IRC Logs

Log of #ghostscript at

 <<<Back 1 day (to 2013/07/11)2013/07/12 
mvrhel_laptop ah. I see windows phone has everything locked down except the "storage folder" and skydrive or files from other apps06:14.01 
kens2 Robin_Watts : ping08:43.32 
Robin_Watts pong08:43.39 
kens2 Figured out why vogue didn't work08:43.48 
  typo on my part, b then a not a then b08:44.00 
Robin_Watts ah, ok.08:44.11 
kens2 If you do it the other way, it opens a debugger in Chrome08:44.14 
Robin_Watts not for me, but OK :)08:45.05 
kens2 Only in the Vogue site08:45.19 
Robin_Watts yeah, that's where I was trying it.08:45.36 
kens2 Also, not after you've run the original Konami code08:45.48 
  If I repeat it, I get little green boxes on each image with their dimensions08:46.11 
Robin_Watts I tried it with a fresh tab.08:46.11 
kens2 Hmm, don't know then08:46.22 
Robin_Watts Ah! I got it this time.08:46.36 
kens2 :-)08:46.39 
  Oh good grief. 'we want to set the resolution really low, get crappy output and a file 10 times bigger that prints really slow, because its what we had before and we think the toner cost will be lower'09:02.26 
tor8 Robin_Watts: in mupdf if you toggle between grayscale and color rendering, we get an assert-abort on pdfref17 page 115009:24.29 
  I'm guessing because we cache color-converted pixmaps...09:24.39 
  and then the color space doesn't match09:24.46 
Robin_Watts tor8: ah.09:43.36 
  Try: mupdf-curl.exe 
tor8 Robin_Watts: rats, I don't have a windows machine up and running now!09:50.21 
  maybe I should install one in a virtualbox on my new linux workstation09:50.34 
Robin_Watts tor8: I need to make some changes to the linux app for it to work there. will do that now.09:57.13 
  tor8: Thought I'd found a way to do it using client messages, but it's dying with a BadValue.11:13.00 
  sebras: Do you speak X ?11:13.09 
tor8 Robin_Watts: Xlib isn't thread safe IIRC11:28.01 
Robin_Watts tor8: yeah, I have it working now.11:28.15 
tor8 Robin_Watts: you can send clientevents to your own window11:28.18 
  but I'd suggest adding a bidirectional pipe or something of the sort that you can select() on in the main loop11:28.52 
  how did you solve it?11:29.00 
Robin_Watts tor8: client events11:29.08 <- linux exe11:29.32 
  actually... it's hung at the end of uploading...11:29.55 
  ah, now it's there.11:30.11 
tor8 argh! the latest skype for osx has fixed the time stamps, but removed the option to have conversations in separate windows...11:43.00 
Robin_Watts tor8: Does the exe work for you?11:44.21 
  Or do I need to do a mac build ? :)11:44.42 
tor8 yup. it works.11:44.47 
  thanks to multi-arch11:44.54 
  it's funny to see the page appear and then the fonts get fixed latr11:45.04 
Robin_Watts So, do you see the pages 'fill in' ?11:45.07 
tor8 what does © and (E) mean in the error read of a block we don't have?11:45.57 
Robin_Watts At the moment, the exact areas of the file that get fetched is driven by the page trying to get data and failing.11:46.00 
tor8 page 1150 is taking ages to load11:46.04 
Robin_Watts tor8: I have various places in the curl fetch routine that can throw the same error.11:46.31 
  so I distinguish them by putting (A), (B), (C) etc in the message.11:46.42 
tor8 ah, right11:46.51 
Robin_Watts There are various things I'd like to tweak.11:47.38 
tor8 how big blocks are you requesting at a time?11:48.40 
Robin_Watts Firstly, at the moment, when I load the hints table, I run through and populate the xref with offsets to where the objects are supposed to be.11:48.56 
tor8 I'm still waiting for page 1150...11:48.59 
Robin_Watts 4K at a time.11:49.05 
  I'll look into page 1150.11:49.17 
tor8 that's going to be slow... :)11:49.17 
  I think we should request 128K at a time at least11:49.27 
Robin_Watts The size of the block is a #define.11:49.30 
  if the hints table is screwed, we can get duff offsets, I think.11:49.58 
tor8 ah, there it arrived. but it's the wrong page.11:50.04 
Robin_Watts so I need to rework that a bit.11:50.05 
tor8 1150g and I end up on page 125611:50.56 
  warning: duplicate object found11:51.12 
  and cannot find object in xref11:51.26 
Robin_Watts duplicate object found is not unexpected.11:51.55 
  cannot find object in xref; that can happen if there are links on the page.11:52.11 
  links are not included in the shared objects, so we don't resolve them until we have completely loaded all pages.11:52.27 
  Did I mention that linearized files are REALLY horrible?11:52.39 
tor8 I have heard some grumbling on that topic, yes :)11:52.54 
Robin_Watts tor8: yeah, I'm not getting the page I expect either.12:11.35 
  will look into that after lunch.12:15.24 
  tor8: If you let the data load, I think it works fine.12:16.53 
  i.e. open the first page, wait for the data to load in the background, then do 1150g12:17.08 
  so something is going wrong if it has to retry or something? odd.12:17.26 
tor8 second time I tried, I ended up with page 1200 instead12:21.59 
  maybe the page tree loading is *too* error resilient?12:22.53 
chrisl Hello ghostbot!13:54.13 
Robin_Watts Did you start that, or did I ?13:54.27 
chrisl Does it matter? I did run the script13:54.51 
Robin_Watts OK. I tried, but I did so as root, so it told me off.13:55.08 
chrisl Yes, running it as root caused all the missing logs occurrences a year or so back13:55.46 
Robin_Watts ok, so as long as it was you that made it work not me, we're fine.13:56.08 
chrisl As you saw, the script will now bale if you try to run it as anything other than the ghostbot user13:56.40 
Robin_Watts tor8: So, currently I'm not finding page objects from the hint stream. I find them from linearly walking the file still.14:10.03 
  I should fix that.14:10.06 
  BUT... I think before I do I should understand why it's missing some occasionally (at least I think it is, that would explain the 1150g thing)14:10.48 
tor8 Robin_Watts: what happens if they end up out-of-order?14:19.23 
Robin_Watts linearized docs are guaranteed in-order.14:19.38 
tor8 or the curl fetch requests come in out-of-order, could that happen?14:19.41 
Robin_Watts the curl fetch is all hidden below the stream level.14:19.58 
tor8 Robin_Watts: I have three commits on tor/master14:20.18 
Robin_Watts The linear walking of the file to read objects happens at the start of every pdf_load_page call currently.14:20.43 
  so it never knows about the out of orderness.14:20.55 
  tor8: Will look.14:21.00 
  I see 4 ?14:23.20 
tor8 sorry, yes, 414:24.11 
  the first one makes it easy to add gtk and opengl based viewers and build all of them at once14:24.51 
Robin_Watts Doesn't the makefile change mean that unless we specifically set NOX11 to "", that we'll get no x build?14:24.51 
  i.e. if NOX11 is undefined, NOX11 is set to "yes"14:25.31 
tor8 are you talking about the NOX11 ?= yes inside the ifeq MINGW section?14:25.31 
Robin_Watts Ah! Lack of context.14:25.53 
  right, sorry, that one's fine then.14:26.00 
tor8 question with that one is whether to keep "mupdf-x11-v8" naming or go with "muview-x11-v8"14:26.53 
Robin_Watts I brought up the idea of renaming it to muview a while ago at a staff meeting and was shouted down.14:27.46 
tor8 you had support from me, IIRC!14:28.09 
  but oh well, let's wait a while longer then14:28.15 
Robin_Watts OK. All 4 look fine to me.14:28.19 
tor8 I hope the top one solves zenikos issues with the link destinations14:28.45 
Robin_Watts Does the last one tie in with zenikos comments yesterday?14:28.49 
  hah. ok.14:28.57 
tor8 yes. I'm making a stricter test though (check that it's both a dictionary and that the /Type is /Page)14:29.19 
  zenikos patch jsut tests is_dict14:29.29 
Robin_Watts tor8: Doesn't this mean that any page with a null /Parent pointer will think it's page 1 ?14:29.59 
tor8 yes. we could test both /Type and /Parent14:30.46 
Robin_Watts tor8: Belt and braces; if the run up to /Parent fails, we could do a slow search of the pagetree.14:31.48 
  and possibly fix parents as we go?14:32.04 
  that might be useful as part of our generic 'repair PDFs' anyway?14:32.44 
tor8 hmm. pdf_insert_page doesn't fix up the /Parent entry of the page object14:32.59 
Robin_Watts I suspect I had heat stroke while I was writing that.14:40.47 
tor8 well, and I semi-fixed it. I shuold do a proper fixup of the whole page creation implementation.14:41.37 
  and then make pdfclean use it.14:41.42 
  but that's for another week14:41.55 
  Robin_Watts: one more commit on tor/master then14:43.21 
Robin_Watts ok. looks fine.14:45.11 
tor8 oops. fixed a refcounting error in that one.14:45.14 
  should be array_insert not array_insert_drop...14:45.24 
Robin_Watts right.14:45.32 
  looks good to me.14:46.10 
tor8 alright, pushed.14:46.14 
henrys gawd I can't believe I'm in the middle of Nebraska with a 2g, yes that's a 2, wireless hotspot. 15:32.09 
kens lol :-)15:32.35 
henrys I have no idea how people live here unless you farm corn.15:33.47 
  you can drive 100 miles and see absolutely nothing but corn and dirt15:34.35 
  very nice speedup kens!15:36.10 
vtorri chrisl: ping15:40.58 
chrisl vtorri: pong15:41.11 
kens thanks henrys15:41.14 
vtorri chrisl: i still have errors with libtiff when compiling gs15:41.32 
chrisl vtorri: well, as I said, I can't reproduce the issue.....15:41.54 
vtorri the LIBTIFF_INT8_T and al types are not defined in the config headers15:42.09 
chrisl <shrug> works for me....15:42.30 
vtorri do you still plan to use the sub-configure of libtiff ? or do you plan to remove it ?15:43.04 
chrisl We do still use the libtiff configure15:43.21 
vtorri and in a near future ?15:43.43 
chrisl No plans to to change, no15:43.50 
vtorri ok15:43.55 
  chrisl: do you plan to allow the use external libraries like libtiff (jpeg, png, etc..) ?15:45.22 
chrisl You already can15:45.36 
vtorri how ?15:48.20 
  i have ghostpdl from the git repo15:48.34 
chrisl Delete the directories, and rerun configure15:48.43 
vtorri and ./gs/configure --help show nothing15:48.47 
  chrisl: btw, there is a more recent version of libtiff than in the git repo, in case you don't know15:50.25 
chrisl You told me before, I knew then, and I still know now15:50.45 
vtorri ok15:50.52 
  sorry to bother you15:50.57 
  gp_win32.c is never compiled....16:08.37 
  the build system is an abomination16:08.54 
chrisl Oh well.....16:09.12 
vtorri i have an undef ref, but i think that you still don't care16:09.33 
chrisl You should be getting gp_unix.c compiled16:10.06 
vtorri ./obj/iapi.o:iapi.c:(.text+0x23a): undefined reference to `gp_local_arg_encoding_get_codepoint'16:11.35 
  function which is defined in gp_win32.c16:11.44 
chrisl vtorri: I told you about that the last time you asked about this stuff16:12.08 
vtorri ???16:12.15 
  i don't think so as it's the first time i go so far in the compilation...16:12.57 
chrisl I found the problem, and have a fix, but I never pushed it because you were still having your problem16:13.19 
  Oh actually, I did push the fix:16:13.44 
vtorri yes, but i never asked you about that undef ref16:13.45 
vtorri hmm, maybe the autotools were not automatically relaunched when is modified16:15.03 
chrisl You need to return autogen.sh16:15.21 
vtorri usually, when is touched, the needed autotools (autoconf, etc...) are relaunched16:15.50 
  it seems it's not the case16:15.56 
chrisl How would that happen, then?16:16.13 
vtorri let's try16:16.21 
  compilation succeeded, but no DLL built16:20.54 
chrisl Well, what did you get?16:21.24 
vtorri only gs.exe in gs/bin16:22.21 
  and a 'find' finds no *.dll16:22.39 
chrisl Well, no, the normal build is a monolithic executable16:24.00 
vtorri ok16:24.13 
  there is a procedure to get the dll ?16:24.28 
chrisl Use the Windows project16:24.36 
vtorri hmmm16:24.51 
chrisl Unix has shared libraries16:24.57 
  vtorri: does mingw not support shared libraries, then?16:29.26 
vtorri it does16:29.42 
chrisl As in Unix style .so shared libs16:29.57 
vtorri shared lib on windows are DLL16:30.08 
  i've built dll with mingw for 7 years16:30.36 
Robin_Watts vtorri: Well, presumably you need a different set of build flags to tell it to build a DLL than you would to build an .so16:32.39 
vtorri $(CC) -shared -o $@ -Wl,--out-implib=$(IMPLIB) -Wl,--export-all-symbols -Wl,--enable-auto-import $(LIB_O)16:32.47 
Robin_Watts which would mean you'd expect to need to make changes to the makefiles to support building a dll.16:33.25 
vtorri that's a make rule that can be used16:33.26 
Robin_Watts because, as we've said before, msys is NOT a supported platform.16:33.43 
vtorri sure16:33.51 
  but which Makefile ?16:33.56 
chrisl vtorri: you can probably edit the top level Makefile, that has the main compiler options in it16:35.21 
Robin_Watts chrisl is the expert here, but AIUI... Generally the structure of the build system is that you have a top level makefile that sets up variables etc, and that then includes the platform independent ones.16:35.36 
  For instance, msvc.mak is the top level makefile for building on Windows using nmake and a VC toolset.16:36.11 
vtorri there are plenty of .mak files that are included16:36.46 
Robin_Watts yes.16:37.17 
  The basic rule is that the top level makefile can be as specific as you like, but the lower level ones need to work on all the variants of make. hence no if...else... blocks because the paradigms for that vary across different make versions.16:38.24 
  So if I was you, I'd copy the autogenerated Makefile, and work from that.16:38.56 
chrisl vtorri: if you look in gs/Makefile and search for CFLAGS_SO 16:39.04 
Robin_Watts Once you have that working to your satisfaction, then maybe we can consider taking changes from that back into the autogenerated one.16:39.36 
vtorri chrisl: that variable is empty16:40.40 
  so i guess it's .so by default16:40.52 
chrisl You may also want to set DYNANIC_LIB_EXT to something other than .so16:42.11 
vtorri hmm, there is a 'so' rule16:44.14 
  in unix-dll.mak16:45.02 
chrisl Unix systems to use .so for shared libs16:45.21 
vtorri it does not work, of course, but that could be a start16:45.23 
chrisl If you edit the "CFLAGS_SO" value, that is what is used when building the "so" target16:46.27 
vtorri i set it to 'dll' ?16:46.57 
chrisl No, 'dll' is not a CFLAG16:47.20 
vtorri ha16:47.26 
  the flags to craete the shared lib ?16:47.37 
chrisl Yes, and you'll probably want to set DYNANIC_LIB_EXT=.dll16:47.57 
  Oh, actually, just DYNANIC_LIB_EXT=dll16:48.34 
  (no period)16:48.39 
vtorri or directly in configure.ac16:48.41 
chrisl I was going to suggest I would look at doing it when I have more time (I'm going out soon)16:49.50 
vtorri ok16:49.57 
  btw, it's DYNANIC and not DYNAMIC :)16:50.15 
chrisl Oops, I should fix that, too - thanks ;-)16:50.37 
  vtorri: I've taken note of the compiler options you posted above for creating the DLL, I'll have a look over the weekend16:53.12 
vtorri usually the import library name is libgs.dll.a and the dll name is libgs-$(major_version).dll16:54.02 
  the dll is installed in bin/ and not lib/ (contrary to unix)16:54.26 
  the import lib is installed in lib/16:54.34 
chrisl vtorri: okay, thanks. I will probably only look at building it, and leave the install stuff for a later date, but I'll see how it goes, and what else I have to do16:56.30 
vtorri and uname returns in MSYS : MINGW32_NT-6.116:56.48 
  so MINGW32* in the case should be sufficient16:57.10 
chrisl I have it using MINGW* in a couple of places already16:58.27 
Robin_Watts Gah. More bloody stream problems. I *hate* the mupdf stream interface.16:58.30 
chrisl bye!17:01.41 
mvrhel_laptop_ So in the windows phone environment, only the office app appears to have file picking capability. Other apps, (adobe reader included) find files by file type association from other apps (e.g. the office app or explorer). 17:28.49 
  microsoft is doing starting a big reorg. they have finally put windows and windows phone under the same group. maybe that will help this mess17:29.40 
Robin_Watts I'm sure windows phone now feels very happy it has a huge ship to cling to the side of.17:31.31 
  I'm not so sure that the windows group is so happy with it's new name: "Titanic"17:31.51 
  tor8: sorted it, I think.17:47.08 
SpNg is there a good editor with syntax highlighting for postscript?18:46.14 
malc_ SpNg: emacs18:48.24 
SpNg malc_: perfect. thank you18:51.01 
mvrhel_laptop_ Robin_Watts: that is funny18:52.15 
  and a fitting description18:52.24 
ray_laptop SpNg: Also "vim" (or "gvim" for Windows) knows about PS highlighting23:08.55 
  I prefer vim to emacs, but that seems to be totally a matter of religion23:09.56 
  both are widely available across platforms now-a-days23:10.45 
  The only thing missing from 'vim' (IMHO), which may be my laziness in setting it up, is the ability to match ] } and } when #ifdef / #else / #endif exists. Totally confuses it :-(23:13.29 
  (not PS -- C)23:13.41 
  I've made a few changes to the gs source files I own in order to work around this -- make sure the open and close block are BOTH in the same #if/#else clause23:15.38 
SpNg ray_laptop: awesome thanks. I actually prefer vim too. I typically use Sublime text during the day, but I'm decently comfortable in vim. I was playing with it earlier after emacs was mentioned. exactly what I was looking for ;-)23:19.17 
ray_laptop SpNg: Thanks. I thought I was the only "vi" neanderthal here ;-)23:31.13 
  I like -- it is quite funny23:33.58 
 Forward 1 day (to 2013/07/13)>>>