Closed dnadlinger closed 8 years ago
I suppose the internal/convert.d unittest fix should go upstream first. Tested on X86 and ARM.
// don't test pad bytes because can be anything
enum testsize =
(FloatTraits!TYPE.EXPONENT + FloatTraits!TYPE.MANTISSA + 1)/8;
assert(rtbytes[0..testsize] == ctbytes[0..testsize]);
If questioning if fixing test is right thing, note that clang doesn't zero pad x87 long doubles either. The last 6-bytes are whatever happens to already be in memory. Makes sense because eliminates extra instructions needed to zero bytes that don't have any affect on long double computations.
See #770.