[Haskell-cafe] Re: Type hackery help needed!

[Niklas Broberg]

Yes, this will surely do the trick, thanks a lot! :-)

I got as far as defining a TypeEq class myself in one of my attempts,
trying to trick the inference engine, but now seeing the full
ingenuity of the TypeCast class I realize how far from the solution I
really was. Again, thanks a million!

/Niklas


On 8/5/06, oleg at pobox.com <oleg at pobox.com> wrote:
>
> > A value of any type should be embeddable inside a build expression
> > of any result type, i.e. a -> b or b -> a cannot hold on Embed in
> > general.
>
> It seems you might benefit from local functional dependencies, which
> are asserted per instance rather than for the whole class. They are
> explained in
>
>         http://pobox.com/~oleg/ftp/Haskell/typecast.html
>
> Incidentally, that web page's source also gives an illustration of their
> use:
>         http://pobox.com/~oleg/ftp/Haskell/typecast.hs
>
> The page itself is written in HSXML, which had to deal with a similar
> problem: in HSXML, 'p' is a polyvariadic function, which has to accept
> as many strings as the user cares to specify -- and in addition,
> arbitrary HSXML fragments like 'em', 'code', 'cite', etc. The latter
> are too built by polyvariadic functions. So, at some point I had to
> force the constraints and tell the functions that their arguments are
> over and they better give some data structure. Incidentally, that's
> where [] notation comes in quite handy. The brackets are not just the
> embellishment; by lucky accident, they actually force the
> constraints. BTW, the rendering of the above HSXML code checks local
> file links (and inserts the file sizes into the HTML code, while at
> it). There is a version of the HSXML rendering that, in addition to
> formatting Haskell code, passes it to a Haskell system to verify its
> typing -- and to optionally run it as well. The complete source code
> for the renderers is
>         http://pobox.com/~oleg/ftp/Haskell/HSXML.tar.gz
>
>