Excerpts from Hendrik Tews's message of Fri Apr 04 16:01:57 +0200 2008:
> "Nicolas Pouillard" <nicolas.pouillard@gmail.com> writes:
> 
>    > Can I do the same with a syntax extention that is only
>    > loaded at runtime via Dynlink? Assume a syntax extention X
>    > that is only available at runtime and not at compiletime.
>    > After loading X with Dynlink, all I can do is invoke the
>    > callback of X, but this will add the grammar extension of X
>    > to Camlp4.PreCast.Syntax.
> 
>    You can do it but you will to resort to some kind of
>    registering of your extensions, as it's done in camlp4 for the
>    default grammar.
> 
>    > Is there a way to obtain two functions parse_implem (as
>    > above) that mix X with different syntaxes (eg Ocaml + X and
>    > Revised + X)? For that, I believe, it would be necessary to
>    > extract parse_implem from a syntax module such that it
>    > survives clearing that Syntax.
> 
>    Yes but your extension X has to be a functor, then you can get
>    X(OCamlOriginal), X(OCamlRevised).
> 
> But this only works if I can statically link against X. There is
> no way to access the functor X through the Dynlink interface.

Dynlink cannot access the X functor, but it can register itself.

> With the standard interface of Camlp4 syntax extensions the
> loaded module applies X only to PreCast.Syntax. This could be
> repeated but one would have to clear PreCast.Syntax in between.

You  can  either  reuse  the  current setup, by clearing the PreCast.Syntax in
between but also setup another registration functor.

> Changing the Camlp4 interface one could chieve that X is applied
> to two or maybe three times (to PreCast.Syntax1 ... Syntax3). But
> how about the general case, where X is applied to n Syntax
> modules and n depends on the input? I believe this is impossible.

I think that's doable (not tested), basically you need three primitive functors:

Register : (Syntax -> Syntax) -> Unit
Clear    : Unit -> Unit
Apply    : Syntax -> Syntax

-- 
Nicolas Pouillard aka Ertai