Crash if trying to write a parameter or a constant in a Fortran application

[Rob Latham]

On Tue, Jul 09, 2013 at 04:23:37PM -0500, Wei-keng Liao wrote:
> One problem is Fortran allows ones to define a parameter array.
> Such as
>       integer, dimension(8192), parameter :: varp=0
> 
> I am not aware of a way to tell a subroutine argument is
> parameter or not in Fortran or C. In the above case, segmentation
> fault will still occur if writing the entire 8192 elements, though
> I wonder if it is rare to see one using large parameter variables
> in Fortran.
> 
> If we want to document this, we might want to generate an error
> code corresponding to it.
> 
> One possible solution to forbidding parameter buffers is to declare
> the intent of I/O buffer as inout
>           <type>, intent(inout) :: buf
> but it works only for F90.

I think that's a great solution.  I hope Philippe Wautelet can let us
know if that is OK for him.

==rob

> Wei-keng
> 
> On Jul 9, 2013, at 3:40 PM, Rob Latham wrote:
> 
> > On Tue, Jul 09, 2013 at 11:17:15AM -0500, Wei-keng Liao wrote:
> >> NetCDF 3 and 4 pass the test, because they do internal caching. So
> >> byte-swap is done on the cache buffer and the memory contents of
> >> Fortran parameters are not touched.
> > 
> > Ah, right, thank you for reminding me about the netcdf internal cache.
> > 
> > I'm tempted to just document this in the release notes, unless there's
> > a reasonably small threshold: for example, "if number of items is less
> > than 1024, make a copy for byteswapping in case it's a parameter"
> > 
> > ==rob
> > 
> >> Wei-keng
> >> 
> >> On Jul 9, 2013, at 10:14 AM, Rob Latham wrote:
> >> 
> >>> On Tue, Jul 09, 2013 at 08:53:24AM -0500, Rob Latham wrote:
> >>>> Does serial netcdf exhibit this behavior?  
> >>> 
> >>> Here's my serial netcdf-f90 version of your test case, but it's buggy:
> >>> I am apparently trying to write to a variable that does not exist?
> >>> Our policy is to be as netcdf-like as possible, or have a good reason
> >>> not to be, and I'd like to know how serial netcdf handles the case of
> >>> byte swapping data passed to it in this un-modifiable parameter form.
> >>> 
> >>> ==rob
> >>> 
> >>> -- 
> >>> Rob Latham
> >>> Mathematics and Computer Science Division
> >>> Argonne National Lab, IL USA
> >>> <netcdf_parameter.f90>
> >> 
> > 
> > -- 
> > Rob Latham
> > Mathematics and Computer Science Division
> > Argonne National Lab, IL USA
> 

-- 
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA