paul@cosc.canterbury.ac.nz (06/13/90)
The following program illustrates a bug in the system(3f) function In Sun
FORTRAN 1.0 (I think) on a Sun 4/330 running 4.0.3.
character*80 fred
read(*,fmt='(a)' ) fred
write(*,fmt='(a)') fred
status = system('date')
read(*,fmt='(a)' ) fred
write(*,fmt='(a)') fred
end
When executed with stdin connected to the terminal the program behaves as
expected - the first line is read and written, the date(1) output is
printed, then the second line is read and written. When input is
redirected to come from a file, however, the program dies at the second
read because end of file is true. The file that input is coming from does
have two lines though!!!!
Because the input file is file buffered (read in 8192 byte chunks), the
whole file is read by read(2) in respone to the first read of the program.
The first line is returned, and the second is buffered somewhere. When
the second read is done the buffered second line should be returned.
Instead a read(2) on the file is attempted, which of course returns EOF.
It seems that system(3f) is flushing the input buffer when it shouldn't
be. When input comes from the terminal there is no problem as the input
is line buffered. I tried an equivalent C program which worked as
expected.
Is this a known bug? Is there a patch? A workaround (could use fork and
exec - can the equivalent of a setbuf be done from Fortran?). A fixed
later version?