COMPILER BUG: Passing components of complex arrays to Fortran intrinsic procedures

[DataScientist]

What is wrong with this Fortran program?

complex :: z(5)
call random_number(z%re)
write(*,*) z
end

Intel ifort classic 2021 throws the following error message:

error #6672: This intrinsic procedure or system subprogram must not have an expression or constant as an argument. [RANDOM_NUMBER]

For comparison, gfortran compiles and runs flawlessly. Is this an ifort compiler bug? Generally speaking, what is the easiest way of assigning real/imaginary components of complex numbers separately?

[Steve_Lionel]

Yes, that's an ifort bug. There have been others related to %re and %im in the past. 

[DataScientist]

Is there a changelog for the newer compiler releases to see if this bug has been fixed in ifort 2022? Thanks.

[DataScientist]

Would be also great if Intel provided a direct link to the bug report page in the main page of this forum.

[Steve_Lionel]

There haven't been bugfix change logs for years. Only people with paid support can file an official report. But the Intel support engineers often pick up issues from this forum.

[DataScientist]

That's unfortunate. I changed the title to make it more visible.

[Steve_Lionel]

I filed bug report 05244368 on your behalf. I reported a similar issue with %re and %im in a I/O list about a year and a half ago.

To answer your other question, the complex-part-designator (%re or %im) is the easiest way of doing this - that's what it's for.

[DataScientist]

I know this is not your responsibility anymore. So thank you very much for taking the time to report it.

[Steve_Lionel · Accepted solution]

This is fixed in ifort 2021.6 and ifx 2022.1 (oneAPI HPC Toolkit 2022.2 - darned confusing.)

[DataScientist]

fantastic, thank you for the update.