Chris Rowley writes:
 >   I forgive Sebastian Rahtz!! (Well for one thing, anyway
 >                                ... there's still hyperref:-) )

you obviously haven't looked at passivetex yet

 > for never having supplied us humble coders with what he claims is `the
 > obvious algorithm for float positioning (including spanning floats)'.

its from the same family as the Carlisle algorithm for word
counting. it goes

WHILE it_doesnt_look_nice()

implementation of "it_doesnt_look_nice" is in local libraries

 > Not necessarily help in designing this wonderful algorithm but in
 > formulating, or finding in the literature, the rules and heuristics for
 > judging whether a float placement is acceptable and/or good.

one data point you have to consider is "what the picture looks
like". it makes a difference whether it is a line diagram or a
photograph, for instance, and how dark it looks.

I suspect this "floating silver bullet" doesn't exist.  Shock, Horror!
Was Knuth wrong in his entire initial premise that computers could
solve typography? You have to wonder about a man who rates Georges
Perec over Patrick O'Brian.