The R language has some hacks specifically for mingw-w64 that
were caused by our handling of NaNs in sqrt(x). R uses a
special valued NaN to mean 'Not Available' and expects it to
be retained through various calculations. Our sqrt(x) doesn't
do this, instead it normalises such a NaN (retaining sign).
From:
http://wwwf.imperial.ac.uk/~drmii/M3SC_2016/IEEE_2008_4610935.pdf
"6.2.3 NaN propagation
An operation that propagates a NaN operand to its result and
has a single NaN as an input should produce a NaN with the
payload of the input NaN if representable in the destination
format."
There might even be a slight speed-up from this too.
Thanks to Duncan Murdoch for finding the reference.
96fc528397