(I originally posted this last week, but mail delivery failed. 
Reposting via web interface.)

26 nov 2007 kl. 22.47 skrev Frank Mittelbach:
> anything other than my first four options (from the original mail) or only some of them ?

One I would expect is the ability to put the _center_ of the float area at a particular position. 
Since this area varies in size, it would not be a special case of (2) or (3).

>     * The ratio of t1 to t2 is fixed by the design and a float AAA can be
>      placed into the middle position if neither t1 nor t2 become too
>       small. (Downside of this kind of layout might be that the positioning of
>       the floats drastically varies from page to page.)

Also, this is the thing I would expect to use for a golden ratio positioning; by enforcing a 
golden ratio t1:t2, the same horizontal line will split the column and the float in parts by this 
ratio. (To make this work, t1 would have to include any top floats and t2 any bottom floats.)

>     * The end position of t1 is fixed (vertically) so that a middle float
>       always starts on the same point on a page. Further restriction then that
>       t2 is not getting smaller as a certain value.
> 
>     * The starting starting position of t2 is fixed so that the bottom of the
>       middle floats always appear on the same vertical position on the page,
>       again with some further restrictions to the size of t1 this time.

I'm not so enthusiatic about those two. Having a text block of fixed size above or below all 
floats seems ... kind of annoying, really; a top or bottom float should always be preferable 
to this.

If you want a typographical rationale, it might be good to think about the reader's 
expectations when following the text. A middle float is an _interruption_ in the running text 
-- what you're reading continues further down in the same column -- but if there in all 
columns are floats whose starting or ending positions all align then it's easy to interpret 
these floats as being aligned to the boundary between two different texts. IOW, upon seeing

  ttttt ttttt ttttt ttttt
  ttttt ttttt ttttt ttttt
  ttttt ttttt ttttt ttttt

  AAAAA BBBBB CCCCC DDDDD
  AAAAA BBBBB CCCCC DDDDD
  AAAAA BBBBB CCCCC DDDDD
  AAAAA BBBBB       DDDDD
        BBBBB ttttt DDDDD
  ttttt       ttttt
  ttttt ttttt ttttt ttttt
  ttttt ttttt ttttt ttttt
  ttttt ttttt ttttt ttttt
  ttttt ttttt ttttt ttttt

it's easy to think that what follows the third line of the first column is the first line of the 
second column.


>     * An obvious extension to the above could be that there are a list of
>       vertical starting points to choose from and some mechanism/logic to
>       select one of them

With facilities for info in the log file explaining the choices made, I presume.

>     * ...other ideas...

It might be a good idea to allow for some kind of offset to values used for ratio calculations. 
Some designers might want to make proportions relative to the \textheight area, whereas 
others prefer them relative to the \paperheight area. I suspect the preferences may even 
vary depending on whether the float stays in the column or sticks out into the margin.

Lars Hellström