Image Rendering Regression

Details

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: None
• Fix Version/s:
• Component/s: None
• Labels:
None
• Environment:
Windows, Java 1.5

Description

Running ant testimage , I find several tests failing.

ARCHIVERGB – the version of the PNG file in trunk\pdfbox\src\test\resources\input\rendering is far closer to the AI than is the one that is output. I believe the problem has to do with the Stencil Mask.

HOTRODCMYK – same as ARCHIVERGB

test-landscape2 – not coming out in Landscape format any more

A few others are reported as "not identical" but they are visually identical – false-positives IMO. The above 3, however, have regressed since those image snapshots were taken ... version 1.2 IIRC.

Activity

Hide
Daniel Wilson added a comment -

Andreas' fix works great!

Show
Daniel Wilson added a comment - Andreas' fix works great!
Hide
Daniel Wilson added a comment -

Very nice correction on the FlateFilter.

And ... yes, the code was right and the test result was wrong in the past on the test-landscape2 case. Thanks for correcting that ... sorry I saw that one wrong.

Show
Daniel Wilson added a comment - Very nice correction on the FlateFilter. And ... yes, the code was right and the test result was wrong in the past on the test-landscape2 case. Thanks for correcting that ... sorry I saw that one wrong.
Hide
Andreas Lehmkühler added a comment -

I updated the landscape image in revision 1157533

Show
Andreas Lehmkühler added a comment - I updated the landscape image in revision 1157533
Hide
Andreas Lehmkühler added a comment -

I myself introduced the issue in revision 1081149 which is part of 1.6.0.

The problem was the calculation of the row length within the FlateFilter for predictor values >= 10. I fixed that in revision 1157529

Show
Andreas Lehmkühler added a comment - I myself introduced the issue in revision 1081149 which is part of 1.6.0. The problem was the calculation of the row length within the FlateFilter for predictor values >= 10. I fixed that in revision 1157529

People

• Assignee:
Unassigned
Reporter:
Daniel Wilson