First page Back Continue Last page Overview Graphics
Specific Choices
Keep trailing zeros.
Normalize after each calculation.
Special values: +0, -0, +INF, -INF, NaN.
Errors raise exceptions OR can set flags and allow the calculation to continue.
And Facundo will go into more detail in his talk following this one...
Notes:
So what were the specific design choices made? Most are dictated by Cowlishaw's spec, so I'll just point at a couple of big ones.
We keep trailing zeros, as many as were entered or generated by a calculation. Values are normalized after each step of a calculation, to a precision of 28 places (configurable). Several special values are defined including positive and negative zeros, infinities, NaN. Error handling is supported two ways: by default, exceptions are thrown, or you can instead set things so that the calculation proceeds but warning flags get set.