[MLton] SML documentation tool(s)

[Stephen Weeks]

> when to write an ML library and when to make an ML binding for an
> existing C library? For some reasons we better trust(?) and are keen
> on an ML implementation but in the case of a mature C one we should
> not reinvent the wheel (see gmp as an example). What are the good
> arguments for choosing one over the other?

As I see it, here are the main tradeoffs.

benefits of ML library
 * portability
 * safety
 * naturally produce ML style interface
 * easier to build

benefits of ML binding for C library
 * less work (both initial and ongoing)
 * more likely to be correct and/or fast if we don't have domain knowledge

For libm, gmp, opengl, and other large libraries, the amount of saved
work and improved correctness and performance clearly outweighs any
advantage we might get from writing in SML.  For things like popt,
base64 conversion, URL manipulation, etc, the tradeoff is probably the
other way.  The right way to go is usually pretty clear.

> Another thing: other than arguing about the documentation we could
> collect the ideas what libraries would we like to have. A Wiki page
> seems natural to this. I guess it's worth putting on the main page
> but some people might disagree here, so let's discuss it.

The main page does seem a bit much, but I could see a link off of
Development or StandardML.  Note that we already have a libraries page
too.