Bizarre pointer behaviour with allocatable scalars in user defined types

[IanH]

I have some code which is behaving strangely when I set a pointer to point to an allocatable scalar that is a member of a user defined type.

When I run the attached short stretch of code I get different output from the WRITE statements in the AddElement subroutine, with only the first WRITE giving meaningful output. The confusing extract is

e%content(pos)%element%dbg = 'CL'
p => e%content(pos)%element
WRITE (*,"('1>',A)") e%content(pos)%element%dbg
WRITE (*,"('2>',A)") p%dbg

I would have thought that the two lines of output would have to be the same, even in the face of a gross error somewhere else.

The code works as I would expect if I remove one level of procedure call (set this_is_driving_me_crazy to .FALSE., as far as I can tell B to B is the same code as A to A), or change the allocatable scalar to be a rank one, size one array (lines marked "! C" replace similar lines).

What am I missing????

All this with 11.0.066.

Thanks for any ... pointers ...

IanH

[gib]

Quoting - IanH

I have some code which is behaving strangely when I set a pointer to point to an allocatable scalar that is a member of a user defined type.

When I run the attached short stretch of code I get different output from the WRITE statements in the AddElement subroutine, with only the first WRITE giving meaningful output. The confusing extract is

e%content(pos)%element%dbg = 'CL'
p => e%content(pos)%element
WRITE (*,"('1>',A)") e%content(pos)%element%dbg
WRITE (*,"('2>',A)") p%dbg

I would have thought that the two lines of output would have to be the same, even in the face of a gross error somewhere else.

The code works as I would expect if I remove one level of procedure call (set this_is_driving_me_crazy to .FALSE., as far as I can tell B to B is the same code as A to A), or change the allocatable scalar to be a rank one, size one array (lines marked "! C" replace similar lines).

What am I missing????

All this with 11.0.066.

Thanks for any ... pointers ...

IanH

It is interesting that ifort 10.1.011 cannot compile this program. The line
TYPE(Element), ALLOCATABLE :: element
gives
Error: This scalar name is invalid in this context [ELEMENT]
pointing to the member 'element'.

[Steven_L_Intel1]

Quoting - gib

It is interesting that ifort 10.1.011 cannot compile this program. The line
TYPE(Element), ALLOCATABLE :: element
gives
Error: This scalar name is invalid in this context [ELEMENT]
pointing to the member 'element'.

Expected - allocatable scalars is a new 11.0 feature.

[Steven_L_Intel1]

This is a compiler bug. It is not handling the p%dbg reference properly. I have reported it as issue DPD200111953.

[IanH]

Quoting - Steve Lionel (Intel)

This is a compiler bug. It is not handling the p%dbg reference properly. I have reported it as issue DPD200111953.

Thanks for the help. I'll work around it in the meantime by using a size one allocatable array.

[Steven_L_Intel1]

The developers tell me that this bug has been fixed for the next release (May/June timeframe). Thanks for reporting it.