Nala Ginrut <nalagin...@gmail.com> writes:
> On Tue, 2016-01-05 at 08:49 +0100, to...@tuxteam.de wrote:
>> Sorry for intervening from the peanut gallery, but if I understood Mark
>> correctly, he only was proposing to introduce a second function for the
>> C API (to keep backward compatibility at the linking-to-C level). At the
>> Guile source level, I guess all can be subsumed under one function.
Yes, Tomás is correct. For the Scheme API, I would prefer to keep just
one procedure 'pointer->procedure' that accepts an optional
#:return-errno? keyword argument, as we had previously agreed.
> If we want to combine them in Scheme level, we have to change the name
> "pointer->procedure" in C level, since it's registered with SCM_DEFINE.
That's right, the C function name in the SCM_DEFINE construct would need
to have a different name and be private, perhaps something like
'scm_i_pointer_to_procedure_with_keywords'.
> Dunno if it breaks the ABI too.
As long as there still exists a 'scm_pointer_to_procedure' function with
the same signature and semantics as it has now, that will preserve ABI
compatibility.
More specifically, here's what I'd suggest:
* A new, static, 'pointer_to_procedure' C function that inherits the
signature and body of 'scm_pointer_to_procedure' but with a new and
required 'return_errno' argument. The other functions below would be
wrappers for this function.
* A new private 'scm_i_pointer_to_procedure_with_keywords' C function,
defined using SCM_DEFINE and bound to 'pointer->procedure' in Scheme,
that uses 'scm_c_bind_keyword_arguments' and calls
'pointer_to_procedure'.
* The C API function 'scm_pointer_to_procedure', which has the same
arguments as in Guile 2.0.11 and calls 'pointer_to_procedure' with
'return_errno' set to SCM_BOOL_F.
* A new C API function 'scm_pointer_to_procedure_with_errno', which is
identical to 'scm_pointer_to_procedure' except that it calls
'pointer_to_procedure' with 'return_errno' set to SCM_BOOL_T.
The only change to foreign.h would be to add a prototype for
'scm_pointer_to_procedure_with_errno'.
What do you think?
Regards,
Mark
