Can't handle resultset with 128 bit int

[gijzelaerr]

keyerror_bug.zip

[ghost]

The solution is to consider mapping hge to Numpy object array

[gijzelaerr]

adding a 128 bit column definition to our cffi layer:

 typedef struct { monetdbe_types type; __int128 *data; size_t count; char *name; __int128 null_value; double scale; int (*is_null)(__int128 *value); } monetdbe_column_int128_t;

is not supported by cffi:

cffi.FFIError: Unsupported type: '__int128'.  Please look at http://cffi.readthedocs.io/en/latest/cdef.html#ffi-cdef-limitations and file an issue if you think this type should really be supported.

https://foss.heptapod.net/pypy/cffi/-/issues/457

[aris-koning]

Is it not possible to cast the array of int128 returned by monetdbe as a numpy structured array? In that case the array in memory is still properly aligned and in theory could directly be cased into an array of 128 bits integers if needed. E.g. if there is native implementation available on the user side to directly handle 128 bits. Otherwise the user can always access the individual high and low part of the huge integer through the structured array interface of numpy.

[gijzelaerr]

one of the issues is that cffi doesn't support int128 directly yet, so we first need to work around that to even get access to values.