djsundog/circuitpython
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

unix: Fix modffi to be able to return double on x86 machines.

Browse Source
This commit is contained in:
Damien George 2014-09-06 23:04:42 +00:00
parent 03281b3850
commit 91fe0d4880
1 changed files with 13 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
unix/modffi.c
View File

@ -334,9 +334,19 @@ mp_obj_t ffifunc_call(mp_obj_t self_in, mp_uint_t n_args, mp_uint_t n_kw, const
valueptrs[i] = &values[i];
}
ffi_arg retval;
ffi_call(&self->cif, self->func, &retval, valueptrs);
return return_ffi_value(retval, self->rettype);
// If ffi_arg is not big enough to hold a double, then we must pass along a
// pointer to a memory location of the correct size.
// TODO check if this needs to be done for other types which don't fit into
// ffi_arg.
if (sizeof(ffi_arg) == 4 && self->rettype == 'd') {
double retval;
ffi_call(&self->cif, self->func, &retval, valueptrs);
return mp_obj_new_float(retval);
} else {
ffi_arg retval;
ffi_call(&self->cif, self->func, &retval, valueptrs);
return return_ffi_value(retval, self->rettype);
}
error:
nlr_raise(mp_obj_new_exception_msg(&mp_type_TypeError, "Don't know how to pass object to native function"));