Possible compiler bug

[sarai]

Hi,

while compiling a program I am working on, I get the following error:

"...catastrophic error: **Internal compiler error: segmentation violation signal raised**..."

The code compiles with both gfortran 9 and 10. My issue is that I cannot locate the cause of the error.

I am compiling with the latest ifort compiler available through oneAPI for mac. But I have also tried with previous versions of the compiler, with the same result.

I have 'reduced' the code to a minimal (nonsensical) example  that triggers the error message.  This example also compiles with the gnu compilers.

Compiler flags: -warn all -stand f18 -i8

Thanks!

[Steve_Lionel]

Internal compiler error is ALWAYS a compiler bug, regardless of the validity of the source.

[sarai]

Is there a way to identify what triggers the bug?

[andrew_4619]

There is no 'easy' way to find the cause. One strategy is to progressively add or remove the compiler options. The presence of some specific option is often a trigger  "-warn all" covers a number of option BTW.

The other option is to pair back the code (which it seems you have worked on) to create a minimalist reproducer.  This can be very time consuming.

[sarai]

Thanks for the suggestion! Unfortunately, even without any of the compler flags,  encounter the error.

[jimdempseyatthecove]

Then remove sections of code in a search pattern. Use !dir$ IF(.FALSE.) / !dir$ END IF to encapsulate code removed. e.g. the code part of the suspected procedure. When you locate the offending procedure, move the !dir$ IF(.FALSE.) down and/or  !dir$ END IF up while marking the former location(s) with comment. Eventually you will narrow in on the problem section of code.

Jim Dempsey

[Steve_Lionel]

The specific line that triggers the error is:

call output%printf()

in classA.F90. The Intel compiler developers would have to figure out why. I did note that if I commented out that line, the program still failed to build with other "normal" errors, probably caused by the code reduction - I have not investigated those.

[sarai]

Thanks! I also notice that moving the use  statement

   use global_out, only: output

to the routine that calls

  call output%printf()

also eliminates the issue. I have attached a new set of files that will compile when the offending line is commented out.