Potential regression in gfortran 4.3.3

[Mikael Morin]

Harald Anlauf wrote:
> Hi,
> 
> using Tobias' snapshots for x64_64, I am seeing a potential regression
> in gfortran 4.3.3 between revs. r142852 and r142864.  I am still working
> to reduce the problem to a manageable size which might take some time,
> so this is just a head-up.  Anyway, the problem appears to be related to
> passing arrays which are sections of pointer components of derived types.
> I am suddenly getting an array bounds violation from r142864, but not from
> previous versions (nor from other compilers).
> 
> The code in question is a subroutine call:
> 
>     call tq_tvgh (var_f% av (k_lev:,1,1),  ...)
> 
> where checks on var_f% av are positive:
> 
>  associated: T
>  shape     :          43           4           1
>  lbound    :           1           1           1
>  ubound    :          43           4           1
> 
> but I get:
> 
> At line 1167 of file ../../../../oo-model/mo_tovs.f90
> Fortran runtime error: Array reference out of bounds for array 'var_f', lower bound of dimension 2 exceeded(1 < 9221120237041090560)
> 
> Mabe this rings a bell.
> Mikael, any ideas whether your recent fix to PR38487 could be relevant here?
It is unlikely, but that of pr35983(r142860) could be related, even if
it passed the tests.
Does it happen with trunk as well?

> 
> Anyway, I'll work on a reduced example.
Thanks!

> 
> Cheers,
> Harald
>