Drawing floating-point lines with Bresenham

Hi!

   I've written a graphics library that sup****ts fast Bresenham line 
drawing.  I can give an integer starting point, integer number of 
pixels, error increments for minor/major steps, and an initial error 
value for the first pixel to draw.

Now I would like to draw lines with floating-point endpoints.

Does anyone know how to initialize the error terms to correctly draw 
lines with non-integer endpoints?!

A link to an actual implementation in a graphics library would be nice!

(There _should_ be some driver code out there as many hardware 
accelerators seem to be integer-bresenham only).

thanks for your help,
Thomas Kindler