PR26890 SIZE parameter interacts with same variable in IO list.

[Jerry DeLisle]

Paul Thomas wrote:
> 
> This is, of course, why the construct is non-standard (incidentally, 
> have you checked my assertion of this?).
>

Your example works fine with ifc.  With gfortran the variable is set to zero in 
data_transfer_init
>>
>> The easy fix is to use another integer variable to accumulate the size 
>> and then assign it to the SIZE parameter as part of the 
>> finalize_transfer.
> 
> 
> I just suggested that on Bugzilla.
> 

Our mail is crossing the ocean simultaneously.

>>
>>
>> The problem is, to do this I have to add a variable to the dtp 
>> structure.  This will effect compatibility with previous versions of 
>> the library.
> 
> 
> Does it have to be an extra field on the dtp structure? Surely, a static 
> variable would do the job, wouldn't it?
> 

Using a static variable might lead to non thread safe behavior.  Thats the main 
reason for the dtp structure.  Incidentally, the internal_unit is a static 
structure and this is why we have pr26766.  I will be fixing that soon.

>>
>> Also, I notice that at the end of the dtp structure there is the 
>> following:
>>
>>       char pad[16 * sizeof (char *) + 34 * sizeof (int)];
>>
>> What is this for?  Can I add an integer and reduce the size of this 
>> pad accordingly?  This sure is not documented anywhere.  And I do not 
>> see it used anywhere. 
> 
> 
> If it is not used, it could be used for temporaries.  It's name imples 
> that it is being used for alignment purposes but the space allocated 
> belies that.
> 

I think I will try using this pad area and see what happens.

Jerry