Bug#454550: Here is a conflict of our interests
reassign 454550 poppler 0.6.4-1 forwarded 454550 https://bugs.freedesktop.org/show_bug.cgi?id=9860 thanks Hi, The PS issue has been solved by using libspectre in Evince 2.22. The PDF problem remains, so I will reassign this bug to Poppler. -- Cheers, Sven Arvidsson http://www.whiz.se PGP Key ID 760BDD22 signature.asc Description: This is a digitally signed message part
Bug#454550: Here is a conflict of our interests
Hello Zack, The -DDOINTERPOLATE Ghostscript option was added to Evince at my request, logged as the following GNOME bug report (now closed): http://bugzilla.gnome.org/show_bug.cgi?id=319049 Fixing your bug by removing -DDOINTERPOLATE unconditionally would reopen the bug above. IMHO, we should formulate the rules for the desired Evince behaviour that suit us both before doing anything with the code or compilation options. -- Alexander E. Patrakov -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#454550: Here is a conflict of our interests
On Thu, Feb 7, 2008 at 5:06 AM, Alexander E. Patrakov [EMAIL PROTECTED] wrote: Hello Zack, The -DDOINTERPOLATE Ghostscript option was added to Evince at my request, logged as the following GNOME bug report (now closed): [...] Thanks for the counterexample. I agree that your file is rendered better with -dDOINTERPOLATE / equivalent behavior from the PDF renderer (I care about consistency between PDF and PS renderers as well as the behavior of each). Have you looked at my test cases? Do you agree that those are rendered better without interpolation? I think this is going to have to get fixed inside the renderers, because a document could easily have some images that should be interpolated and others that shouldn't; a global toggle (even if accessible in the UI) is not good enough. The heuristic that comes to mind is to enable interpolation only for images that are not being scaled up by more than 10% or so of their natural size. My test cases, and the other test cases in the Gnome bug to which this was forwarded, involve low-resolution bitmaps representing mathematical objects, scaled to large multiples of their natural size; it is very important to have sharp lines in that case. zw -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#454550: Here is a conflict of our interests
Zack Weinberg wrote: The heuristic that comes to mind is to enable interpolation only for images that are not being scaled up by more than 10% or so of their natural size. Here is a better approach that has correct limits and also handles non-uniform scaling and rotation. Note that the source bitmap defines a function of two real (not integer!) variables (xdest and ydest) that has a constant value in each rectangle corresponding to a source pixel (and discontinuities at the edges of source pixels). The value of the destination pixel should be the integral from this (piecewise-constant) function over the area covered by this destination pixel. -- Alexander E. Patrakov -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#454550: Here is a conflict of our interests
Zack Weinberg wrote: Have you looked at my test cases? Do you agree that those are rendered better without interpolation? Yes, and you have described them very well. Rendering them with interpolation produces clearly wrong results. However, questions stand 1) whether representing such sets of rectangles with low-resolution bitmaps is valid according to the PostScript and PDF specifications (hint: importing this into CorelDraw under Windows gives a warning about a low-resolution bitmap), and 2) if the standard does allow interpolation in this case, how widespread is this assumption of non-interpolated rendering in (broken) software that creates such files. In either case, your don't interpolate on upscaling heuristic looks good enough in theory, but is not currently implementable in practice for PostScript files processed via Ghostscript (because there is no such command line option for Evince to pass to Ghostscript). And that's a bug in Ghostscript. -- Alexander E. Patrakov -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#454550: Here is a conflict of our interests
On Thu, Feb 7, 2008 at 10:45 AM, Alexander E. Patrakov [EMAIL PROTECTED] wrote: Yes, and you have described them very well. Rendering them with interpolation produces clearly wrong results. However, questions stand 1) whether representing such sets of rectangles with low-resolution bitmaps is valid according to the PostScript and PDF specifications (hint: importing this into CorelDraw under Windows gives a warning about a low-resolution bitmap), and 2) if the standard does allow interpolation in this case, how widespread is this assumption of non-interpolated rendering in (broken) software that creates such files. I can't speak to point 1, but the test cases I provided were generated by Matlab, which is very widely used; I looked for a toggle to disable this behavior and could not find one. If you look at the upstream bugs that this one is forwarded to, and the bugs that those are merged with, and so on, you will see several more people with exactly the same sort of low-resolution bitmap generated by other scientific software, complaining loudly about the interpolation. Thus, whether or not this tactic fits the letter of the specifications, I think these low-resolution bitmaps (good phrase, btw) must be rendered uninterpolated. Note also that this is what every PostScript printer does. In either case, your don't interpolate on upscaling heuristic looks good enough in theory, but is not currently implementable in practice for PostScript files processed via Ghostscript (because there is no such command line option for Evince to pass to Ghostscript). And that's a bug in Ghostscript. Yes, that's what I meant when I said this has to be fixed in the renderers. zw -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]