Thank you for all your comments and additions, Andreas!
It's good to be back after quite a long time (enough to forget the good old habit of using JUnit!)
(In reply to comment #22)
> (In reply to comment #21)
> > The problem with getLengthBase() seems to point to a difficulty with
> > property-inheritance: strictly speaking, the footnote should inherit the
> > computed value for start-indent(),
> Sorry, I obviously meant "end-indent".
What I still cannot understand is why there is no such problem with start-indent = "body-start", which is resolved ok ...
(In reply to comment #17)
> Right, my first instinct would be to include footnotes for the table-header
> only on the first page that is spanned by the table, and for the table-footer
> only on the last page.
It's an interesting idea, and probably the easiest to implement.
Personally, I would have placed them all in the first page spanned by the table, although this would be a bit more problematic in terms of relative order between the footnotes.
What do other think in this regard?
Anyway, I think this is a situation not perfectly covered by the specs, which forbid footnotes in static-contents but say nothing about footnotes in table headers / footers, which are not so different.
The condition defining where a block area returned by fo:footnote is permitted does not explicitly take into account the situation when a single fo:foonote generates several anchor areas in different pages, although the definition
"The term anchor-area is defined to mean the last area that is generated and returned by the fo:inline child of the fo:footnote."
could maybe be read as a justification for placing all the notes in the last page where the table appears ...