On Windows, long is a 32 bits type, so some tests must not be here
I'm going to give my two cents to this discussion.
So, seeing the EINA_VALUE_TYPE_LONG Documentation, it doesn't specify any data type size, so I think it relies on machine/compiler internal implementation, which is ok for C and C++, but starts raising issues when related to bindings (e.g. C# long). Not big issues, but they wouldn't exist if, instead of having long as a possibility, we just rely on the user to decide if he/she wants a signed or unsigned 32 or 64 bit integer using INT, UINT, INT64 and UINT64, which is way more accurate to his/her intention in having a specific sized integer. After all, we already don't have EINA_VALUE_TYPE_LONG_LONG, and instead we have EINA_VALUE_TYPE_INT64.
I you agree, I think that, as @vtorri stated, deprecating eina_value_type_long should be the way.