How to issue system call from running fortran program

swerdna · 12-26-2006, 12:35 AM

Hi, I'm using gfortran 4.1 on Suse 10.1 to compile/run a fortran 90/95 program. The program loops iteratively towards the solution to a model. I need to display graphs at each iteration to make sure it's not calculating junk. I figure I can display these if the program can issue a sytem call to invoke one of the plotting apps that come with Suse (e.g. gnuplot, xmgrace). Google tells me to use the fortran intrinsic "system(command)" where "command" is a character variable containing the desired command.

My problem is that gfortran doesn't recognise "call system(command)" as a valid statement.

Here's the extracted lines of code:

Code:

character :: command
command = "xmgrace data/reconstituted.out"
call system(command)

and here's the compile-time dialogue showing the rejection of the code

Code:

john@suse101:~/Desktop/model> gfortran -std=gnu -W -Wall -pedantic model.f90
 In file model.f90:97

call system(command)
                   1
Warning: Intrinsic 'system' at (1) is not included in the selected standard
john@suse101:~/Desktop/model>

If I change the standard from gnu to f95 I still get a rejection as follows:

Code:

john@suse101:~/Desktop/model> gfortran -std=f95 -W -Wall -pedantic model.f90
/tmp/ccIk3JRP.o: In function `MAIN__':
model.f90:(.text+0xebc): undefined reference to `system_'
collect2: ld returned 1 exit status

I believe that "system(command)" used to work in fortran77, but I'm not keen to go back to that to get system calls.

Any advice on what I'm doing wrong, or on alternate methods to invoke graphics, would be much appreciated.

Thanks
Swerdna

swerdna · 12-26-2006, 08:51 AM

Hi folks, I got it solved:

Code:

john@suse101:~/Desktop/model> gfortran -std=f95 -W -Wall -pedantic model.f90

is no good because std=f95 forces compliance with f95, but calling "system" isn't supported [or just won't work for me] via the gfortran/f95 variant.

On the other hand using -std=gnu instead of -std=f95 does allow calling "system" because it was recently implemented in gfortran [and so you'll need ver 4.1+ of gfortran]. But you'll still get the warning message because of including "-pedantic" in the options. Despite this warning the executable works fine.

Thanks
Swerdna

sundialsvcs · 12-26-2006, 09:42 AM

If the code that uses this non-standard feature can be isolated, it is my understanding that code compiled to different standards can still be linked into the same executable... you are still using the same compiler.

In other words, if the standards-level "matters" to your code, I believe that you can spin-off those routines into separate compile-units so that the majority of the code is enforced to the higher standard. If that matters to your project.

swerdna · 12-27-2006, 07:44 AM

Quote:

Originally Posted by sundialsvcs

If the code that uses this non-standard feature can be isolated, it is my understanding that code compiled to different standards can still be linked into the same executable... you are still using the same compiler.

In other words, if the standards-level "matters" to your code, I believe that you can spin-off those routines into separate compile-units so that the majority of the code is enforced to the higher standard. If that matters to your project.

Thanks Sundialsvcs. It makes sense to separate the system-calls part of the work because that's tied up with a plotting package and is really a separate utility to generate plots from within programs generally. So I can and I will relocate that as a general-purpose utility.
Cheers
Swerdna