I declared my variables as unsigned long long int and I have also changed the return type to this. It worked.

You've got a typo in the first two math equations, instead of r variable, b should be used.

I just tried it, there are no issues.

it is IMPOSSIBLE to have 4000000000 as answer

It's a wrong assumption that long is always 32 bits. long is not less than int and not less than 32 bits. In 64-bit Linux it's 64 bits.

I tried to change the function to unsigned long but I get the same error.

You are probably converting to unsigned long when it's too late and the result is already truncated (at best; or even more likely undefined).

I tried to change the function to unsigned long but I get the same error.

And what's the message now? VS C++ uses 32 bit, but here it's 64 bit... and there are better ways to force 64 bits on different platforms? Many comments for a small solution of a simple kata. Fact is, there's no overflow here and no VS C++;-)!

long is platform dependant (has 4 bytes on 32 bit platforms, then you have to use long long or __int64 (eg. look here https://msdn.microsoft.com/en-us/en-en/library/s3f49ktz.aspx)).

The most correct way of doing this is to use stdint.h header (from C99) and then use int64_t which is platform independant and is always 64bit.

ExtendedTest3
✘ Expected: equal to 4000000000
Actual: -294967296

C++ version has prototype: long sumTwoSmallestNumbers(std::vector n), it is IMPOSSIBLE to have 4000000000 as answer

It's a "long" which has 8 Byte or 64 Bit, try sizeof(m) which is 8... so there's no problem, because all the others are only int.

It is rather bad as 9999999999 overflows (0x2_540B_E3FF) and 32 bit architectures use 0x7FFFFFFF as max for signed int.