[poppler] Followup Bug 32349 & Poppler: More shading fun ;-)

Andrea Canciani ranma42 at gmail.com
Fri Dec 31 09:05:44 PST 2010 

On Fri, Dec 31, 2010 at 4:50 PM, Andrea Canciani <ranma42 at gmail.com> wrote:
> On Fri, Dec 31, 2010 at 4:38 PM, Leonard Rosenthol <lrosenth at adobe.com> wrote:
>> I did "Save as PNG" on each of the files and then zipped them up.  You can find them at <http://dl.dropbox.com/u/1644474/Gradient_Issues.zip>.
>
> They look quite different from what I would expect to be the correct result.
> The other cases might be harder to check, but it's very easy to see that
> radial gradients in radial-gradient-one-stop should not contain colors
> different from red.
>
> I will try to modify the test a little so that it generates integer coordinates.
> This should make it easier to verify what the behavior should be.

I also made the gradient fully opaque and a little simpler.

The result is here, alpng with the expected output:
http://people.freedesktop.org/~ranma42/cairo/radial/reduced2/
The gradients in the files have the following extremes:
         /Coords [ -0.25 0 0.125 0 0 0.375 ]
         /Coords [ -0.375 0 0 24.625 0 25 ]
         /Coords [ -24.625 0 25 0.375 0 0 ]
         /Coords [ 0 0 0.375 0.25 0 0.125 ]
which by manual inspection define tangent circles (exactly).

Adobe Reader 9.4.1 on Ubuntu 10.4 displays these files as:
http://people.freedesktop.org/~ranma42/cairo/radial/acroread2/

Sorry for getting a little off-topic, but I believe that this is the hardest
radial shading case (which seems to be confirmed by the inconsistency
in the results from different PDF viewers) and so far nobody has been
able to confirm if the result I would expect is actually the correct one.

Andrea

>
> Andrea
>
>>
>>
>> Leonard
>>
>> -----Original Message-----
>> From: Andrea Canciani [mailto:ranma42 at gmail.com]
>> Sent: Friday, December 31, 2010 10:01 AM
>> To: Leonard Rosenthol
>> Cc: Thomas Freitag; Albert Astals Cid; poppler
>> Subject: Re: [poppler] Followup Bug 32349 & Poppler: More shading fun ;-)
>>
>> On Fri, Dec 31, 2010 at 3:28 PM, Leonard Rosenthol <lrosenth at adobe.com> wrote:
>>> Not sure what version of Acrobat/Reader on what OS platform you tested with, but Acrobat X Pro on Windows just opened and rendered these documents quite fine.  (though they look really weird!)
>> Yes, they are supposed to look weird, they're very artificial tests
>> which trigger different cornercases in cairo.
>>
>> Adobe Reader 9.4.1 on Ubuntu 10.4 produces these results:
>> http://people.freedesktop.org/~ranma42/cairo/radial/acroread/
>>
>> The test is structured as a 4x7 matrix. To indicate the errors I'll
>> number rows from 1 to 4 (top to bottom) and columns from 1
>> to 7 (left to right):
>>  - radial-gradient-one-stop looks incorrect in (2,2), (2,6), (3,2), (3,6)
>>  - all other tests look incorrect in (4,2).
>>
>> Can you confirm that the proposed rasterization (i.e.
>> http://people.freedesktop.org/~ranma42/cairo/radial/*.image.*)
>> is the correct one?
>> Does Acrobat X Pro agree with it or with Adobe Reader 9.4.1?
>> This is essentially the same question as http://forums.adobe.com/message/3316526
>>
>> Thank you for looking into this
>>
>> Andrea
>>
>>>
>>> Leonard
>>>
>>> -----Original Message-----
>>> From: poppler-bounces+leonardr=adobe.com at lists.freedesktop.org [mailto:poppler-bounces+leonardr=adobe.com at lists.freedesktop.org] On Behalf Of Andrea Canciani
>>> Sent: Friday, December 31, 2010 8:25 AM
>>> To: Thomas Freitag
>>> Cc: Albert Astals Cid; poppler
>>> Subject: Re: [poppler] Followup Bug 32349 & Poppler: More shading fun ;-)
>>>
>>> On Fri, Dec 31, 2010 at 3:10 AM, Thomas Freitag
>>> <Thomas.Freitag at kabelmail.de> wrote:
>>>> I made a lot of mistakes, but I'm quite close now, s. attached rendering of
>>>> Andrea's PDF. What is still up to do, is
>>>
>>> Wow, that's really good!
>>>
>>>> a) speed it up again
>>>> b) implement the shading extend in a correct way (I'd already a look at it,
>>>> it's also wrong implemented in Gfx)
>>>> c) run it again through my own regression test.
>>>
>>> I'd like to provide another set of "difficult" radial gradients,
>>> because the test
>>> I posted isn't checking some interesting cases (in particular the case where
>>> start and end circle are tangent):
>>> http://people.freedesktop.org/~ranma42/cairo/radial/
>>>
>>> This folder contains some pdf files and the expected rasterization.
>>> Adobe Reader seems to be unable to display these files (it just hangs),
>>> so I also created some "reduced" versions:
>>> http://people.freedesktop.org/~ranma42/cairo/radial/
>>>
>>> Adobe Reader can display the reduced files, but all of them look incorrect
>>> in the second column (start circle tangent inside end circle).
>>>
>>>>
>>>> But first I need a little bit to sleep :-)
>>>
>>> Sorry for trying so hard to make your code fail ;P
>>> FWIW I'm doing the same against my own radial gradient implementation
>>> in pixman (which right now fails with tangent circles, but I already have a
>>> patch to fix it).
>>>
>>> Andrea
>>>
>>>>
>>>> Thomas
>>>>
>>>> Am 30.12.2010 12:20, schrieb Thomas Freitag:
>>>>>
>>>>> I just recognized a bug with Andrea's PDF when drawing non centered
>>>>> circles. Want to fix that first before sending a new patch.
>>>>>
>>>>> Thomas
>>>>>
>>>>> Am 30.12.2010 11:29, schrieb Albert Astals Cid:
>>>>>>
>>>>>> A Dijous, 30 de desembre de 2010, vàreu escriure:
>>>>>>>
>>>>>>> Sorry, I attached Your "new.png", not mine. Here the correct one
>>>>>>> produced with the above changes.
>>>>>>
>>>>>> This looks good enough for me, can you send the full patch to the list?
>>>>>>
>>>>>> Thanks!
>>>>>>
>>>>>> Albert
>>>>>>
>>>>>>> Thomas
>>>>>>>
>>>>>>> Am 30.12.2010 10:31, schrieb Thomas Freitag:
>>>>>>>>
>>>>>>>> Hi Albert!
>>>>>>>>
>>>>>>>> I changed the SplashRadialPattern::getColor a little bit to solve
>>>>>>>> this. It is still a little bit different from the old one, in my
>>>>>>>> opinion a little bit better, but that's just a flavour, and in other
>>>>>>>> cases it could be a little bit more worse. So if You agree, I can send
>>>>>>>> a complete new patch.
>>>>>>>>
>>>>>>>> BTW, as You probably see, Andrea attached his PDF to the closed bug
>>>>>>>> 32349. I saw already yesterday, that also the new rendering with the
>>>>>>>> latest patch is quite better than the old rendering, but it's still
>>>>>>>> not shown what Acrobat reader shows. I'll have an additional look in
>>>>>>>> it, but I fear that this will be again a bigger job (the PDF has 128
>>>>>>>> radial shadings, and only some of them are still buggy!). So (if not
>>>>>>>> solvable in time) I would prefer to close this thread first, and then
>>>>>>>> reopen the bug or create a new one.
>>>>>>>>
>>>>>>>> Do You agree, or do You have other ideas?
>>>>>>>>
>>>>>>>> Thomas
>>>>>>>>
>>>>>>>> FYI, here the changes:
>>>>>>>>
>>>>>>>> GBool SplashRadialPattern::getColor(int x, int y, SplashColorPtr c) {
>>>>>>>>
>>>>>>>>   double xc, yc;
>>>>>>>>   int xs, ys;
>>>>>>>>   Guchar *bitmapAlpha;
>>>>>>>>
>>>>>>>>   bitmapAlpha = bitmap->getAlphaPtr();
>>>>>>>>   ictm.transform(x, y,&xc,&yc);
>>>>>>>>   xs = splashRound(xc);
>>>>>>>>   ys = splashRound(yc);
>>>>>>>>   // Because of rounding problems, coordinates could be
>>>>>>>>   // outside the bitmap. Reset them on the outer bound now
>>>>>>>>   // and let it up to the alpha channel if they are shown:
>>>>>>>>   if (xs<  0) xs = 0;
>>>>>>>>   if (xs>= bitmap->getWidth()) xs = bitmap->getWidth() - 1;
>>>>>>>>   if (ys<  0) ys = 0;
>>>>>>>>   if (ys>= bitmap->getHeight()) ys = bitmap->getHeight() - 1;
>>>>>>>>   if (bitmapAlpha[ys * bitmap->getWidth() + xs])
>>>>>>>>
>>>>>>>>       bitmap->getPixel(xs, ys, c);
>>>>>>>>
>>>>>>>>   else
>>>>>>>>
>>>>>>>>       return gFalse;
>>>>>>>>
>>>>>>>>   return gTrue;
>>>>>>>>
>>>>>>>> }
>>>>>>>>
>>>>>>>> Am 29.12.2010 23:28, schrieb Albert Astals Cid:
>>>>>>>>>
>>>>>>>>> First page, left of the "title". A few white pixels in there.
>>>>>>>>>
>>>>>>>>> Albert
>>>>>>>>>
>>>>>>>>> A Dimecres, 29 de desembre de 2010, Thomas Freitag va escriure:
>>>>>>>>>>
>>>>>>>>>> Am 29.12.2010 18:53, schrieb Andrea Canciani:
>>>>>>>>>>>
>>>>>>>>>>> On Wed, Dec 29, 2010 at 4:48 PM, Thomas Freitag
>>>>>>>>>>>
>>>>>>>>>>> <Thomas.Freitag at kabelmail.de>    wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> ...
>>>>>>>>>>>> I made a mistake when solving the problem with
>>>>>>>>>>>> altona_visual_1v2a_x3.pdf. I find now a better way to solve it,
>>>>>>>>>>>> which
>>>>>>>>>>>> also gives a better look of the printer paper back again.
>>>>>>>>>>>
>>>>>>>>>>> I'd like to point yo to another pdf whose rendering regresses with
>>>>>>>>>>> the
>>>>>>>>>>> patch: https://bugs.freedesktop.org/attachment.cgi?id=41506
>>>>>>>>>>
>>>>>>>>>> Albert, can You please just change two lines in the former patch?
>>>>>>>>>> 1. In SplashOutputDev.cc in the constructor of SplashRadialPattern,
>>>>>>>>>> replace
>>>>>>>>>> + bitmap = new SplashBitmap(splashRound(width) + 1,
>>>>>>>>>> splashRound(height)
>>>>>>>>>> + 1, colorMode != splashModeMono1, colorMode, gTrue);
>>>>>>>>>> through
>>>>>>>>>> +      bitmap = new SplashBitmap(splashRound(width) + 2,
>>>>>>>>>> splashRound(height) + 2, colorMode != splashModeMono1, colorMode,
>>>>>>>>>> gTrue);
>>>>>>>>>> 2. In Splash.cc in the painting routine radialShadedFill, where the
>>>>>>>>>> smaller extend circle is painted, replace
>>>>>>>>>> + drawPartEllipseLine(&pipe, 2* yo - curY, xMin, xMax);
>>>>>>>>>> through
>>>>>>>>>> +                drawPartEllipseLine(&pipe, curY, xMin, xMax);
>>>>>>>>>> (curY is alfready recalculated two line before!)
>>>>>>>>>>
>>>>>>>>>> Or just reapply the attached patch.
>>>>>>>>>> This solves the rendering regressions mailed by Andrea.
>>>>>>>>>>
>>>>>>>>>> Thomas
>>>>>>>>>>
>>>>>>>>>>> In the last row, half of the inner circle is transparent with
>>>>>>>>>>> poppler/master+radialsh.patch.
>>>>>>>>>>>
>>>>>>>>>>> Andrea
>>>>>>>>>>>
>>>>>>>>>>> PS: Sorry for removing most of the thread from this message, but
>>>>>>>>>>> gmail
>>>>>>>>>>> squashed it to just one level.
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> poppler mailing list
>>>>>>>>>>> poppler at lists.freedesktop.org
>>>>>>>>>>> http://lists.freedesktop.org/mailman/listinfo/poppler
>>>>>>>>>>>
>>>>>>>>>>> .
>>>>>>
>>>>>> .
>>>>>>
>>>>>
>>>>
>>>>
>>> _______________________________________________
>>> poppler mailing list
>>> poppler at lists.freedesktop.org
>>> http://lists.freedesktop.org/mailman/listinfo/poppler
>>>
>>
>

More information about the poppler mailing list