This program
program test real a(10) open(1, file='f1',access='direct', recl=20) do i=1, 10 a(i) = i-1 end do write(1, rec=1)(a(i),i=1,5) write(1, rec=2)(a(i),i=6,10) close(1) open(1, file='f1',access='direct',recl=8) read(1, rec=4)(a(i),i =5,9,4) print*,a end
works incorrect in visual Fortran (incorrect):
0.0000000E+00 1.000000 2.000000 3.000000 9.000000 5.000000 6.000000 7.000000 0.0000000E+00 9.000000
Result in WATCOM (correct):
0.0000000 1.0000000 2.0000000 3.0000000 6.0000000 5.0000000 6.0000000 7.0000000 7.0000000 9.0000000
Why?
The default unit length of RECL in Visual Fortran is a word (4bytes). If you compile with option 'Use bytes as RECL= unit for unformatted file' (/assume:byterecl), you will get what you expect.
You should not need to know when writing the code how exactly are the record long in whatever units. Setting recl=20
or recl=5
and depending on exact behaviour of your compiler and exact real
size is a recipe for future problems.
You should ask the compiler for the size of the record in whichever units it uses:
integer :: recl5, recl2 inquire(iolength=recl5) (a(i),i=1,5) open(1, file='f1',access='direct', recl=recl5) ... inquire(iolength=recl2) (a(i),i =5,9,4) open(1, file='f1',access='direct',recl=recl2)
Note: In Fortran 90 it could be much shorter by using subarray notation.
Note2: For some compilers your program will never work as the compiler is allowed to use record end marks into the file and you have to always open it with the same recl
. But that is not common.