topic Re: Internal Error trying to compile with OneAPI 2025 in Intel® Fortran Compiler

Internal Error trying to compile with OneAPI 2025

JacobB — Thu, 20 Mar 2025 09:08:45 GMT

Hello

I am trying to build a software (WRF-HYDRO) with the 2025 version of the fortran compiler. One file io_manager.f90 (See attached files) causes an internal error.

In previous versions of OneAPI I used IFORT (instead of IFX) and it went through. Now IFORT is no more available in the new version.

Any help on how to solve the problem or find a workaround would really be appreciated. The compiler crashes when trying to allocate.

best

jac

Re: Internal Error trying to compile with OneAPI 2025

andrew_4619 — Thu, 20 Mar 2025 11:20:29 GMT

I would look at that but there are module dependencies (netcdf) that are not included. An example that is self contained would get more response if that is easy to do.

Re: Internal Error trying to compile with OneAPI 2025

braycarla — Fri, 21 Mar 2025 05:51:01 GMT

Ran into similar internal errors with OneAPI 2025 and it seems a bit unstable right now. Faced this while testing llamacpp too. It maybe worth checking compiler flags.

Re: Internal Error trying to compile with OneAPI 2025

JacobB — Fri, 21 Mar 2025 07:14:38 GMT

Sorry for that

in a more generic way, IFX issues an internal compilation error in the following case

if I declare an object

class(class_type), allocatable :: class_name

the compilation error is issued when I try to allocate

allocate(class_name)

best

jac

Re: Internal Error trying to compile with OneAPI 2025

JacobB — Fri, 21 Mar 2025 07:15:19 GMT

thanks

do you have a suggestion ?

best

jac

Re: Internal Error trying to compile with OneAPI 2025

JacobB — Fri, 21 Mar 2025 07:54:50 GMT

to complete the picture class_type is an abstract class

Re: Internal Error trying to compile with OneAPI 2025

Andrew_Smith — Fri, 21 Mar 2025 09:21:59 GMT

It is not legal to allocate an instance if an abstract type

Re: Internal Error trying to compile with OneAPI 2025

Steve_Lionel — Fri, 21 Mar 2025 20:29:14 GMT

If the usage is invalid (and the one Andrew mention is), an ICE is still a compiler bug.

Re: Internal Error trying to compile with OneAPI 2025

jimdempseyatthecove — Sat, 22 Mar 2025 18:14:58 GMT

>>It is not legal to allocate an instance if an abstract type

If the object you are trying to allocate is and abstract class, then your allocate statement will require a type specification:

ALLOCATE ([type::] object[(s-spec[, s-spec]...)] [, object[(s-spec[, s-spec]...)] ]... [[coarray-spec]]...[, alloc-opt[, alloc-opt]...])

type	Is a data type specifier. If specified, the kind type parameters of each object must be the same as the corresponding type parameter values, and each object must be type compatible with the specified type.

At least this is my read of the reference manual.

Jim Dempsey

Re: Internal Error trying to compile with OneAPI 2025

Steve_Lionel — Sat, 22 Mar 2025 18:30:23 GMT

Yes, in which case you are no longer allocating an instance of abstract type, as you must specify a non-abstract extension of the base type.

Re: Internal Error trying to compile with OneAPI 2025

JacobB — Sun, 23 Mar 2025 08:00:26 GMT

Thanks for the clarification

I proceeded as suggested and now it compiles

best

jac