[Haskell-cafe] Avoiding undecidables

[Daniel Fischer]

Am Samstag 05 Dezember 2009 21:51:26 schrieb Michael Snoyman:
> On Sat, Dec 5, 2009 at 10:33 PM, José Iborra <pepeiborra at gmail.com> wrote:
> > Since you have already decided MPTCs are ok, in my opinion undecidable
> > instances are fine here.
> > But I realize this is not the usual stance, so I might be wrong.
> >

As far as I understand, all UndecidableInstances does is tell the compiler "Maybe instance 
checking won't terminate, try anyway", so they're dangerous *only during compilation*, 
once things compile, everything's dandy.

> Sounds reasonable to me. I'm waiting for the boogey man to jump out
> though
>
> and explain why undecidables here will get me gored to death by a raptor
> [1].

They won't. They might suck you into a black hole, but raptors are very specialised.


> >
> > instance Convert Integer Double where convert = Just . fromIntegral
> > instance TotalConvert Double    -- that's all

instance TotalConvert Integer Double

> >
> Interestng approach. The current approach is *almost* the same, just
> leaves
>
> off the default definition. I'd be wary of putting in a definition like
> that; although it saves a line of typing, it let's a partial function get
> in which could cause trouble for unsuspecting users. But it is a
> possibility.

But only if they declare misbehaved instances of TotalConvert. Whenever a reasonable 
instance TotalConvert x y exists, the Convert instance should satisfy the demands of the 
default definition.

>
> > Cheers,
> > pepe
>
> [1] http://xkcd.com/292/