Why isn't -rectypes command line option the default

Why isn't -rectypes command line option the default

[Erik de Castro Lopo]

Hi all,

If I try to define a recursive type like this:

    type 'a t = 'a * 'a t Lazy.t

the compiler won't accept it unless I add -rectypes to the compile
command.

I know I can avoid this by using a tag:

    type 'a t = Node of 'a * 'a t Lazy.t

but I was wondering if there was a reason why -rectypes wasn't the
default.

Cheers,
Erik
-- 
+-----------------------------------------------------------+
  Erik de Castro Lopo
+-----------------------------------------------------------+
"The very definition of hell is having to maintain someone 
else's Perl code." -- Anonymous

Re: Why isn't -rectypes command line option the default

[Zheng Li]

Erik de Castro Lopo <mle+ocaml@mega-nerd.com> writes:
> but I was wondering if there was a reason why -rectypes wasn't the
> default.
It's consciously disabled to avoid mysterious report of errors. Many careless
coding errors can be typed with recursive types, for example

# fun x-> x::x;;
- : ('a list as 'a) -> 'a list = <fun>

in this case, the programmer is probably thinking of connecting @ two
list. With recursive types, instead of a immediately type error report in
place, the function is typed without any problem, and finally some accident may
be happened far away from here, where the original function is likely to be under
several layers of composition and hard to recognized. 

So the main consideration is safety, though it's really fun to play with
recursive types sometimes.

-- 
Zheng Li
http://www.pps.jussieu.fr/~li

Re: [Caml-list] Re: Why isn't -rectypes command line option the default

[Erik de Castro Lopo]

Zheng Li wrote:

> So the main consideration is safety, though it's really fun to play with
> recursive types sometimes.

Ok, I can see why they aren't on by default, but using -rectypes on 
the whole file exposes to the whole file to these errors.

Wouldn't it be nicer to be able to specify the rectypes behaviour
per type? Maybe something like this would work:

	rec type 'a t = 'a * 'a t Lazy.t

Cheers,
Erik
-- 
+-----------------------------------------------------------+
  Erik de Castro Lopo
+-----------------------------------------------------------+
"Mutable state is actually another form of manual memory management: every 
time you over-write a value you are making a decision that the old value is 
now garbage, regardless of what other part of the program might have been 
using it." -- Paul Johnson

Re: [Caml-list] Re: Why isn't -rectypes command line option the default

[james woodyatt]

On Mar 20, 2007, at 19:13, Erik de Castro Lopo wrote:
>
> Wouldn't it be nicer to be able to specify the rectypes behaviour  
> per type? Maybe something like this would work:
>
> 	rec type 'a t = 'a * 'a t Lazy.t

You know... I have found myself on several occasions wishing for  
something like that.  A nitpick: I think I'd be happier with "type  
rec" as opposed to "rec type" since that seems symmetrical with the  
"let rec" phrasing.  Yes, I would expect "type rec a = ... and b  
= ... and c = ..." to allow all three types to be recursive.


—
j h woodyatt <jhw@conjury.org>
http://jhw.vox.com/

Re: [Caml-list] Re: Why isn't -rectypes command line option the default

[Erik de Castro Lopo]

james woodyatt wrote:

> You know... I have found myself on several occasions wishing for  
> something like that.  A nitpick: I think I'd be happier with "type  
> rec" as opposed to "rec type" since that seems symmetrical with the  
> "let rec" phrasing.

Sure, either way would be really great.

Erik
-- 
+-----------------------------------------------------------+
  Erik de Castro Lopo
+-----------------------------------------------------------+
"The question of whether a computer can think is no more interesting 
than the question of whether a submarine can swim." -- edsger dijkstra

Re: [Caml-list] Re: Why isn't -rectypes command line option the default

[Ville-Pertti Keinonen]

On Mar 21, 2007, at 4:13 AM, Erik de Castro Lopo wrote:

> Wouldn't it be nicer to be able to specify the rectypes behaviour
> per type? Maybe something like this would work:
>
> 	rec type 'a t = 'a * 'a t Lazy.t

It would probably be possible to allow that by default with no extra  
keyword, since errors are unlikely when writing a type explicitly.

But the ability to write recursive type aliases would not make  
actually creating recursively typed values all that convenient, you'd  
have to give the type explicitly (I'd rather just use a variant and a  
constructor).

Note that in the example using -rectypes posted by Zheng Li, the type  
'a t is never actually used for anything - presumably it's intended  
as an illustration to the reader, and could just as well be a comment  
-  the type inferred for fib is compatible, but not named int t.

Re: [Caml-list] Re: Why isn't -rectypes command line option the default

[Gerd Stolpmann]

Am Mittwoch, den 21.03.2007, 13:13 +1100 schrieb Erik de Castro Lopo:
> Zheng Li wrote:
> 
> > So the main consideration is safety, though it's really fun to play with
> > recursive types sometimes.
> 
> Ok, I can see why they aren't on by default, but using -rectypes on 
> the whole file exposes to the whole file to these errors.
> 
> Wouldn't it be nicer to be able to specify the rectypes behaviour
> per type? Maybe something like this would work:
> 
> 	rec type 'a t = 'a * 'a t Lazy.t

You can do that:

type 'a t = T of 'a * 'a t Lazy.t

Recursive types are accepted without -rectypes when the recursion
crosses a variant type. Note that the runtime costs are usually
negligible.

Gerd
-- 
------------------------------------------------------------
Gerd Stolpmann * Viktoriastr. 45 * 64293 Darmstadt * Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------