Regression with dev_null.f90

[Thomas Koenig]

Steve Kargl wrote:

> The program I sent was bogus because it did not use feof()
> to check for EOF.  This suggests that the implementation
> of READ(...,end=100) is not properly checking for EOF on
> at least the platforms we've identified.

We don't use C i/o like your example program, we use the Unix system
call read() if mmap() fails, as it is supposed to for /dev/null.

Can you open PR for this?