[clisp-list] verify FFI-code, plz

[<Joe...@t-...>]

Yaroslav Kavenchuk wrote:

>  (:return-type int))
>       (unless res
>It is right?
No. You didn't try it out, did you?
This is not C. What branch is (unless <integer>) going to take in Lisp?

>         (LLVMCreateJITCompiler provider 2 err)
>          (LLVMDisposeMessage err)
Compare with
>   if(LLVMCreateJITCompiler(&engine, provider, 2, &error) != 0) {
>    LLVMDisposeMessage(error);
Obviously you cannot use the same object in Lisp. One time you need
a pointer to pointer-sized slot, another time the value of that slot.

>Or 'LLVMCreateJITCompiler' will be called with a value of
>'err', rather than its address?
I find "value of err" is confusing in this sentence.
err "represents" the object you asked to allocate on the stack,
You can view it as a pointer to this object.
The Lisp side likes to ignore pointers and consider the value that object holds.
The C sides values the pointer itself.
So it's suitable for the &error call.

Note that this is not the only way to interface to these functions.
I wrote:
>I'd say you're better off using an explicit c-pointer for [out]error
Perhaps you'd get less confusion by defining
LLVMCreateJITCompiler with (OutErr (c-ptr c-pointer) :out)
as you initially wanted. You'll be returned a pointer that you can pass
back or interpret as a string with memory-as.
Beware: This pointer is not the same as the above err object!
It does not represent a pointer to a pointer, rather than the string data itself.

(memory-as error 'string) ; not ffi:c-string!
This is a special pattern that reads a string from memory, not a pointer to a string.

It appears that this is not documented in the impnotes.
Strange... That is an oversight.

Regards,
 Jörg Höhle