Re: arrays and derived data types in namelists

[Paul Brook <paul at codesourcery dot com>]

On Wednesday 24 November 2004 09:23, Paul Thomas wrote:
> Is there a strategy in the collective (or not!) mind for implementing the
> missing features in namelists?  Most specifically, a lot of the code that I
> work with includes arrays, and occasionally components of derived data
> types.
>
> I have done a bit of tidying up work on list_read.c and thought to try to
> implement arrays, at very least, before submitting (a) patch(s).

If possible tidy up work should be submitted separately from new features.

> I notice that the descriptor dimension array and the derived components are
> missing from namelist_type.  Is it intended to add them to this structure
> or a link back to the array descriptor itself?

I think for arrays it makes sense to pass the array descriptor, rather than 
the g77-style data pointer plus duplicating most of the array descriptor 
information. This gives the compiler the option of reusing an existing 
descriptor.

We've thrown a few ideas around on IRC, but didn't reach any conclusions. 
We need some sort of interface for registering arrays and derived types.
Two strategies have been suggested:
- Extend the existing st_set_nml_var_* routines, adding new ones for arrays 
and derived types. Use these to construct the type descriptions at runtime.
- Have a single st_set_nml_descriptor entry point. Build an in-memory 
description of the namelist type at compile time, and pass this. Similar to 
how formatted IO is done, but using a compiler/library friendly format rather 
than a human readable string.

Both of these have advantages and disadvantages. I personally favour the 
latter.

Paul