Pitch: Method to sum numeric arrays

[scanon (Steve Canon)]

@Jens

Should the following trap (because the particular order causes an overflow) or do the best that it can and print 7:

let a: [Int8] = [127, 8, -128]
print(a.sum())

?

To expand on this a bit, there are multiple related questions lurking here, which effect both floating-point and integer:

• Is the accumulation order specified? Is vectorization (which implicitly re-associates) allowed? Is the re-association allowed to be different depending on platform or runtime?
• Is the accumulator width required to be the same as the element type? Can the compiler or runtime use a wider accumulator to prevent overflow? Is the compiler or runtime required to use a wider accumulator to prevent overflow?
• Is the runtime required to produce a result as though computed in a single operation without intermediate overflow or rounding?