Memory (de)allocation again

[Richard E Maine]

On May 18, 2006, at 2:16 AM, François-Xavier Coudert wrote:

> In one of my runs, the ASSOCIATED returns .true., and the DEALLOCATE
> statement then exits with istat /= 0. I think that's a sign of a bug
> in gfortran, but I'm not sure: is that information enough to conclude?

That is absolutely not enough information to conclude such a thing.  
Could be a bug in gfortran, but that is certainly nowhere near  
sufficient information  to know (and I didn't study the provided  
debug data as it doesn't mean much to me without research). Indeed,  
my first reaction is that it sounds typical of a quite common user  
coding error.

When users ask this question (ok, not about the same program, but the  
same general situation), it tends to make me extra suspicious of user- 
code bugs, because to me the question illustrates that the user has  
forgotten about (or never knew) a critical and fundamental issue  
relating to pointers. If they have forgotten about that issue, odds  
of their pointer-using code having bugs seems high.

The issue is, of course, that pointer association status can be  
undefined and that it is illegal to even use the associated intrinsic  
on a pointer with undefined association status. If code does that, it  
is quite typical for the associated intrinsic to return bogus  
results. Symptoms like getting a true value from associated, but then  
being unable to deallocate or reference the pointer are very typical;  
yes, I've seen variants of that question a lot.

Another possibility is that memory has gotten corrupted from any of  
the usual suspects (exceeding dimension limits, argument list  
mismatches, errors in using pointers, etc.).

-- 
Richard Maine                |  Good judgment comes from experience;
Richard.Maine@nasa.gov       |  experience comes from bad judgment.
                             |        -- Mark Twain