiles for us18:46.24 
  so that we always have these ones that do the simple conversions18:46.37 
chrisl Oh, okay, I misunderstood what Ray was doing then - ignore me :-)18:47.03 
mvrhel2 no problem. I think it is a great idea. Two people thought of it...18:47.27 
chrisl It makes sense for the colour spaces we'll *always* need to be there implicitly, I think18:48.01 
mvrhel2 chrisl: so on the CIELAB TIFF stuff, when the user specifies a CIELAB output profile there is a special setting in the TIFF output that indicates the data is CIELAB ICC data18:48.26 
  and photoshop then interprets it correctly18:48.34 
henrys in pcl's first erase page *pgs->color->color_space->type == gs_color_space_index_DeviceGray18:49.08 
mvrhel2 I had fixed that for them quite some time ago. but this file had a special issue with respect to things getting initialized for the ICC work flow18:49.13 
chrisl mvrhel2: yep, as I just replied, my suggestion about trying our profile was just as a diagnostic thing, not that I thought it was actually the problem.18:49.42 
mvrhel2 henrys: ok. well, if you try to set the -sOutputICCProfile to lab.icc then you should see a nice red background when you open it with photoshop18:50.37 
  (and go out to tiff24nc)18:50.44 
  henrys: is the pgs->icc_manager->defaultgray set?18:51.24 
Robin_Watts walks dogs.18:51.30 
mvrhel2 I need to go get the kids from track camp.18:51.40 
  brb18:51.42 
Robin_Watts mvrhel2 is training Auden to be a bounty hunter.18:52.12 
henrys yes the defaultgray is set.18:54.33 
tkamppeter Robin_Watts, another Ubuntu package, with patched CUPS filters to use -dNOINTERPOLATE, is on its way ...18:59.29 
Robin_Watts No ray today ?19:41.10 
kens Haven't seen him today19:41.27 
tkamppeter Robin_Watts, with my results the decision is made, Ghostscript will return to be the ...toraster filter to generate the CUPS Raster format in Ubuntu Oneiric. Thank you for your great work!19:48.18 
Robin_Watts tkamppeter: Excellent stuff.19:48.35 
  Though I should say that the work wasn't all mine; I merely finished off stuff that both mvrhel2 and ray did before me.19:49.02 
tkamppeter so also thanks to mvrhel2 and ray_laptop.19:50.57 
Robin_Watts mvrhel2: I've put a patch on bug 692235. That bug belongs to you officially, but I've done some digging on it before.19:54.02 
  I'm hoping that Russell and Martin (the two commenters on the bug) will test it and if they are both happy, we can commit and close it.19:54.34 
kens Night alll.19:58.58 
Robin_Watts night.19:59.05 
  Aha, ray.20:00.59 
  Any chance you could try out my patch from bug 692355 to see if it solves it for you please?20:02.11 
chrisl Robin_Watts: is it just a Windows 7 setup that we know showed the problem that you need? I can do it, if you like?20:03.15 
Robin_Watts chrisl: Windows 7 works for henrys.20:03.36 
  but it fails for ray_laptop.20:03.50 
  and for someone else I thought.20:03.55 
chrisl Yeh, me!]20:04.02 
  I pointed it out to Ray]20:04.10 
Robin_Watts basically, I'd like someone for whom it failed before to tell me it now works.20:04.14 
  Ah, then you'd be perfect! :)20:04.22 
chrisl I also can't seem to hit the return key reliably20:04.27 
Robin_Watts dinner. bbiab.20:05.43 
mvrhel2 ok. back from indian buffet for lunch...20:19.57 
chrisl henrys: is your Win7 VM a 32 bit setup?20:27.50 
Robin_Watts henrys: Just to be clear here, were you testing gswin32c.exe or gswin32.exe ?20:37.24 
  because the "Ghostscript Image" window works in both.20:37.43 
  it's the text window in the gswin32.exe that doesn't work.20:38.01 
chrisl Robin_Watts: the Window title shows perfectly with your patch, both 32 and 64 bit exes20:38.18 
Robin_Watts Fab.20:38.25 
chrisl Both Ray and I were testing on 64 bit, so I wondered if the source of the problem was the 32 bit to 64 bit ABI - we'll probably never know......20:39.06 
Robin_Watts now, how can I push just one of my git commits...20:39.52 
  bingo.20:41.47 
  mvrhel2: I burbled above about bug 69223520:42.31 
  chrisl: Thanks for testing.20:42.57 
mvrhel2 hi Robin_Watts: ok thanks. I will try to take a look at this this afternoon20:43.11 
chrisl Robin_Watts: np20:43.19 
Robin_Watts mvrhel2: It's a low priority bug, so don't feel you have to spend time on it. If you don't have any objections to the approach, feel free to assign it to me.20:43.58 
mvrhel2 ok20:44.13 
  brb20:44.16 
chrisl I've had enough - good night all!21:10.14 
henrys sorry I was away - windows 7 32 - I ran gswin32.exe21:17.42 
  I did run it from a cygwin prompt.21:18.15 
  running it from an explorer window gets the same result.21:20.40 
  language location - english and US FWIW21:23.06 
ray_laptop henrys: saw your comment -- you said you ran on Win7 32 but didn't say whether or not it worked. I commented on the bug that Win7 64 works for me. I didn't mention that BOTH gswin32.exe and gswin64.exe are OK.21:51.25 
henrys I said it worked earlier and Robin_Watts's question accounted for that context.21:52.42 
ray_laptop henrys: thx21:53.01 
henrys very odd21:53.12 
  ray_laptop:is there anything other than region and language that might be at play here?21:53.51 
  anything other conf setting that should affect the title.21:54.21 
ray_laptop henrys: not that 21:57.05 
  henrys: not that I know of. So I think it is OK now21:57.21 
henrys mvrhel2:so defaulting the language to the simple profiles is going to change all the files too - do you want to pool that in with the other changes?22:18.06 
mvrhel2 henrys: good question. 23:27.21 
  xps should default to what we currently have23:27.37 
  I think pcl is the odd man out here23:27.47 
  henrys: I think we should just wait until we get the SMALL_PROFILES build option put in place23:28.25 
  but first I need to figure out why my very simple fix seg faults everything during the clusterpush23:39.01 
  :(23:39.04 
tkamppeter ray_laptop, I also want to thank you for your parts of the work on the performance problem bug 691755. This fix was really important.23:39.53 
 Forward 1 day (to 2011/07/22)>>> 
ghostscript.com
Search: