Hi All, As can be seen in Bugzilla #50852 [1], the current implementation of tagged PDF has fundamental limitations that prevent some features from working. Among others, an empty table cell will not show up in the structure tree, such that a screen reader will effectively shift other cells by one column.
Also, building a structure tree using a preliminary XSLT does not scale well, and is not even necessary since the structure tree is almost readily available in form of the FO tree. Of course it’s possible to warp the current code into something that more or less does the job, but besides not solving the performance issue that would make it too difficult to understand and maintain. I’d like to work with Peter on re-architecturing the tagged PDF code. The broad idea is to integrate the construction of the structure tree into the current processing pipeline (XSL-FO -> FO tree -> Layout Managers -> Area Tree -> Rendering etc.). We will work on a temporary branch forked off Trunk so that interested parties can follow our progress. Once the work is done we will call for a vote to merge it back to Trunk. Any comments or suggestions are welcome. Thanks, Vincent [1] https://issues.apache.org/bugzilla/show_bug.cgi?id=50852 See also comment #21 https://issues.apache.org/bugzilla/show_bug.cgi?id=50852#c21 And mailing list: http://markmail.org/message/mn7jdbxmjdq7ey52