Segmentation fault (core dumped) in a GFortran subroutine

AlexBB · 12-20-2014, 03:44 PM

Hi there. I am trying to adjust to my particular requirements a Fast Fourier Transform package I found on the web. It works but I need to do some modifications. So, at one moment I place a few lines of code into a subroutine. I need to dump an array into a file to examine it subsequently with gnuplot. It is a very simple array with 2310 members. This is the code:

Code:

10 FORMAT (E15.5)
  open(unit=2,file="fft_1.dat",status="old")
  IOstatus = 0
  counter = 0
  do while (IOstatus.eq.0)
    counter = counter + 1
    if (MOD(counter,2).ne.0) then
      write (2,10) r (counter)
    endif
  enddo
  close (unit=2)

What happens is the program breaks down. There are about 2600 lines of code and it is not immediately apparent at what point. If I comment the above lines out, everything works fine.

There are two ALLOCATE and two DEALLOCATE statements in that routine. I am sure if I rewrite the subroutine without them everything will work with and without this code block.

Why do I have this problem? Thanks, A.

suicidaleggroll · 12-22-2014, 08:32 AM

You've written an infinite loop. It shouldn't seg fault, but it should never end either. Clearly there's more at play than the small code chunk you've posted here.

99% of seg faults in fortran can be found by adding the -fbounds-check flag to your gfortran compilation.

btmiller · 12-22-2014, 11:38 AM

Probably the infinite loop causes the counter variable to keep incrementing until such time it is larger than the length of the array "r", at which point the program will begin reading garbage until it tries to access an invalid memory address, which will cause a segfault generally.

suicidaleggroll · 12-22-2014, 11:45 AM

Quote:

Originally Posted by btmiller

Probably the infinite loop causes the counter variable to keep incrementing until such time it is larger than the length of the array "r", at which point the program will begin reading garbage until it tries to access an invalid memory address, which will cause a segfault generally.

For some reason I read that write as "r, counter", rather than "r(counter)", probably due to the oddly-placed space.

Yes, the infinite loop will eventually exceed the limits of r and dive into the ether, ultimately causing a seg fault.

AlexBB · 12-22-2014, 11:51 AM

I made the space between the r and "(" myself, manually. You are correct: I should have put IOstatus into the read statement. O' my!!! Thank you.

JeremyBoden · 12-22-2014, 07:58 PM

It might be easier to just print as separate lines and re-direct the printer to file output.