| <<<Back 1 day (to 2018/03/28) | 20180329 |
sebras | vtorri_: I tested pocorgtfo15.pdf on 1.12.0 source code with ASAN/valgrind and I can't reproduce your crash. | 10:20.55 |
| vtorri_: neither can I reproduce the crash using latest git HEAD source code. | 10:21.19 |
kens | FWIW I've tried several different versions of MuPDF on WIndows, and all have been succesful for me. | 10:21.27 |
sebras | kens: most of the images seem to be flate/dct | 10:21.57 |
kens | Oh, I forgot to look | 10:22.08 |
| Anyway, knowing the SHA1 of the code vtorri is using would be highly helpful, as well as the conditions for the crash, does this occur on opening the file, or on one of the 100 pages ? | 10:22.37 |
sebras | kens: hm.. but the file is also a valid zip-archive which contains furhte pdfs.. | 10:23.28 |
kens | Really ? That's well, weird.... | 10:23.42 |
sebras | kens: the name of the document is Proof of Code or Get The Fuck Out, so it seems they'ave packed the source code inside. | 10:24.15 |
kens | I don't see a PK at the begining of the file. | 10:24.24 |
moolc | sebras: furhte? taiwan is getting to you :) | 10:24.25 |
kens | The file does start 'oddly' with ILDA but that's valid for PDF as long as the %!PDF is in the first 1Kb | 10:25.04 |
sebras | moolc: mandarin _is_ a language that is quite special, so I wouldn't be surprised... | 10:25.17 |
kens | Is object 9999 the zip ? | 10:25.19 |
moolc | sebras: so they finally converted formoza to mandarin... news to me | 10:25.53 |
sebras | kens: I'm not sure. | 10:26.14 |
kens | I see what you mean, I can open it as a zip archive and its packed with stuff | 10:26.28 |
sebras | moolc: it is a mix of mandrain, minnan, japanese and the old aboriginal languages... | 10:26.44 |
kens | including more zip files | 10:26.46 |
sebras | kens: those zips don't contain pdfs though. | 10:27.41 |
moolc | sebras: treasure trove for a linguist.. or at least to sinologist.. cool | 10:27.41 |
kens | sebras, no apparently not | 10:27.50 |
| Offset 0x243a1e into the file is a stream which is a Zip | 10:30.36 |
| Object 9990 | 10:30.49 |
| Its between the info dict and the startxref keyword, so I think that's the thing | 10:32.35 |
| Its also not a valid stream (no /Length) and isn't referenced anywhere in the file as far as I can see | 10:33.12 |
| I'm sort of surprised a zip application is prepared to ignore 2Mb of 'garbage' (from the zip app point of view) while looking for a start of Zip. | 10:34.24 |
sebras | kens: the zip central directory is towards the end. | 10:34.54 |
kens | Presumably MuPDF is 'fixing' the invalid stream. It could be that vtorri is using a set of source where that code had a problem. | 10:35.03 |
| sebras, yes I'm seeing it as the stream in object 9990 | 10:35.28 |
| immediately after the /ID and before the startxref | 10:35.40 |
vtorri_ | sebras i open it then press the right arrow for next pages, then q to quit | 11:33.31 |
| i have then a seg fault | 11:33.40 |
| sebras i will open a bug with backtrace anyway | 11:34.03 |
kens | Did you get hte 'fie has unsaved changes' message ? | 11:34.22 |
| My code isn't totally up to date, but trying what you say worked OK for me (windows build) | 11:34.49 |
vtorri_ | kens something like that (i'm at work, so i can't be sure) | 11:39.37 |
kens | OK no problem | 11:39.44 |
| I'm sure otr8 and sebras will look at it | 11:39.52 |
vtorri_ | the messages was moreor less : invalid pdf, fixing it, then unsaved changes, then an abort | 11:40.22 |
kens | Hmm, I don't get a message baout the PDF being invalid, I just get offered to 'save the changes' | 11:40.43 |
vtorri_ | i will compile with debug symbols and give bt | 11:40.52 |
kens | No worries, when you have time | 11:41.02 |
kens | lunches | 11:44.32 |
paulgardiner | I'm seeing some strangeness linking with mupdf under VS2015: symbol conflicts between mupdf itself and thirdparty; e.g., _jpeg_free_large occurring both in jmemobs and jmemcust. Any thoughts? | 14:14.17 |
| Yeah, I know "don't use Windows" :-) Any other thoughts? | 14:14.49 |
kens | You are building MuPDF or building something which includes MuPDF and the 3rd party libraries already included in MuPDF ? | 14:15.30 |
paulgardiner | The latter, but the conflict seems to be between mupdf and mupdf's thirdparty libraries. | 14:17.03 |
kens | Probably best to get Robin_Watts to look at it | 14:17.53 |
Robin_Watts | paulgardiner: I bet this is a consequence of you manually listing libraries. | 14:18.32 |
| libmupdf already includes libthirdparty, but if you list both, then it'll try to include libthirdparty twice. | 14:18.59 |
paulgardiner | But it says one occurance in jmemobs and the other in jmemcust. | 14:19.46 |
Robin_Watts | Oh. Urm... jmemnobs = the no backing store memory handler for jpeglib. | 14:21.09 |
| jmemcust is the custom memory handler for jpeglib. | 14:21.25 |
| Should you really have both included? | 14:21.34 |
| Mupdf doesn't include jmemcust | 14:22.09 |
paulgardiner | I'd guess not, but this is just the VS2005 project automatically updated (which seemed to work for just building the mupdf app), so I haven't changed any config or defines. | 14:23.03 |
Robin_Watts | paulgardiner: OK, so then jmemcust must be coming from the SO build? | 14:24.24 |
| Did you try my suggestion of just not including jpeglib and jbig2dec in MuPDF? | 14:24.44 |
paulgardiner | Apparently not. libmupdf.lib(jmemcust.obj) libthirdparty.lib(jmemnobs.obj) | 14:25.54 |
Robin_Watts | ah, ok. | 14:26.42 |
paulgardiner | Not including jpeglib and jbig2dec would require turning off the symbol obfuscation in epage and then mupdf would be using possibly versions it isn't happy with. | 14:27.29 |
Robin_Watts | ok, so jmemcust is indeed part of libmupdf, rather than libthirdparty, so that is how we build stuff. | 14:29.03 |
| So we're back to my previous question about libs being included twice. | 14:29.43 |
| When I saw you this morning you said "I'm having to manually specify libs to include rather than VS doing it automatically as mupdf does it. But it's working, so I'm not going to worry about it." | 14:30.47 |
| Well, it's not working, so I'd worry about it :) | 14:30.55 |
paulgardiner | You misheard. I said "I'll probably just fiddle about and get it working and not worry about it". I never got there. | 14:31.34 |
Robin_Watts | ok, so let's try and fix that now. | 14:32.03 |
paulgardiner | I still have no idea why I need to provide the libraries explicitly. If I leave them out, I have many missing symbols | 14:32.40 |
| Perhaps it's the way I have added the projects to the solution, or the way I've specified dependancies | 14:33.19 |
Robin_Watts | paulgardiner: I'm saying let's figure that out now. | 14:33.33 |
paulgardiner | Sounds like a fine plan. | 14:33.50 |
Robin_Watts | Where can I find your solutions files to look? | 14:34.21 |
paulgardiner | Hang on. I'll just check I have everything pushed. Thanks for looking | 14:35.43 |
| It's on my socontrol branch. | 14:40.17 |
| It's a pain to build. You need to init the mupdf submodule, but you don't want to init all submodules because so now has three independent copies of mupdf. | 14:41.06 |
| You also have to run ./scripts/release.py smart-office-lib-windows to build the so lib the project needs. | 14:42.39 |
| Building mupdf, other than initializing the submodule is part of the VS project. | 14:43.16 |
Robin_Watts | paulgardiner: 2 mins. | 14:43.51 |
| paulgardiner: OK, back. | 14:51.06 |
| So which submodule do I want to init? | 14:53.37 |
| And what's the downside to me initing all the submodules other than disc space? | 14:54.28 |
paulgardiner | epage\platform\windows\apps\smart-office-nui\lib\mupdf | 14:57.28 |
| Downside just time and space | 14:57.41 |
| There's one more complication: the command to build smart-office-lib.lib places it in epage\platform\windows\apps\smart-office-lib. You have to put a copy in epage\platform\windows\apps\smart-office-nui\lib. | 15:00.06 |
Robin_Watts | ok. checked out and inited. | 15:00.08 |
| So what next? (Walk me through this, step by step if you would) | 15:00.27 |
paulgardiner | You need a submodule iniit from within the mupdf submodule unless you specified recursive | 15:01.12 |
| Then issue ./scripts/release.py smart-office-lib-windows from a command prompt | 15:01.47 |
Robin_Watts | paulgardiner: And that requires what VS version? | 15:02.26 |
| 2010 ? | 15:02.32 |
paulgardiner | I'm using 2015. 2010 may be okay | 15:03.17 |
| I think the free 2010 may have problems actually | 15:03.38 |
Robin_Watts | I have full 2010. | 15:03.45 |
paulgardiner | I think that should work. Wont be like for like with what I'm doing here, but I'd imagine the same problems will show up | 15:04.23 |
Robin_Watts | That script specifically requires 2015. I may have 2015 community edition installed. Let's see. | 15:05.02 |
paulgardiner | I have only the community edition | 15:05.44 |
| of 2015 that is | 15:05.59 |
Robin_Watts | paulgardiner: "compiler is out of heap space" | 15:11.17 |
| and this is on PC with 32Gig :) | 15:11.31 |
paulgardiner | WT.....? | 15:12.33 |
| I've not seen any problem like that here. | 15:12.56 |
Robin_Watts | building the romfss. | 15:13.35 |
paulgardiner | I have just 16Gig | 15:14.02 |
jogux | might be worth checking what service pack of 2015 etc. | 15:16.10 |
paulgardiner | I'm on 14.0.25431.01 Update 3 | 15:17.17 |
Robin_Watts | I'm behind you. | 15:18.04 |
| let eme update. | 15:18.07 |
| how do I update? | 15:18.35 |
paulgardiner | I've just discovered one other local change. I've just pushed it as a further commit on that branch | 15:19.51 |
Robin_Watts | Well, updating this has blown a hole in the rest of the afternoon I think. | 15:25.25 |
| let me look at the solutions files with an editor. Where are they? | 15:25.50 |
paulgardiner | Ooops sorry, hang on | 15:31.13 |
Robin_Watts | I found them. | 15:31.21 |
paulgardiner | epage\platform\windows\apps\smart-office-nui\lib\mupdfproj | 15:31.26 |
Robin_Watts | right. | 15:31.33 |
| So first observation... | 15:31.36 |
| you've edited the OutDir and InDir for Debug|Win32, but none of the others. | 15:31.58 |
| Actually, what I really want to do here is to start from a fresh MuPDF, update the projects myself, and then see if the dependencies work. | 15:33.58 |
paulgardiner | I think they did | 15:34.14 |
Robin_Watts | So it's the moving of the project files that has caused this problem? | 15:34.30 |
paulgardiner | Possibly, but I don't see why it should. | 15:34.48 |
Robin_Watts | Have you included these projects within a larger workspace? | 15:35.14 |
paulgardiner | I think I just systematically replaced ../.. by ../mupdf | 15:35.29 |
Robin_Watts | This update is going to take hours. | 15:35.34 |
| and I'll lay long odds it won't solve the problem at the end of it :( | 15:35.59 |
paulgardiner | I added the 4 projects to epage\platform\windows\apps\smart-office-nui\viewer-sample\viewer-sample.sln | 15:36.54 |
Robin_Watts | paulgardiner: If you do the build using the existing solution, do you need to manually specify dependencies ? | 15:37.38 |
paulgardiner | Not sure what you mean. The existing solution for what? | 15:38.45 |
Robin_Watts | Update the solution from VS2005 -> VS2010, and that gives you a new solution, plus 9ish new projects. | 15:40.40 |
| If modify (move) all those projects, and that solution do you get something that works with the dependencies correct? | 15:41.15 |
| I'm trying to figure out if it's the modifications that broke stuff, or the reimporting of the projects into a new solution. | 15:41.39 |
paulgardiner | Ah right. I didn't try that. When I moved the projects, I took on only the ones I needed | 15:41.52 |
Robin_Watts | this update says it's proceeding, but it's stuck at about 5% | 15:42.16 |
paulgardiner | I'll redo it here. I remember it taking a while, but not that long | 15:42.55 |
| Actually, I may have branch with them on | 15:43.58 |
| Ah, my mupdf master branch has a commit adding 2015 project files. Not moved or changed from the update I believe | 15:47.39 |
| I wonder... perhaps when moving the project files and systematically editing the references to the mupdf source files, perhaps I accidentally changed a string that didn't refer to a mupdf source file | 15:54.13 |
Robin_Watts | Update completed. Rebooted. Compiler still out of heap space. ass. | 17:28.26 |
| ok, so it was getting confused by the presence of extra fonts in the lib. Was trying to build a 19Meg font into romfss and having conniptions. | 17:48.48 |
| Forward 1 day (to 2018/03/30)>>> | |