Thanks: that was exactly what I was looking for.

> The reason is simple enough: variants in ocaml are not based on Remy's
> rows, but on a type system with kinded variables, more in Ohori's 
> style.
> This is described in detail in "Simple type inference for structural
> polymorphism", which you can find at
>   http://wwwfun.kurims.kyoto-u.ac.jp/~garrigue/papers/
>
> The main reason for this choice is avoiding making rows explicit,
> i.e. having a multi-sorted type algebra. Note also that ocaml object
> types, while originally based on Remy's system, are hiding their
> row-variables, and can be described in the same formalism.
Just to make it crystal clear for me: You say that the "main reason for 
this choice is avoiding making rows explicit", does this mean that the 
O'Caml team feels that row-based type information is too complicated 
for an average user since you either steer away of hide it in an object 
model?

> Pragmatic reasons for this choice are given in Section 3 of
>   "Typing deep pattern-matching in presence of polymorphic variants"
> which you can find at the same place.

Sincerely,
Gilles Dubochet.