[Haskell-cafe] Generics for constructing Rows

[José Pedro Magalhães]

Hello,

On Thu, Aug 20, 2009 at 16:54, Max Desyatov <explicitcall at googlemail.com>wrote:

> Sean Leather <leather at cs.uu.nl> writes:
>
> > I'm not sure the problem you're running into is strictly a generic
> > programming (GP) one. Typically, GP takes code that is often written
> > and generalizes it, so that it doesn't have to be written for multiple
> > datatypes.
>
> That seems to be GP problem, as your solution doesn't scale well when I
> wan't to add/remove/change fields in the `Row` record.  The perfect way
> as I see it, would be just editing `Row` data declaration, nothing else.
> Studying few papers about GP in Haskell, I reckon this could be
> represented as generic traversal, using my `Row` declaration with
> `Either`.  I don't see really good way to write a generic producer from
> `[String]` to version of `Row` without `Either`.  But SYB doesn't
> provide a way for passing type-class-parametric functions to gmapT, and
> SYB-with-class has large overhead of its usage.  I don't have enough
> time to find out how this can be written in SYB-with-class, if it really
> can be
> written.  The restriction of EMGM was described in my initial message.


Indeed SYB doesn't work here because Typeable-based run-time type comparison
only works for monomorphic types. Doing something like

readRow l = gmapT (mkT (\(Left (Just ri) :: E Name) -> Right $ l `atMay` ri
> >>= readMay))
>

would work, but this, of course, is not what you want. I'm guessing the
polymorphic typeOf previously described by Oleg [1] could help here, were it
integrated in SYB.

I don't think syb-with-class will help you here, since it only adds
modularity to the type-based function extension. I think you would still
have to write a case for every field in the Row record.

Multirec would possibly work, were it not for the fact that it doesn't
support parametric datatypes yet...


Cheers,
Pedro

[1] http://osdir.com/ml/haskell-cafe@haskell.org/2009-03/msg00212.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.haskell.org/pipermail/haskell-cafe/attachments/20090821/47903166/attachment.html