On Mon, May 28, 2012 at 11:33 PM, Liam Healy <lnp@healy.washington.dc.us> wrote:

On Sun, May 27, 2012 at 4:48 PM, Luís Oliveira <luismbo@gmail.com> wrote:

On Sun, May 27, 2012 at 9:15 PM, Luís Oliveira <luismbo@gmail.com> wrote:
> Did you notice the function lisp-string-to-foreign? Seems like a
> cleaner way to implement translate-into-foreign-memory.

More importantly, how can you be translating a string without
performing allocation in this case? Also, in general,
translate-into-foreign-memory for strings should definitely refuse to
accept string types without an explicit size. (Which is not
implemented yet, IIRC.)

In any case, why does Ryan's example need translate-into-foreign-memory?

Yeah, the new method is the wrong thing, I need to revert this change. When foreign-funcall-form is called and fsbvp=T, it calls translate-objects with indirect=T, which in turn calls #'expand-to-foreign-dyn-indirect on each of the arguments. When an argument is a foreign-string-type, the trouble begins. The foreign string is represented by a pointer, as indicated by the (:actual-type :pointer) in the define-foreign-type. All libffi really needs is a pointer to the pointer, and there is a method for expand-to-foreign-dyn-indirect for foreign-pointer-type, but foreign-string-type is not a subclass of foreign-pointer-type.

It seems like the resolution is to define a method for expand-to-foreign-dyn-indirect for the foreign-type-alias class that does a recursive call using the actual-type. Does that make sense? As a side point, expand-to-foreign-dyn has a similar method but it is defined for enhanced-typedef (subclass chain: enhanced-typedef < foreign-typedef < foreign-type-alias); is there any reason why that method shouldn't be defined for foreign-type-alias instead?

Liam