different records, same field name?

different records, same field name?

[mfmorss]

I have not used OCaml, just done some reading about it and toyed with the 
toplevel to see if it might be a useful tool here.  I have "Practical 
OCaml," which unfortunately is a rather wretched reference.

In any case, is it possible to define and use different types of records, 
which nevertheless share some field names?  I have had difficulty doing 
this in the toplevel.

Re: [Caml-list] different records, same field name?

[Jon Harrop]

On Monday 03 March 2008 16:51:57 mfmorss@aep.com wrote:
> I have not used OCaml, just done some reading about it and toyed with the
> toplevel to see if it might be a useful tool here.  I have "Practical
> OCaml," which unfortunately is a rather wretched reference.
>
> In any case, is it possible to define and use different types of records,
> which nevertheless share some field names?  I have had difficulty doing
> this in the toplevel.

Field names will shadow others so you define the record types in separate 
modules:

  module Vec2 = struct
    type t = {x: float; y: float}
  end

  module Vec3 = struct
    type t = {x: float; y: float; z: float}
  end

  { Vec2.x = 2.; y = 3. }

-- 
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
http://www.ffconsultancy.com/products/?e

Re: [Caml-list] different records, same field name?

[Basile STARYNKEVITCH]

mfmorss@aep.com wrote:
> 
> In any case, is it possible to define and use different types of records, 
> which nevertheless share some field names?  I have had difficulty doing 
> this in the toplevel.

It is not possible, for a simple reason:

Suppose you have two record type tr1 & tr2 sharing a common field name f

   type tr1 = { f : int; f1 : string }

   type tr2 = { f2 : float;  f: int }

Then the compiler won't be able to infer the type of function ff 
retrieving this field f

   let ff x = x.f

So the requirement that each field name is different is the price to pay 
to have type inference.

Regards
-- 
Basile STARYNKEVITCH         http://starynkevitch.net/Basile/
email: basile<at>starynkevitch<dot>net mobile: +33 6 8501 2359
8, rue de la Faiencerie, 92340 Bourg La Reine, France
*** opinions {are only mines, sont seulement les miennes} ***

Re: [Caml-list] different records, same field name?

[DooMeeR]

mfmorss@aep.com a écrit :
> I have not used OCaml, just done some reading about it and toyed with the 
> toplevel to see if it might be a useful tool here.  I have "Practical 
> OCaml," which unfortunately is a rather wretched reference.
> 
> In any case, is it possible to define and use different types of records, 
> which nevertheless share some field names?  I have had difficulty doing 
> this in the toplevel.

Sharing names for record labels, as well as for variant tags, is not a 
good idea. Indeed, only the last name is remembered (as for everything 
in OCaml), so the type system will infer the last type which has this label.

One way to separate names is to use modules, for instance:

module People = struct
   type t = {
     name: string;
     age: int;
   }
end;;

Then you can use labels People.name and People.age, for instance:

let x = {
   People.name = "x";
   age = 69;
};;

Notice how, when defining this variant, you don't have to use "People." 
for every label.

x.People.name

The toplevel gives you:
- : string = "x"

If you know you won't have any name clash, you can open module People 
and you won't have to use "People." everytime:

open People;;

let x = {
   name = "x";
   age = 69;
};;

-- 
Romain Bardou

Re: [Caml-list] different records, same field name?

[Kuba Ober]

On Monday 03 March 2008, Basile STARYNKEVITCH wrote:
> mfmorss@aep.com wrote:
> > In any case, is it possible to define and use different types of records,
> > which nevertheless share some field names?  I have had difficulty doing
> > this in the toplevel.
>
> It is not possible, for a simple reason:
>
> Suppose you have two record type tr1 & tr2 sharing a common field name f
>
>    type tr1 = { f : int; f1 : string }
>
>    type tr2 = { f2 : float;  f: int }
>
> Then the compiler won't be able to infer the type of function ff
> retrieving this field f
>
>    let ff x = x.f
>
> So the requirement that each field name is different is the price to pay
> to have type inference.

I think that type inference in OCaml sometimes gets in the way of other 
language features. I'd much rather see OCaml go in a direction where type 
inference is performed on "as far as it can go" basis.

In the cited example, if the compiler couldn't infer the type, it should 
simply remind the user to use explicit type as appropriate in such case.

While I do admit that this may cause some "silly" behavior, where changes in 
unrelated code can cause compile errors elsewhere, this should be naturally 
shielded by module boundaries. So, as far as I can see, it wouldn't suddenly 
break precompiled modules, at least.

Cheers, Kuba

Re: [Caml-list] different records, same field name?

[Michael Wohlwend]

Am Montag 03 März 2008 20:32:02 schrieb Kuba Ober:
>
> I think that type inference in OCaml sometimes gets in the way of other
> language features. I'd much rather see OCaml go in a direction where type
> inference is performed on "as far as it can go" basis.

I'm also the opinion, that type inference should be a feature. In the case of 
the record labels, if the compiler cannot decide the type of a record, an 
error could be issued to let the user manually fix it. The duplicate record 
label thingy arises regulary on the list, but I think it will stay a feature 
of ocaml :-)

 Michael

Re: [Caml-list] different records, same field name?

[Jon Harrop]

On Monday 03 March 2008 19:32:02 Kuba Ober wrote:
> I think that type inference in OCaml sometimes gets in the way of other
> language features. I'd much rather see OCaml go in a direction where type
> inference is performed on "as far as it can go" basis.

Other languages already do that, of course, and it just brings a different set 
of trade-offs. OCaml is roughly the best you can get in this respect.

Breaking decideability seems like a bad idea because you end up annotating all 
types and benefitting much less from type inference, as is idiomatic in 
Haskell and imposed in Scala.

The alternative of flagging an error when an ambiguity arises is used by F# 
but it makes no attempt to search current definitions for structurally- 
compatible classes. Indeed, I already find it confusing and a little ugly to 
have to add type annotations as you develop because your final product often 
ends up with unnecessary annotations as later developments remove 
ambiguities.

-- 
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
http://www.ffconsultancy.com/products/?e

Re: [Caml-list] different records, same field name?

[Berke Durak]

DooMeeR a écrit :
> mfmorss@aep.com a écrit :
>> I have not used OCaml, just done some reading about it and toyed with 
>> the toplevel to see if it might be a useful tool here.  I have 
>> "Practical OCaml," which unfortunately is a rather wretched reference.
>>
>> In any case, is it possible to define and use different types of 
>> records, which nevertheless share some field names?  I have had 
>> difficulty doing this in the toplevel.
> 
> Sharing names for record labels, as well as for variant tags, is not a 
> good idea. Indeed, only the last name is remembered (as for everything 
> in OCaml), so the type system will infer the last type which has this 
> label.
> 
> One way to separate names is to use modules, for instance:
> 
> module People = struct
>   type t = {
>     name: string;
>     age: int;
>   }
> end;;
> 
> Then you can use labels People.name and People.age, for instance:
> 
> let x = {
>   People.name = "x";
>   age = 69;
> };;
> 
> Notice how, when defining this variant, you don't have to use "People." 
> for every label.
> 
> x.People.name
> 
> The toplevel gives you:
> - : string = "x"
> 
> If you know you won't have any name clash, you can open module People 
> and you won't have to use "People." everytime:
> 
> open People;;
> 
> let x = {
>   name = "x";
>   age = 69;
> };;
> 

That's one instance were Alain's openin extension shows its usefulness.

   http://alain.frisch.fr/soft.html#openin

let sprint_time () t =
   open Unix in
   sprintf "%04d-%02d-%02d" t.tm_year t.(tm_mon+1) t.tm_day

Here's one thing we could do as part of the OSR.  Define a list of
"standard" syntax extensions and libraries.

I would certainly vote for openin to be included by default.

Then let the fine Debian Ocaml maintainers write a wrapper scripts
around the ocaml compiler that will use those extensions and have those
libraries accessible by default; name the resulting binaries in a consistent way,
such as by appending an "s" (for "standard") at the end of command names,
as in "ocamlcs", "ocamlopts".  Put those in a package called "ocaml-osr".

Tell people new to Ocaml to use those instead;  the existing users will be told
to just add an "s".

ocamlcs, ocamlopts would expand to

   ocamlc -I +pcre -I +lablgtk -I +extlib -I +whatever... -pp camlp4o -pp pa_openin ..

possibly depending on installed packages.
--
Berke DURAK

Re: [Caml-list] different records, same field name?

[David Teller]

On Tue, 2008-03-04 at 10:52 +0100, Berke Durak wrote:
> Here's one thing we could do as part of the OSR.  Define a list of
> "standard" syntax extensions and libraries.
> 
> I would certainly vote for openin to be included by default.

Agreed on both accounts. Do you wish to launch a thread on that
subject ?


Cheers,
 David
-- 
David Teller
 Security of Distributed Systems
  http://www.univ-orleans.fr/lifo/Members/David.Teller
 Angry researcher: French Universities need reforms, but the LRU act brings liquidations.