English version
Accueil     À propos     Téléchargement     Ressources     Contactez-nous    

Ce site est rarement mis à jour. Pour les informations les plus récentes, rendez-vous sur le nouveau site OCaml à l'adresse ocaml.org.

Browse thread
parameterized pattern
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: 2006-11-14 (23:12)
From: Don Syme <Don.Syme@m...>
Subject: RE: [Caml-list] parameterized pattern

> I just did a quick scan of some F# docs and 
> I saw nothing. What did you have in mind?

.NET type parameters are extensional, i.e. "you can always find out what
'a is at runtime".  In particular in C# you can just write "typeof(T)",
and in F# "(type 'a)", in each case getting a System.Type value.
Supporting exact runtime types was a design decision we made in the
early design stages for .NET generics. 

As a result all .NET languages that support generics (polymorphism) have
extensional polymorphism. It gets used heavily in the kind of
meta-activities we're all familiar with: marshalling, pretty-printing,
debugging (yes, Visual Studio 2005 shows you the values of "T", except
when they've been optimized away). It also gets used internally in some
libraries for adhoc optimization purposes, e.g. generating efficient
comparison functions for default .NET comparers based on type arguments,
and the F# matrix library uses it to detect when generic matrices are
really floating point matrices, hence thunking out to more efficient
matrix routines.  

There are downsides to extensional polymorphism (e.g. you can wind up
take up extra registers passing type parameters), but they don't seem to
bite in practice.  At the last ML Workshop a group at Cambridge
University recently reported on an experiment to modify core-OCaml to
pass runtime types, and IIRC saw no significant performance loss. 

FWIW if you're interested I'd also like to mention the huge impact OCaml
had on the design of .NET generics and C# 2.0, which I've never properly
described on this list.  It was seeing and experiencing polymorphic
programming in OCaml and SML that made us persevere with the long and
torturous process of convincing Microsoft that they should add such an
"experimental and academic" feature as generics to their flagship
runtime and languages.  Of course we were in reality just making 1970s
ideas work in practice, but at least now even Visual Basic has generics.


-----Original Message-----
From: caml-list-bounces@yquem.inria.fr
[mailto:caml-list-bounces@yquem.inria.fr] On Behalf Of brogoff
Sent: 09 November 2006 16:22
To: skaller
Cc: caml-list@yquem.inria.fr
Subject: Re: [Caml-list] parameterized pattern

On Thu, 9 Nov 2006, skaller wrote:
> On Thu, 2006-11-09 at 05:19 +0000, Jon Harrop wrote:
> > On Thursday 09 November 2006 01:45, brogoff wrote:
> > > It's a pity, as I've often wished that OCaml supported the
> > > polymorphism that GCaml has, but I don't think that's going to
> > > It would probaby make more sense to create a separate language at
> > > point, since OCaml is complicated enough.
> >
> > I think F# provides some form of extensional polymorphism.

I just did a quick scan of some F# docs and I saw nothing. What did you
have in

> > I'm not convinced that it is a good idea yet...

For almost any given language feature, there will be people who like it,
people who don't. Do you think having class based OO in OCaml is a good
I find it useful, especially since OCaml records are far too
restrictive, but
I hope that in some future ML that there are other approaches as the
class/object system is complex, and the interactions with "core ML +
is tricky.

That said, the class system is being used and it won't go away, and some
people really like it.

> Well FYI Felix has traditional (open) overloading, but since it
> doesn't allow traditional C++ style dependent name lookup because
> that would destroy parametricity of polymorphic functions,
> something else was needed.
> So it now has first order typeclasses to solve this problem.

Did you consider GCaml style generic functions?

-- Brian

Caml-list mailing list. Subscription management:
Archives: http://caml.inria.fr
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
Bug reports: http://caml.inria.fr/bin/caml-bugs