On Mon, 11 Apr 2022 19:13:04 -0700
Hal Murray <halmurray@sonic.net> wrote:
I just pushed an scons change that should determine what
strerror_r() returns at configure time.
Works.
Good.
I have some hack code that tests the printf rounding.
I found the change between 26.1199 and 26.1200.
The rounding looks good to me.
float
26.119949340820312500 26.1199 41d0f5a8
26.119951248168945312 26.1200 41d0f5a9
So that step is the smallest possible in a float? If so, then that
is good.
double
26.119949999999999335 26.1199 403a1eb50b0f27bb
26.119950000000002888 26.1200 403a1eb50b0f27bc
That looks good.
Fred's theory is that the rounding error is happening before the
printf(). Somewhere. Probably related to implicit long doubles.