On Nov 22, 2007, at 4:58 PM, Duncan McGregor wrote:
On 22 Nov 2007, at 21:01, Timothy Wall wrote:
The return value part of the cast ensures that the compiler
properly handles the value returned. Defining your JNA method
interface is equivalent to doing the cast; all type information is
passed to the libffi layer, which handles stack setup and capture
of the return value. Currently all returned structures are
assumed to be as pointers by JNA; you might write up some simple
tests with a custom shared lib to see if small structs are handled
properly as an argument and/or return value. libffi for powerpc
looks like it handles small structs, as does x86; JNA may need to
be tweaked a bit to pass structure size information down to
libffi, though.
So fighting my way down through libffi, I'd pass the instance of
the Structure in along with an ffi_type describing it, pass that to
ffi_prep_cif, let the invoke happen giving it the pointer to the
structure as the return value, as currently happens.
So I guess that libffi uses the contents of the ffi_type to work
out how to copy the value to the buffer?
I'm in so deep I can't even see the bottom any more.
That's basically it, if you want to do auto-magic structure-by-value.
Given that Structure results are currently return by pointer, how
would you signal that you wanted return by value? It would seem
logical to make that the default, but would break current code.
An easier workaround would be to simply replace the structure with an
int or long (depending on structure size) and provide "toInt/fromInt"
methods on the Structure definition for convenience.
9 messages in net.java.dev.jna.usersRe: [jna-users] Re: EXC_BAD_ACCESS Ca...
