Is using Data.Dynamic considered a no-go? - Haskell-Cafe - Haskell.org

newer
How to deal with utf-8?

Is using Data.Dynamic considered a no-go?

older
Re: [Haskell] Google Summer of...

Peter Verswyvelen

12 Feb 2009 12 Feb '09

12:43 p.m.

Haskell seems to have pretty strong support for dynamic casting using Data.Typeable and Data.Dynamic. All kinds of funky dynamic programming seems to be possible with these "hacks". Is this considered as being as bad as - say - unsafePerformIO? What kind of evil is lurking here? Cheers, Peter

Attachments:

attachment.html (text/html — 426 bytes)

Reply

Sign in to reply online Use email software

Show replies by date

Don Stewart

12 Feb 12 Feb

12:49 p.m.

bugfact:

Haskell seems to have pretty strong support for dynamic casting using Data.Typeable and Data.Dynamic.

All kinds of funky dynamic programming seems to be possible with these "hacks".

Is this considered as being as bad as - say - unsafePerformIO? What kind of evil is lurking here?

Inefficiencies and runtime errors? -- Don

Reply

Sign in to reply online Use email software

Achim Schneider

12:58 p.m.

Peter Verswyvelen wrote:

Haskell seems to have pretty strong support for dynamic casting using Data.Typeable and Data.Dynamic. All kinds of funky dynamic programming seems to be possible with these "hacks".

Is this considered as being as bad as - say - unsafePerformIO? What kind of evil is lurking here?

Uncertainty. Noone on #haskell seems to know how Typeable-dispatches compare to pattern matching wrt. performance, and I was either too lazy or too uninterested to measure it myself. Combined with a custom type class and existentials, they safe you from writing catch-all ADTs and thus give code flexibility (and conciseness), while retaining static typing, modulo unexhaustive pattern warnings. This is how XHB uses Typeable, and, IMHO, it's a Good Thing. -- (c) this sig last receiving data processing entity. Inspect headers for copyright history. All rights reserved. Copying, hiring, renting, performance and/or quoting of this signature prohibited.

Reply

Sign in to reply online Use email software

Lennart Augustsson

1:04 p.m.

They are not unsafe in the way unsafePerformIO is, but I regard them as a last resort in certain situations. Still, in those situations they are very useful. -- Lennart 2009/2/12 Peter Verswyvelen :

Haskell seems to have pretty strong support for dynamic casting using Data.Typeable and Data.Dynamic. All kinds of funky dynamic programming seems to be possible with these "hacks". Is this considered as being as bad as - say - unsafePerformIO? What kind of evil is lurking here? Cheers, Peter

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

Jonathan Cast

1:24 p.m.

On Thu, 2009-02-12 at 19:04 +0100, Lennart Augustsson wrote:

They are not unsafe in the way unsafePerformIO is,

I beg permission to demur: newtype Unsafe alpha = Unsafe { unUnsafe :: alpha } instance Typeable (Unsafe alpha) where typeOf _ = typeOf () pseudoSafeCoerce :: alpha -> Maybe beta pseudoSafeCoerce = fmap unUnsafe . cast . Unsafe Note that pseudoSafeCoerce = Just . unsafeCoerce

but I regard them as a last resort in certain situations. Still, in those situations they are very useful.

But I would agree with both of these. As long as you *derive* Typeable. jcc

Reply

Sign in to reply online Use email software

Lennart Augustsson

1:53 p.m.

You're quite right. You should only be allowed to derive Typeable. (Which could be arranged by hiding the methods of typeable.) On Thu, Feb 12, 2009 at 6:24 PM, Jonathan Cast wrote:

On Thu, 2009-02-12 at 19:04 +0100, Lennart Augustsson wrote:

...
They are not unsafe in the way unsafePerformIO is,

I beg permission to demur:

newtype Unsafe alpha = Unsafe { unUnsafe :: alpha } instance Typeable (Unsafe alpha) where typeOf _ = typeOf ()

pseudoSafeCoerce :: alpha -> Maybe beta pseudoSafeCoerce = fmap unUnsafe . cast . Unsafe

Note that

pseudoSafeCoerce = Just . unsafeCoerce

...
but I regard them as a last resort in certain situations. Still, in those situations they are very useful.

But I would agree with both of these. As long as you *derive* Typeable.

jcc

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

Peter Verswyvelen

2:55 p.m.

It would be interesting to see when you HAVE to use dynamics, e.g. when no other solution is possible in Haskell... Right now if I use it, it feels that I'm doing so because I'm too new to Haskell. On Thu, Feb 12, 2009 at 7:53 PM, Lennart Augustsson wrote:

You're quite right. You should only be allowed to derive Typeable. (Which could be arranged by hiding the methods of typeable.)

On Thu, Feb 12, 2009 at 6:24 PM, Jonathan Cast wrote:

...
On Thu, 2009-02-12 at 19:04 +0100, Lennart Augustsson wrote:

...
They are not unsafe in the way unsafePerformIO is,

I beg permission to demur:

newtype Unsafe alpha = Unsafe { unUnsafe :: alpha } instance Typeable (Unsafe alpha) where typeOf _ = typeOf ()

pseudoSafeCoerce :: alpha -> Maybe beta pseudoSafeCoerce = fmap unUnsafe . cast . Unsafe

Note that

pseudoSafeCoerce = Just . unsafeCoerce

...
but I regard them as a last resort in certain situations. Still, in those situations they are very useful.

But I would agree with both of these. As long as you *derive* Typeable.

jcc

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

Don Stewart

3:40 p.m.

Notably, extensible exceptions use dynamics, in conjunction with type classes and existentials. A number of solutions to the 'expression problem' involve dynamics. bugfact:

It would be interesting to see when you HAVE to use dynamics, e.g. when no other solution is possible in Haskell...

Right now if I use it, it feels that I'm doing so because I'm too new to Haskell.

On Thu, Feb 12, 2009 at 7:53 PM, Lennart Augustsson wrote:

You're quite right. You should only be allowed to derive Typeable. (Which could be arranged by hiding the methods of typeable.)

On Thu, Feb 12, 2009 at 6:24 PM, Jonathan Cast wrote: > On Thu, 2009-02-12 at 19:04 +0100, Lennart Augustsson wrote: >> They are not unsafe in the way unsafePerformIO is, > > I beg permission to demur: > > newtype Unsafe alpha = Unsafe { unUnsafe :: alpha } > instance Typeable (Unsafe alpha) where > typeOf _ = typeOf () > > pseudoSafeCoerce :: alpha -> Maybe beta > pseudoSafeCoerce = fmap unUnsafe . cast . Unsafe > > Note that > > pseudoSafeCoerce = Just . unsafeCoerce > >> but I regard them >> as a last resort in certain situations. >> Still, in those situations they are very useful. > > But I would agree with both of these. As long as you *derive* Typeable. > > jcc > > > _______________________________________________ > Haskell-Cafe mailing list > Haskell-Cafe@haskell.org > http://www.haskell.org/mailman/listinfo/haskell-cafe > _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

Sterling Clover

13 Feb 13 Feb

12:20 a.m.

SYB makes very heavy use of Typeable as well, although not, as I recall dynamics as such. Cheers, Sterl. On Feb 12, 2009, at 3:40 PM, Don Stewart wrote:

Notably, extensible exceptions use dynamics, in conjunction with type classes and existentials.

A number of solutions to the 'expression problem' involve dynamics.

bugfact:

...
It would be interesting to see when you HAVE to use dynamics, e.g. when no other solution is possible in Haskell...

Right now if I use it, it feels that I'm doing so because I'm too new to Haskell.

On Thu, Feb 12, 2009 at 7:53 PM, Lennart Augustsson wrote:

You're quite right. You should only be allowed to derive Typeable. (Which could be arranged by hiding the methods of typeable.)

On Thu, Feb 12, 2009 at 6:24 PM, Jonathan Cast wrote:

...
On Thu, 2009-02-12 at 19:04 +0100, Lennart Augustsson wrote:

...
They are not unsafe in the way unsafePerformIO is,

I beg permission to demur:

newtype Unsafe alpha = Unsafe { unUnsafe :: alpha } instance Typeable (Unsafe alpha) where typeOf _ = typeOf ()

pseudoSafeCoerce :: alpha -> Maybe beta pseudoSafeCoerce = fmap unUnsafe . cast . Unsafe

Note that

pseudoSafeCoerce = Just . unsafeCoerce

...
but I regard them as a last resort in certain situations. Still, in those situations they are very useful.

But I would agree with both of these. As long as you *derive* Typeable.

jcc

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

...
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

5943

Age (days ago)

5944

Last active (days ago)

Download

8 comments

6 participants

tags

participants (6)

Achim Schneider
Don Stewart
Jonathan Cast
Lennart Augustsson
Peter Verswyvelen
Sterling Clover