Shading pattern regressions

[timvandermeij]

This is a tracking issue after the excellent triage done by @THausherr on several shading pattern issues. The PDF files below worked after #4192 (with the viewer preview at http://107.21.233.14:8877/a2a6295988a73d2/web/viewer.html), but not anymore with the current master.

Below is a list of files that definitely regressed. I have looked through the entire 4-shading-pattern category to create this list. We need to determine the exact cause of the issue.

1. https://f.cloud.github.com/assets/2175324/261931/2532062c-8d0a-11e2-8fa6-845fdbcb735f.png (from Unsupported shading type: 4 #2948, rename .png to .pdf)
2. http://www.archimate.nl/content/bestanden/archimate_made_practical_2008-04-28.pdf (from Error rendering images or graph or something like this #3462)
3. http://files.shareholder.com/downloads/AMDA-MMXS5/0x0x827039/370849E8-1114-474D-963D-74CE277CA3B6/TRIP_1Q15_Investor_Presentation_v3.pdf (from PDF is missing background & has extra artefacts #6183, although this also looked bad in Implements shading types 4-7 #4192, but at least the right pattern was there)
4. http://ctan.cs.uu.nl/graphics/pgf/contrib/pgfplots/doc/pgfplots.pdf (from Interpolation is handled incorrectly. #6231, especially page 128 looked much better in http://107.21.233.14:8877/a10a71131aab410/web/viewer.html)
5. https://issues.apache.org/jira/secure/attachment/12647093/coons-function.pdf (from Shadings rendered with additional lines #6284, looked almost perfect in http://107.21.233.14:8877/a10a71131aab410/web/viewer.html aside from the lines to the bottom of the page)

I have bisected this with file 1 and found the following:

1e8d70a is the first bad commit
commit 1e8d70a
Author: Yury Delendik [email protected]
Date: Thu Feb 13 08:20:00 2014 -0600

Prescales shading pattern packed data indices

When we find the cause of this problem, we must add tests to avoid any new regressions.

/cc @yurydelendik

[THausherr]

https://issues.apache.org/jira/secure/attachment/12652009/pattern-shading-2-4-idMatrix.pdf
The old version has a small gouraud shaded triangle at the bottom left, the current version doesn't.

Postscript code can be found here:
https://issues.apache.org/jira/browse/PDFBOX-2141

[THausherr]

https://issues.apache.org/jira/secure/attachment/12658599/LATTICE1.pdf
worked partly in the old version (wrong size, tiles missing), but nothing in the new one.

[THausherr]

https://issues.apache.org/jira/secure/attachment/12658600/LATTICE2.pdf
worked partly in the old version (very small), but totally weird stuff in the new version.

[THausherr]

Another one:
https://issues.apache.org/jira/secure/attachment/12646800/eci_altona-test-suite-v2_technical_H.pdf

(That file is impossible to render correctly (even Adobe Reader doesn't render it perfectly) and too complex to debug, but it is a sort of success indicator)

[THausherr]

Yeah, you did it! A lot of problems I mentioned have been fixed. The Gouraud shadings (types 4 and 5) work fine now, although only those without a function. If you look at the ECI file, you will see that it is almost perfect, only the first rectangle is missing, which is a type 1 shading (that isn't supported yet by PDF.js, so it seems). I'll open some new issues about stuff unrelated to the regression that I found in the meantime.

Re "we must add tests to avoid any new regressions", that is why I mentioned that some of the files (those I created for PDFBox) are Apache licensed, so you can use them. I mentioned the PostScript source because that makes it easy to modify them, and they can be converted to PDF with ghostscript.

[timvandermeij]

This issue will be closed soon as the patch will be merged. @THausherr Could you create separate issues for the remaining problematic PDFs in this issue that are not related to the regression? Especially for the three that you provided here as they are not tracked elsewhere. Thank you!

[timvandermeij]

All PDFs in this issue have either been fixed or are now tracked in separate issues.

[THausherr]

You meant to type "now", I guess...

[timvandermeij]

Indeed I did. Corrected :)