[patch, libgfortran] PR24224 Generalized internal array IO not implemented.

[Jerry DeLisle]

Paul Thomas wrote:
> Jerry,
> 
> This is OK once you have responded to a few niggles and a question on 
> the functioning of the patch:
> 
>>  
>> +   array_loop_spec *ls;  /* For traversing arrays */
>>
> I know that this is in keeping with the originator's style but could you 
> start to set comments above and aligned with the line of code.
> 

OK, will do.

>>
>> !             carry = 1;
>> !             record = 0;
>> !             rank = GFC_DESCRIPTOR_RANK(ioparm.internal_unit_desc);
>> !             ls = current_unit->ls;
>> !             for (i = 0; i < rank; i++)
>> !               {
>> !                 if (carry)
>> !                   {
>> !                     ls[i].idx++;
>> !                     if (ls[i].idx > ls[i].end)
>> !                       {
>> !                         ls[i].idx = ls[i].start;
>> !                         carry = 1;
>> !                       }
>> !                     else
>> !                       carry = 0;
>> !                   }
>> !                 record = record + (ls[i].idx - 1) * ls[i].step;
>> !               }
>>
> I think that it is worthwhile to break this out into a separate function 
> while you are about it.  I think that this section now appears four 
> times between list_read.c and transfer.c?
> 

I thought so also, will do.

>>
>> !   !   for (i=0; i<rank; ++i)
>>      {
>> !       /* Initialize the loop specification */
>>        !       internal_unit.ls[i].idx = 1;
>> !       internal_unit.ls[i].start = desc->dim[i].lbound;
>> !       internal_unit.ls[i].end = desc->dim[i].ubound;
>> !       internal_unit.ls[i].step = desc->dim[i].stride;
>> !       !       /* Determine the number of records */
>> !       !       record_count += (desc->dim[i].ubound - 
>> desc->dim[i].lbound)
>> !                       * desc->dim[i].stride;
>>      }
>>    return record_count;
>>
> ...and this too.  This will make both this and the namelist code look a 
> lot tidier.
> 

Agree,

> Now for the question: I wonder if this behaves as it should?
> 
> [prt@localhost mytests]# cat arrayio_test.f90
> character*4, dimension(8) :: abuf = (/"0123","4567","89AB","CDEF", &
>  "0123","4567","89AB","CDEF"/)
> character*4, dimension(2,4) :: buf
> character*8 :: a
> equivalence (buf,abuf)
> read(buf(2, 1:3:2), '(a)') a
> print *,a
> end
> 
> [prt@localhost mytests]# ./a.out
> 456789AB
> 
> g95 and ifc yield
> 
> 4567          ie. the end of record is acted upon.
> 
> Paul
> 

I will study this a bit.  Without looking yet, I think s_alloc_r_at is not 
catching the end of record.  This may be a new bug.  (I wonder if this really 
should be a run time error, trying to read past end of record)

I will report back on this.

Thanks for comments,

Jerry