advance = 'no'. Error: A non-advancing READ immediately following a non-advancing WRITE

[riad_h_1]

Dear all,

since the last update of ifort, I get the runtime error: "A non-advancing READ immediately following a non-advancing WRITE on the same unit number is not allowed, unit 5". 

I show below a simplified version of the part where the error is produced (this sample works well with gfortran and nagfor). I need to use advance='no' in the read statement since the length of the record is not known. 

The error message is somewhat ambiguous and surprising since the units are not the same.

In this example, the error is reported at the 2nd iteration. 

  use, intrinsic :: iso_fortran_env, only:  uin=>input_unit, uout=>output_unit, uerr=>error_unit, iostat_eor
  
  implicit none
  integer, parameter               :: lstr = 1000, mstr = 100
  integer                          :: iostat, size, i
  character(len=lstr)              :: iomsg
  character(len=mstr)              :: buf
  character(len=:   ), allocatable :: line

  do i = 1, 10

     ! print the prompt:
     write(uout, '(a)', advance='no') "--> "

     ! read until end-of-record:
     line = ''
     do
        read(uin, '(a)', advance='no', iostat=iostat, iomsg=iomsg, size=size) buf

        if (iostat > 0 .or. (iostat < 0 .and. iostat /= iostat_eor)) then
           write(uerr, '(a)') trim(iomsg)
           stop
        end if

        line = line // buf(:size)
        if (iostat == iostat_eor) exit
     end do

     write(uout,'(a)')'line = '//trim(line)
  end do

end

[riad_h_1]

Hi  Ferdinand and thank you for your comment and example.

Yes, I indeed forgot  to put the flush statement in the sample but is well present in my real code. Unfortunately (and even with your example) it crashes with the aforementioned message when compiling with ifort 19.1 (for mac) but it doesn't when compiling with the previous version (19.05).

Best regards

[Ferdinand_T_]

Maybe this issue is caused by changes in how the fortran runtime interacts with buffered terminal in/output?

Consider this (which crashes on the same error):

read(5, *)
write(6, '(a)', advance='no') 'x'  !; flush(6)
read(5, '(a)', advance='no')
end

But if you comment in the flush statement, it works.

So as a workaround, you could try to flush(uout) in line 14.
 

EDIT (in response to #3) ---------------------------------------------
Results for the above (short) example, generated on Red Hat Enterprise Linux Server release 7:

VERSION     NO FLUSH  WITH FLUSH
19.0.3.199  crash     ok
18.0.5.274  crash     ok
17.0.6.256  crash     crash
16.0.4.258  ok        ok

[Ferdinand_T_]

Hi Riad,

I had tested on linux, see my updated post (#2).

Alternatively, would if work if you simply ignored the error? On my setup I got iostat=126, so how about adding line 19 as follows:

if (iostat == 126) cycle

Of course, this does neither fix nor explain the origin of the error...
Best, Ferdinand

[riad_h_1]

I will adopt your solution (but IOSTAT values are certainly not standard, most likely compiler dependent) and, in the meantime, I will report the problem to the Online Service Center.

Thanks Ferdinand

Riad

[andrew_4619]

If you use ISO_FORTRAN_ENV Module then you have  constants such as IOSTAT_END or IOSTAT_EOR which is standard Fortran