David

 > Has anybody of the LaTeX3 team yet taken a look at the ltxgrid package
 > by Arthur Ogawa?

yes, i did. both are independent developments that happened more or less in
parallel and i learned about it only afterwards.

 > While it does not have the figure placement
 > folderol of xor, it does offer nice ways of extending output
 > routines, both in the area of the actual code as well as magical
 > penalties (<-20000), can make longtable play together with multiple
 > columns and so on.

the main goal for the xor proto-type for me was to work on a conceptual
algorithm for float placements that offers flexibility while maintaining
usability, eg with a faily bounded running time. i was not concerned at that
time with a more generic approach to swapping output routines in and  out,
though as you said it is something worth having for greater flexibility and
probably clearer design.

 > It does not do what xor does out of the box, but it offers much nicer
 > hooks for playing together with others.  The magical penalty stuff is
 > basically what I had been proposing, only that there is missing an
 > allocation macro for the same, and that there is missing a _context_
 > for the same, since one might want to have one and the same penalty
 > behave differently whether one is in an output routine for a
 > particular column, or just rushing a footnote box through penalty
 > processing in order to get margine notes and so on.

yes, and context is something concidering as a general concept, eg something
that you could offer as a data type that can be set and queried ...

 > It is probably not a bad idea to skim for some ideas...  although you
 > probably have done so already.

yes, but not very thoroughly back in 2000 and since then xor got to a sudden
hold for private reasons. so it is certainly worthwise to redo that exercise

frank