Assigning class pointers results access violation in debug mode with 11.1.051

[thomas_boehme]

Unfortunately, we are experiencing a problem with the newly released IVF 11.1.051.
It seems like class pointers no longer work correctly. Any attempt to associate a single class pointer results in an access violation in debug mode (See TypePtr => BaseInstance statement in code). This does not occur when the class pointer is wrapped in a type (see ClassPtr%Ptr => BaseInstance).

Unfortunately, this bug makes it impossible for us to use the new version, as it also prohibits the use of select type statements (see code below).

The problem doesonly occur in debug mode, release modeworks fine. Further, it did notoccur withv11.0.46.

Best regards,
Thomas

[plain]! SelectTypeBug.f90 !**************************************************************************** ! ! PROGRAM: ClassPtrBug ! ! PURPOSE: Demonstrate access violations with class pointers in debug with IVF 11.1.051 ! !**************************************************************************** program ClassPtrBug implicit none ! Variables TYPE :: BaseType INTEGER :: val END TYPE TYPE, EXTENDS (BaseType) :: ExtType INTEGER :: val2 END TYPE TYPE :: PtrType CLASS (BaseType), POINTER :: Ptr END TYPE TYPE (BaseType), TARGET :: BaseInstance TYPE (ExtType), TARGET :: ExtInstance CLASS (BaseType), POINTER :: TypePtr Type (PtrType) :: ClassPtr ! Body of ClassPtrBug TypePtr => BaseInstance ! This line causes an access violation in debug mode ClassPtr%Ptr => BaseInstance ! This line works ok SELECT TYPE (Instance => ClassPtr%Ptr) ! This select type also causes access violation in debug mode TYPE IS (BaseType) WRITE (*,*) 'This is the base type' TYPE IS (ExtType) WRITE (*,*) 'This is the base type' END SELECT end program ClassPtrBug [/plain]

[Steven_L_Intel1]

Thanks, we'll look into this.

[Steven_L_Intel1 · Accepted solution]

Escalated as issue DPD200141581.

The error is triggered by having /gen-interfaces enabled. (Note that /warn:interface implies /gen-interfaces). So as a workaround, turn off both Generate Interface Blocks and Check Routine Interfaces on the Diagnostics property page.

[haifeng2009]

I am curious what are these statements

"CLASS ..."
"TYPE, EXTENDS ..."
"SELECT TYPE..."
"TYPE IS ..."

Are they Fortran (standard)? or some kind of extension? Thanks.

[Steven_L_Intel1]

Fortran 2003 Standard

[rogcar]

Hi,

There is another issue related to this. Let's say the integers received any valid value. If you try to printon the screen using the working pointer (ClassPtr%Ptr%val for the base class), only zero is printed,no matter ifyouare using debug or release mode. Notice that in debug mode, if you try to check their values using the watch window,the pointer shows the correct value.

Changing Generate Interface Blocks and Check Routine Interfaces on the Diagnostics property page did not solve this problem.

Regards,
Roger

[Steven_L_Intel1]

Roger, we'd appreciate a test case for this. Thanks.

[rogcar]

Hello, Steve

Check the attached file. Its a quite simple example. It uses some F2003 features. One thing that I noticeis that the problem is related not to the pointer, but to the function that returns values.

Thanks for your help
Roger

[Steven_L_Intel1]

What are the compiler options used? Copy and paste from the Command Line tab or attach a .vfproj file.

[rogcar]

Quoting - Steve Lionel (Intel)

What are the compiler options used? Copy and paste from the Command Line tab or attach a .vfproj file.

.vfproj file attached.

Thanks
Roger

[Steven_L_Intel1]

This was fixed in Composer XE 2011.