try and tail call

try and tail call

[Christophe Raffalli]

consider this piece of code:

  (try
    let b = f a in
    (fun () -> g b)
  with
    (fun () -> h a)) ()


Is the call to "g" a tail call and is it omptimized as such ?

If not, how to encode a feature like this one:

  try
    ...
  end
    ... (* the exception raised here are not handles by with but 
propagated *)
  with
    ...

By the way I think this "try ... [end ...] with ..." syntax is usefull 
anyway because you have often a bug when
you write

try
  let b = f a in (* you know this may raise Not_found *)
  g b (* you assume wrongly that this can not raise Not_found *)
with
  Not_found -> h a

The unwanted capture of Not_found could be avoided with try ... end ... 
with ...

Christophe Raffalli

 

Re: [Caml-list] try and tail call

[Chris King]

On 10/21/06, Christophe Raffalli <raffalli@univ-savoie.fr> wrote:
>
> consider this piece of code:
>
>   (try
>     let b = f a in
>     (fun () -> g b)
>   with
>     (fun () -> h a)) ()
>
>
> Is the call to "g" a tail call and is it omptimized as such ?

Yes, it is, at least in the assembly code output by ocamlopt.  (Neat
trick, btw.)

> If not, how to encode a feature like this one:
>
>   try
>     ...
>   end
>     ... (* the exception raised here are not handles by with but
> propagated *)
>   with
>     ...

What's typically used is something along the lines of:

match (try Some (f a) with Not_found -> None) with
| Some b -> g b
| None -> h a

though I wouldn't be surprised if the method you used above is
slightly faster since it avoids creating unnecessary option objects
(the anonymous functions are statically allocated).

> By the way I think this "try ... [end ...] with ..." syntax is usefull
> anyway because you have often a bug when
> you write
>
> try
>   let b = f a in (* you know this may raise Not_found *)
>   g b (* you assume wrongly that this can not raise Not_found *)
> with
>   Not_found -> h a
>
> The unwanted capture of Not_found could be avoided with try ... end ...
> with ...

Many people agree with you... I find almost always that I want
something like the syntax you describe above, rather than what Caml
provides natively.  Martin Jambon provides an example syntax extension
in his Camlp4 tutorial[1] which allows the use of "let-try-in-with"
blocks:

let b = try f a in g b
with Not_found -> h a

Here, only exceptions raised by "f a" are caught by the try block, and
"g b" occurs in a recursive context.

Hope this helps,
Chris

[1] http://martin.jambon.free.fr/extend-ocaml-syntax.html#lettry

Re: [Caml-list] try and tail call

[ketty]

[-- Attachment #1: Type: text/plain, Size: 353 bytes --]

Hi,

On 10/21/06, Christophe Raffalli <raffalli@univ-savoie.fr> wrote:
>
>
>   try
>     ...
>   end
>     ... (* the exception raised here are not handles by with but
> propagated *)
>   with
>     ...


You could do it like this:

exception Propagate of exn

try
  ...
  try ...
  with x -> raise (Propagate x)
with
  | Propagate x -> raise x
  | ...

[-- Attachment #2: Type: text/html, Size: 804 bytes --]

Re: [Caml-list] try and tail call

[Christophe Raffalli]

>
> You could do it like this:
>
> exception Propagate of exn
>
> try
>   ...
>   try ...
>   with x -> raise (Propagate x)
> with
>   | Propagate x -> raise x
>   | ...
>
does not work, calls in the second ... are not tail call. (no tail call 
in a try)

Christophe

Re: [Caml-list] try and tail call

[Christophe Raffalli]

Chris King a écrit :
> On 10/21/06, Christophe Raffalli <raffalli@univ-savoie.fr> wrote:
>>
>> consider this piece of code:
>>
>>   (try
>>     let b = f a in
>>     (fun () -> g b)
>>   with
>>     (fun () -> h a)) ()
>>
>>
>> Is the call to "g" a tail call and is it omptimized as such ?
>
> Yes, it is, at least in the assembly code output by ocamlopt.  (Neat
> trick, btw.)
>
OK, but when g b is in fact a piece of code using a lot of values, it 
seems that the call is no more tail call and
the stack keeps far to much values ... anyway, I am rewriting my code in 
a completely different (and better) way to avoid the problem ...
but it would be nice to know exactly what call are tailcalls and what 
value are stored in closure in all cases.

Christophe