| | <<<Back 1 day (to 2018/03/13) | 20180314 |
| ray_laptop | mvrhel: Robin_Watts: I've created a file made of 6 copies of an image that is 4610x3456 on a page which if run at -r1000 has 95,551,752 colors transformed of which 83,338,169 are cache misses resulting in a total of 83,338,215 calls to TetrahedralInterp16. A profile shows it to be 27% of the processing time. | 04:22.07 |
| | so if anyone wants to work on speeding up Tetrahedral interpolation, I can provide this file (no photo copyright restrictions since I took it) | 04:23.22 |
| | note the 83m Tetrahedral calls run in 3.15 seconds in a profile build | 04:26.37 |
| kens | Hmm that's not great timing for a report on RC1 | 09:12.16 |
| | So 123 pages of bitmaps to look at this time | 09:14.12 |
| kens | begins at page 0 | 09:14.35 |
| chrisl | Hmm, I think that's related to deekej's patch :-( | 09:52.18 |
| kens | Oh :( | 09:52.25 |
| chrisl | Actually, I think that might be just a ommission, so we might be okay | 10:00.26 |
| kens | Well it doesn't affect the rendering tests anyway | 10:00.51 |
| chrisl | No, it's not going to invalidate the testing, nor should it require a new rc | 10:01.45 |
| | I'll discuss it with deekej when he appears | 10:02.03 |
| kens | OK | 10:02.09 |
| chrisl | Although, I'm going to be out for a couple of hours from ~10:20.... hopefully he'll spot it on gs-devel | 10:02.47 |
| kens | Yeah I'll be out at the same time.... | 10:02.59 |
| chrisl | I figured that.... | 10:03.20 |
| | Wednesday's not a normal day for me to be out, but I had to rearrange | 10:03.51 |
| | kens: I have a concern about the release testing..... | 10:04.22 |
| kens | Oh | 10:04.28 |
| | What's the problem ? | 10:04.33 |
| chrisl | Given the image issues that came up last time, it occurred to me that we might have missed similar issues because of the jpeg passthrough diffs | 10:05.04 |
| kens | Which one, the downscaling or the ICC profile ? | 10:05.27 |
| chrisl | The downscale | 10:05.34 |
| kens | Hmm, yes that's possible I guess | 10:05.43 |
| chrisl | I'm wondering if it's worth doing a test with passthrough disabled | 10:06.07 |
| kens | You think the JPEG passthrough shopwing diffs is hiding some of the scaling problems ? | 10:06.11 |
| | We can certainly do that. | 10:06.18 |
| | Once I remember how :-) | 10:06.24 |
| chrisl | I'm not convinced it's hiding issues - but I'm not quite convinced it isn't, if you see what I mean | 10:06.46 |
| kens | "unknown unknowns" | 10:07.00 |
| chrisl | Yes :-) | 10:07.13 |
| kens | Well, lets do the ones we have and then run another sequence, it doesn't take long to create the bitmaps | 10:07.43 |
| | I'll check the code and work out how to disable it after lunch | 10:08.07 |
| chrisl | I think it's pretty easy - just set the default to 0 in gdevpsdf.h | 10:08.46 |
| kens | Yeah I thought I had a command line way too, maybe not | 10:09.04 |
| chrisl | Well, either way works | 10:09.19 |
| kens | I'll check it later, just want to eyeball a few more pages before I go out | 10:09.48 |
| chrisl | I haven't even started - just been on the phone for an age | 10:10.32 |
| kens | NP I've just finished page 38 | 10:10.59 |
| | OK pages 0 to 40 completed. Will do more after I get back | 10:13.27 |
| chrisl | deekej: did you see the mail on gs-devel about a patch to unix-inst.mak? | 12:43.14 |
| velix | kens: Did you read my flodding? | 13:23.03 |
| kens | err no sorry | 13:23.16 |
| | I'm only just back | 13:23.27 |
| | OK so no, you can't just render one specific font to paths, its all or nothing | 13:25.54 |
| | You could impleent such a thing by using the exsiting code as a template, but I don't think we'll do it. | 13:26.17 |
| | We also won't attempt to 'unround' rounding errors. If that's what's in the input, that's what we'll put in the output | 13:26.50 |
| velix | kens: Technical background: Some fonts aren't allowed to be embedded. That feature would ROCK office world. | 13:28.18 |
| kens | If the font can't be embedded we won't embed it. We also won't extract the outlines, because that would be illegal | 13:28.43 |
| velix | kens: We're not allowed to embedd our university fonts in PDF... so we're doomed. | 13:28.44 |
| | kens: it's allowed to embedd them as vector paths ;) | 13:29.01 |
| kens | Use the pdfimage devices to get an image as a PDF file | 13:29.04 |
| | velix, that depends entirely on your jurisdiction. If we extract the outlines then we could be held accessory to copyright infringement | 13:29.42 |
| velix | kens: I think, it depends on the license of the font. | 13:30.04 |
| | kens: kens: I'm playing around with extracting the object and mering it overlaying it with another PDF. I'll blog about it when I'm done. | 13:30.29 |
| | kens: But the funny thing is the WMF/EMF thing in office. They're doing pretty hacky stuff there. | 13:30.46 |
| kens | Its not exactly clear *what* they are doing | 13:31.01 |
| velix | kens: Do you have a complex EPS? I could make some tests in old/new Office. | 13:31.24 |
| kens | Off hand, no | 13:31.35 |
| velix | kens: isn't there a good EPS example on the web? The tiger one? | 13:31.48 |
| kens | tiger is very simple | 13:31.55 |
| velix | :D | 13:31.58 |
| kens | You'd want one with Separation, DeviceN, CMYK and CIE colours, overprinting, TYpe 1, 2, 3, 42 and CIDFonts with base types of al lthe preceding fonts | 13:32.51 |
| | Plus patterns, perhaps forms | 13:33.01 |
| velix | Okay, I'll create one in Corel. | 13:33.06 |
| | kens: Did you know that Corel makes heavy use of Ghostscript? | 13:33.15 |
| kens | Some nice shadings, type 6 and 7 would be hard to reproduce | 13:33.18 |
| | Corel uses Ghostscript ? | 13:33.31 |
| velix | yes | 13:33.33 |
| kens | In what way ? | 13:33.39 |
| velix | since 4-5 version numbers | 13:33.40 |
| | kens: Reading several foreign vector formats. | 13:33.56 |
| kens | Hmm.... | 13:34.02 |
| velix | kens: Of course, I can't tell you which. But I think, EPS and others. | 13:34.12 |
| kens | Well technically they can do that. Presumably they are shipping a copy of Ghostscript then ? If so they must include the licence | 13:34.24 |
| velix | kens: yes, they're shipping it. | 13:34.35 |
| | 1 sec | 13:34.36 |
| | I think, they're either using 9.14 or 8.64 | 13:36.02 |
| | https://community.coreldraw.com/talk/coreldraw_graphics_suite_x5/f/coreldraw-graphics-suite-x5/25637/how-ghostscript-is-helping-corel-draw | 13:36.21 |
| kens | Well provided they use Ghostscript as an external process, and include the licence with the copy they ship, they are probably legal | 13:36.29 |
| velix | ahh, better: https://support.corel.com/hc/en-us/articles/216513208-GPL-Ghostscript | 13:36.47 |
| | Even the sourcecode ;) | 13:37.15 |
| kens | Yeah | 13:37.25 |
| | If they modify an open source product they ahve to include their modifications | 13:37.39 |
| chrisl | I'm surprised they admit it, frankly | 13:38.09 |
| kens | 8.64 a very old version | 13:38.13 |
| velix | kens: maybe they're using 9.14 - I've got both in the gs folder. can't tell right now. | 13:38.34 |
| | kens: Need to analyze, which dlls are accesed. | 13:38.42 |
| | accessed | 13:38.44 |
| kens | The link downloads source which is 8.64 | 13:38.46 |
| velix | :) | 13:38.54 |
| | CorelPS2PDF.cpp ?!?! | 13:39.14 |
| | They're creating the PDFs using Ghostscript ? | 13:39.24 |
| kens | Are you surprised ? | 13:39.35 |
| velix | kens: chrisl is. | 13:39.51 |
| kens | They've been able to create PostScript forever, making PDFs would be hard, much easier to get someone else to do it. | 13:40.12 |
| chrisl | I'm surprised they *admit* it, not surprised they're using it | 13:40.14 |
| kens | We get this a lot from various companies | 13:40.19 |
| velix | CString strCommandLine = _T("-q -dSAFER -dNOPAUSE -dBATCH -dEPSCrop -sDEVICE=pdfwrite -dCompatibilityLevel=1.4 -dAutoFilterColorImages=false -dAutoFilterGrayImages=false -dColorImageFilter=/FlateEncode -dGrayImageFilter=/FlateEncode -sOutputFile= | 13:40.31 |
| | :) | 13:40.37 |
| kens | Yes | 13:41.10 |
| | Calling CreateProcess so I guess they're launching a full blown Ghostscript heavyweight process | 13:41.53 |
| chrisl | Well, they'd have to | 13:42.12 |
| kens | Yeah | 13:42.22 |
| velix | But I don't think, they're realling creating the PDFs that way. The output has OGCs and are version 1.7 | 13:42.22 |
| kens | Sure, I'd guess they are importing PostScript that way, by converting to PDF first | 13:42.44 |
| velix | s/realling/really | 13:42.44 |
| | kens: ahhh! | 13:42.50 |
| | Sure, -dEPSCrop ;) | 13:43.19 |
| | We talked about this yesterday. | 13:43.23 |
| kens | I'm more curious about this softek product. | 13:46.05 |
| | It'll either be not very good, or using a PostScript interpreter | 13:46.17 |
| | And I only know 2 likely to be able to do this task | 13:46.37 |
| | They are *claiming* to have implemented an EPS 'parser and interpreter' which would be ambitious | 13:48.01 |
| | velix do you know which version(s) of Word shipped with this Softek product ? | 13:51.29 |
| chrisl | kens: I've worked back to page 60 in gs diffs - I've a list of 8 diffs I need to investigate further | 13:57.45 |
| kens | Oh :-( | 13:57.54 |
| | I'm in the 40s sometwhere | 13:58.03 |
| | And 1 I need to look at | 13:58.09 |
| | I'll carry on | 13:58.18 |
| chrisl | They may not be regressions, there's 2/3 I'll probably get you to look at - color shifts in the pdfwrite/ps2write output | 13:58.42 |
| kens | That sounds ominous | 13:58.59 |
| | ok just done 43 | 13:59.14 |
| ray_laptop | strikes me that do simplify the release, after we'd gone through the bitmaps for RC1, we could compare to that instead of the previous release | 13:59.15 |
| | that way RC > 1 would be much less painful | 13:59.55 |
| kens | chrisl if you want o look at your diffs I'll press on | 14:01.24 |
| chrisl | kens: I'm going to get a coffee, then I'll check them | 14:01.51 |
| kens | OK | 14:01.57 |
| velix | I've used it in 1997, 2003, 2007, 2010, 2013 | 14:09.36 |
| | ohh, and 2000 of course. I was beta tester of 2000 | 14:09.54 |
| kens | Hmm, OK there's a reasonably recent version installed on my daughter's PC I htink (educational edition) I can look at that | 14:10.09 |
| velix | kens: It has been disabled in recent versions. | 14:12.31 |
| kens | I thought you could turn it back on with a registry change | 14:12.51 |
| velix | yes, but I don't know if YOU would do it ;) | 14:12.59 |
| kens | Well, its not *my* PC :) | 14:13.08 |
| velix | Hehe ;) | 14:13.14 |
| kens | Anyway, I only want to look at the software, I just need it to exist | 14:13.28 |
| velix | What harmful scripts can I run in EPS to break hack the PC ? | 14:13.37 |
| kens | Well you can write files to the disk | 14:13.57 |
| | And overwrite them of course | 14:14.14 |
| | But mostly these sorts of security bug revolve around crafted exectuion of memory | 14:15.46 |
| | Buffer overruns and so on | 14:16.07 |
| | Good grief, someone thinks its a smart idea to do a colour swatch as a JPEG..... | 14:20.26 |
| | chrisl I've completed pages 0 - 59. I have one file to investigate further | 14:24.19 |
| | But first, some coffee | 14:24.29 |
| velix | Perhaps their implementation is bugger. | 14:28.01 |
| | oh my... | 14:28.04 |
| | s/bugger/buggy | 14:28.08 |
| kens | Very possibly there's a buffer overrun or similar | 14:28.24 |
| | MS probably paid them a one-off fee so they don't get support and can't fix it. | 14:28.40 |
| | Or something like that. Maybe Softek don't know how to fix the problem | 14:28.56 |
| velix | kens: their website look pretty... old ;) | 14:29.00 |
| kens | So does ours :-) | 14:29.10 |
| velix | Sorry for asking, but can Ghostscript export EMF/WMF ? | 14:29.11 |
| kens | No | 14:29.17 |
| velix | kens: no, yours look nerdish. | 14:29.20 |
| kens | But it would be possible to do | 14:29.24 |
| velix | I had always problems to export WMF/EMF from Corel or Illustrator and place them in Word. | 14:29.51 |
| | EPS never was a problem. | 14:29.56 |
| chrisl | kens: I thought PDF 2.0 dropped transfer functions? | 14:31.19 |
| velix | kens: That's the Tiger in WMF, from Ghostscript website imported to Office 2010 and extracted from the DOCX (ZIP-archive): http://ge.tt/95LRlzo2 | 14:32.36 |
| | As you can see, the EPS is embedded in there. | 14:33.19 |
| | Wow, even Corel can read it. | 14:33.45 |
| kens | chrisl yes it did | 14:33.49 |
| chrisl | Huh, interesting that PDF 2.0 FTS has tests for transfer functions, than! | 14:34.10 |
| kens | :-) | 14:34.27 |
| | Probably held over from the 1.x tests | 14:34.36 |
| | When I open the WMF file the tiger is not correctly rendered in Paint | 14:35.11 |
| velix | PAINT ? :D | 14:35.20 |
| kens | Well its what opened the WMF when I dragged it onto VS | 14:35.35 |
| velix | My Paint cannot even read WMF here :) | 14:35.41 |
| | Oh, it can. | 14:35.51 |
| | Wow. | 14:35.52 |
| | :D | 14:35.57 |
| | It looks the same as in Word. | 14:36.25 |
| kens | Well yes that does contain the EPS. I'm not sure I understnad why | 14:36.27 |
| velix | So their interpretation is bad. | 14:36.30 |
| | Let me print it to PDF from Word. | 14:36.39 |
| kens | Very bad if they can't even get Tiger right | 14:36.42 |
| velix | Woooooow. it's correct. | 14:37.10 |
| | You know what ??? | 14:37.12 |
| | I know why they embedd it ! | 14:37.18 |
| | For viewing and printing on GDI images or printers, they're using some kind of translation | 14:37.39 |
| | For printing on Postscript, they're using the embedded EPS. | 14:37.47 |
| | It's exactly as it should look like. | 14:37.55 |
| kens | Yes I'd have to guess its something like that | 14:38.01 |
| velix | EPS in WMF. Wow ;) | 14:38.09 |
| kens | I didn't know that was possible... | 14:38.17 |
| chrisl | That kind of thing used to be pretty common | 14:38.21 |
| velix | So we're all suprised today. | 14:38.36 |
| kens | I'm baffled as to why they think an EPS in a WMF is better than an EPS | 14:38.53 |
| chrisl | kens: the color shifts I'm seeing are transfer functions.... | 14:38.58 |
| kens | Oh.... | 14:39.04 |
| velix | kens: Because they don#t have to interpret the EPS at all and don't have a problem with the buffer overflow. | 14:39.18 |
| chrisl | Which I didn't think we'd changed. | 14:39.20 |
| kens | But we shouldn't be producing PDF 2.0 files | 14:39.21 |
| velix | kens: the printer or ghostscript will have the trouble. | 14:39.28 |
| kens | Compromising a printer is less of a problem | 14:39.43 |
| chrisl | kens: some of these are straight to ppm | 14:39.43 |
| kens | Then I'm puzzled, I didn't expect any such changes | 14:39.54 |
| | can you give me a number ? | 14:40.00 |
| velix | kens: I've seen videos with Doom running on a Printer :) | 14:40.01 |
| | kens: 42 | 14:40.05 |
| kens | Yes, but that's just silly | 14:40.12 |
| chrisl | kens: https://ghostscript.com/~regression/release__gs/compare70.html#1335 | 14:40.25 |
| kens | Yeah that's odd | 14:40.49 |
| | Its possible I ignore transfer functions from PDF 2.0 files | 14:41.04 |
| | So that would be a difference from 9.22 let me look | 14:41.22 |
| chrisl | That's it being applied now | 14:41.28 |
| kens | (actually I didn't realise we had incorporated the PDF 2.0 tests yet) | 14:41.44 |
| | It wasn't applied before ? That's just weird | 14:42.03 |
| | OK so Acrobat doesn't display the star | 14:42.36 |
| | Oh wait | 14:42.46 |
| | That's an Ink annotation | 14:42.50 |
| | We didn't handle those before I added the code for PDF 2.0, now we do | 14:43.05 |
| | So I reckon that's not a transfer function, its a progression because the annotation is now handled | 14:43.20 |
| | (I believe its a 2.0 annotation) | 14:43.27 |
| chrisl | Er, oh, that's the wrong link - hang on..... | 14:43.48 |
| | Here's a better one: https://ghostscript.com/~regression/release__gs/compare68.html#1225 | 14:44.04 |
| kens | Hmm, yes OK so that's a PDF 1.7 file | 14:44.48 |
| | Interesting, Acrobat displays both images the same | 14:45.43 |
| chrisl | So, maybe we're *not* applying the transfer now? | 14:46.27 |
| kens | Not sure | 14:46.33 |
| | Seems to me there should be a difference, but it may well depend on the colour model, I'm just fiddling with the Acrobat settings | 14:46.53 |
| | OK going to have to check out the code in more detail | 14:48.05 |
| chrisl | There's also this one, same thing: https://ghostscript.com/~regression/release__gs/compare68.html#1217 | 14:48.33 |
| kens | I have a feeling I looked at these before | 14:48.52 |
| | Presuambly when I made the commit that caused the change | 14:49.14 |
| chrisl | The last one that looks dodgy is: https://ghostscript.com/~regression/release__gs/compare87.html#1605 | 14:49.18 |
| kens | That one looks definitively wrong | 14:49.46 |
| | It is the psdcmyk device though | 14:50.13 |
| velix | kens: https://msdn.microsoft.com/en-us/openspecifications/dn646765 | 14:50.16 |
| kens | I'd try a sane device instead | 14:50.19 |
| velix | kens: Microsoft released the complete specification for WMF. | 14:50.24 |
| kens | Well, feel free to implement a WMF device for Ghostscript :-) | 14:50.40 |
| velix | There it is: http://www.schnarff.com/file-formats/wmf/WindowsMetafileFormat(wmf)Specification.pdf | 14:51.00 |
| | Many Postscript Stuff inside. | 14:51.08 |
| | kens: I'm a PHP guy. Can I implement it in PHP? :) | 14:51.39 |
| kens | Ghostscript doesn't understand PHP | 14:51.54 |
| velix | "The Windows Metafile Format (WMF) POSTSCRIPT_PASSTHROUGH record sends arbitrary data directly to the driver. The driver is expected to only process this data when in PostScript mode. For more information, see the POSTSCRIPT_IDENTIFY (section 2.3.6.30 ) Escape record." | 14:52.01 |
| | It's like we thought. | 14:52.14 |
| | It would be nice, if Ghostscript would support this. Since Office "offically" can't read EPS anymore. | 14:52.57 |
| kens | Seems like its a variation on EPS+preview. You just need to wrap up the bitmap preview a bit differently | 14:53.39 |
| | But we can't do that right now | 14:53.46 |
| velix | wow: https://tools.ietf.org/html/draft-seantek-image-wmf-emf-00 | 14:54.07 |
| | Nice draft with good history. | 14:54.15 |
| | Hehe: "Windows Metafiles are historically very buggy." :))) | 14:54.58 |
| kens | chrisl I don't think we should be applying the transfer function unless we apply the halftone | 14:55.20 |
| | Its debatable whether that should be applied in a contone output I guess | 14:55.42 |
| chrisl | kens: Ah, that rings a vague bell | 14:55.55 |
| kens | We have two images with halftones (type 10) applied. One has a transfer function one does not. | 14:57.22 |
| | So if we go by the rule that contone output doesn't apply the halftone, then the images should be the same. | 14:57.51 |
| | But if we do apply the transfer rfunction, even for contone output, then the two shuld be different | 14:58.10 |
| | I can't immediately see the commit which changed this | 14:59.15 |
| chrisl | I can bisect it in a bit | 14:59.53 |
| kens | I'm guessing it was deliberate, I don;t remember why :-( | 15:00.13 |
| chrisl | This one is strange: https://ghostscript.com/~regression/release__gs/compare87.html#1605 | 15:00.16 |
| kens | Yes I'd try that one with a sane device | 15:00.30 |
| | psdcmyk is not a sane device | 15:00.38 |
| chrisl | Running pdfwrite then thru ppmraw I see the dashed line problem, but not the recycle graphic corruption | 15:00.55 |
| kens | That's odd then | 15:01.08 |
| | I'm sure I've not seen that problem before | 15:01.16 |
| | I'd bet the recycle graphic corruption is either a glitch, or specific to that device | 15:01.36 |
| | Funny thing about the dashed line, it looks like its anti-aliased or something, not linework | 15:02.14 |
| velix | kens: This one has been generated by a cloud service, which Microsoft suggests. Not bad: http://ge.tt/8du4nzo2 | 15:02.26 |
| kens | velix I'm kind of busy now, working on the next release | 15:02.51 |
| velix | pardon | 15:03.08 |
| kens | chrisl that dashed line is an image | 15:03.54 |
| chrisl | Uh-oh :-( | 15:04.07 |
| | So, I better bisect that, too | 15:04.46 |
| kens | yes please | 15:04.53 |
| | I don't understand how these got missed in RC1 | 15:05.03 |
| | Oh, the display device renders that image correctly | 15:05.16 |
| chrisl | I'm just going to try it in rc1 | 15:05.26 |
| kens | Ah good plan | 15:05.35 |
| | Hmm, display device is still OK with that image at 300 dpi | 15:06.02 |
| chrisl | You have to put it trough pdfwrite first | 15:06.18 |
| kens | D'oh | 15:06.25 |
| | Hmm, pdfwrtie auto rotates the page | 15:07.08 |
| chrisl | Happens in rc1, too :-( | 15:07.16 |
| kens | Damn, I wonder how we missed that :-( | 15:07.26 |
| chrisl | I get openjpeg warnings from that file | 15:07.55 |
| kens | Non conformant codestream ? | 15:08.34 |
| chrisl | Yeh | 15:08.39 |
| kens | Hmm, let me try this with auto rotate off | 15:08.51 |
| | OK so if we don't rotate the page, then its OK | 15:09.32 |
| | I wonder if we're really looking at a JPEG passthrough problem | 15:09.50 |
| | Previously it got 'fixed' because we decompressed and recompressed it | 15:10.12 |
| | Hmm, but I get no warnings on the pdfwrite output file, so that seems unlikely | 15:10.37 |
| chrisl | Is that a JPEG image? | 15:11.01 |
| kens | Ah, D'oh, its a jp2k | 15:11.03 |
| | We don't pass those through | 15:11.21 |
| chrisl | What about leaving pdfwrite autorotation, and changing the rendered orientation with LeadingEdge? | 15:12.25 |
| kens | Haven't tried | 15:12.33 |
| | Not sure if I can with the display devioce | 15:12.55 |
| | Its odd that Ac robat has no problem with it | 15:13.47 |
| | rotated either way | 15:13.55 |
| chrisl | I don't think the problem is with pdfwrite | 15:14.05 |
| kens | You may be correct, I was slowly coming to that conclusion | 15:14.17 |
| | Can I leave you to bisect those while I go look at my other problem chrisl ? | 15:14.58 |
| chrisl | Sure | 15:15.16 |
| kens | OK the one I had to look at is a progression. It works properly now producing 1 page, whereas before it produced 2 pages (one being an error). Which caused bmpcmp to complain | 15:20.28 |
| chrisl | The dashed line problem is caused by: http://git.ghostscript.com/?p=ghostpdl.git;a=commitdiff;h=38e0c643e62087c089c59c9ad44fc468dd09cd37 | 15:26.13 |
| | Another one for Robin_Watts | 15:26.31 |
| kens | Ah, well that probably explains why its different rotated | 15:26.35 |
| | Oh, its the special sekrit stuff | 15:27.18 |
| Robin_Watts | hmm? | 15:31.47 |
| kens | commit above causes a regression | 15:31.59 |
| velix | kens: PDF-1.5 is the first version, which can do layers. | 15:32.13 |
| Robin_Watts | of which file? | 15:32.14 |
| kens | But only when the page is rotated | 15:32.19 |
| chrisl | https://ghostscript.com/~regression/release__gs/compare87.html#1605 | 15:32.20 |
| Robin_Watts | Ta. | 15:32.24 |
| kens | You have to pdfwrite it first, so that AutoRotatePages rotates the page | 15:32.42 |
| Robin_Watts | ooh, eck. | 15:32.43 |
| kens | velix, possibly yes | 15:32.53 |
| velix | kens: just tried it in Acrobat :) | 15:33.02 |
| kens | It'll be noted in the specification which version introduced it | 15:33.22 |
| | OK so if I 'undef' the TransferFunction in pdf_draw.ps the halftone problem file it looks the same as Acrobat | 15:36.45 |
| | I'm guessing that I added code in .resolveht6 to preserve TransferFunctions from type 6 halftones, the same as the code for type 1 halftones | 15:38.19 |
| | So I believe those two files are 'correct', or at least that the PDF interpreter is correctly dealing whith them. Whether teh rendering code should be applying halftone transfer functions when its not halftoning is a different matter completely | 15:39.12 |
| | I do vaguely recall doing this | 15:39.21 |
| | chrisl I believe it is this commit: | 15:40.45 |
| | http://git.ghostscript.com/?p=ghostpdl.git;a=commit;h=83ef35dce760f68019afd1b07bac8118cd9837a2 | 15:40.45 |
| | I think the PDF 2.0 transfer funxtion deprecation was only the transfer function in the graphcis state, not from halftones. | 15:42.05 |
| chrisl | kens: Yep that's the commit - so I guess we're good on that | 15:43.22 |
| kens | I think so yes | 15:43.27 |
| | Its debatable whether we should apply the transfer function, but that's the graphics library not the interpreter | 15:43.43 |
| | And at least the type 1 & 6 halftones now behave the same, consistency is at least something | 15:44.02 |
| | SO we're just left with the one from Robin ? | 15:44.22 |
| chrisl | Well, that was it, as far as the issues I saw in gs | 15:44.23 |
| kens | Yeah on to PCL and XPS I guess | 15:44.35 |
| Robin_Watts | chrisl: So, you could just revert that commit for the release. | 15:44.49 |
| | but I'm looking now. | 15:44.55 |
| chrisl | I'd rather revert it on master, and pull that into the release | 15:45.27 |
| | kens: actually, there was this one, too: https://ghostscript.com/~regression/release__gs/compare122.html#2642 | 15:48.25 |
| kens | That's a file which is indeterminate | 15:48.43 |
| | It often comes up in the cluster tests | 15:48.58 |
| chrisl | Hmm :-( | 15:49.30 |
| kens | I think we are rendering content there, and the library hands back different results sometimes, I don't know why | 15:50.09 |
| | Its a type 4 imsage or something | 15:50.22 |
| Robin_Watts | I can reproduce the problem here (I rotated the PDF rather than pdfwriting it) | 15:50.36 |
| kens | Either way works, just wanted you not to fall into the same error I did :-) | 15:50.55 |
| chrisl | I'm not seeing anything new in PCL | 15:54.46 |
| kens | No all looks clear to me, no new diffs, no changes | 15:55.12 |
| | XPS diffs, 75 is a progression I beleive | 15:55.48 |
| | As is 19 | 15:56.05 |
| chrisl | Um 75?? | 15:56.30 |
| kens | Sorry 73 | 15:56.37 |
| | can't read again | 15:56.40 |
| | Too much staring at bitmaps | 15:56.46 |
| | 69 I think I recall from when I did the changes, its neither correct nor incorrect, merely different | 15:57.11 |
| | And its pretty small differences | 15:57.21 |
| chrisl | That's what i thought | 15:57.25 |
| kens | So XPS looks fine now | 15:57.36 |
| | chrisl if you want to try a run with JPEGPassThrough disabled I believe that -dPassThroughJPEGImages=false will do it. | 16:28.01 |
| chrisl | kens: I'll look into it..... | 16:29.25 |
| Robin_Watts | chrisl: I think you should revert that commit. | 16:31.36 |
| | I'll work on another one to clean stuff up a bit. | 16:32.01 |
| chrisl | Robin_Watts: Okay. As I said, I'd like to revert it on master, so the two branches don't go too far out of sync | 16:32.12 |
| Robin_Watts | chrisl: Of course. | 16:32.18 |
| chrisl | I'll do that, then - thanks | 16:32.42 |
| Robin_Watts | In an image enumerator, we have 2 rectangles, "rect" (which is the area for which data is supplied), and "rrect" (the area that actually needs to be rendered). | 16:33.12 |
| | When we're using the mitchell scaler, rrect gets boosted a bit to allow for the support pixels required, and that seems dirty to me. | 16:33.55 |
| chrisl | Bum... "error: could not revert 38e0c643e... Tweak bitmap interpolater patch size." | 16:34.17 |
| Robin_Watts | chrisl: Let me have a go. | 16:34.32 |
| chrisl | Robin_Watts: Thanks - given that it needs manual intervention, that's probably safer | 16:35.04 |
| Robin_Watts | Off to a good start. Let's try using the gs tree instead... | 16:35.45 |
| | chrisl: So you're running the tests again without waiting for my fix? | 16:54.13 |
| chrisl | Robin_Watts: I started them before we talked about it - you can abort them if you like | 16:54.51 |
| Robin_Watts | Yeah, let me stop them for now. | 16:55.04 |
| kens | This set of tests were with the JPEG passthrough disabled | 16:55.30 |
| chrisl | I'm just concerned that with the jpeg passthrough touching *so* many files, it *just* might be obscuring other issues | 16:55.47 |
| Robin_Watts | kens: right, but they would still have shown the stuff I'd just fixed. | 16:55.54 |
| kens | Yes htey would | 16:56.00 |
| | I was just mentioning it, so that if we do them again, we get it right :) | 16:56.16 |
| Robin_Watts | OK, so what I was saying before... we have rect, and rrect. | 16:58.07 |
| | I'm going to add drect too. | 16:58.14 |
| | So rect will be the area for which data is supplied. | 16:58.21 |
| | rrect will be the area for which data must be rendered. | 16:58.46 |
| | dreect will be the area for which data must be decoded. | 16:58.56 |
| | That way I won't have to guess how rrect has been munged. | 16:59.14 |
| chrisl | I have to go - I'll pick things up tomorrow morning. | 17:12.38 |
| kens | OK goodnight chrisl | 17:12.49 |
| velix | I've updated bug #699111 | 20:53.37 |
| | oops, wrong channel | 20:53.40 |
| | Forward 1 day (to 2018/03/15)>>> | |
Log of #ghostscript at irc.freenode.net.