Ideas About Tools Support for new Group Iteration Features
==========================================================
Robert E. McGrath
March 10, 2005

An overview of the approach to new features for groups is in:
  http://hdf.ncsa.uiuc.edu/RFC/ReviseGroups/GroupReviseOutline.txt

This note has thoughts about about the impact of the new Group iteration 
feature on tools (h5dump, h5repack, h5diff).  The impact on other software,
such as HDF-EOS5, can't be determined without more information about
the new API and compatibility with 1.6.

Assumptions
===========

It is still uncertain what the features will be or what the API is,
but based on conversations it seems likely that utilities
will be faced with:

   groups with index-by-name  (to an application the same as now, 
                              though different in library.)
   groups with index-by-name and index-by-creation

I would say we should not allow the case of index-by-creation without
any names, or else provide a default result for get name, e.g., return 
the string "1" as the name of the first object.  This will assure that
all existing programs will continue to work.

Assuming this is the general flavor of the API, then a calling program 
should be able to:

  get the name of an object (always,, fake if needed)
  get the creation order of the object, if available
  iterate by name (always, fake it if need be)
  iterate by creation order, if available


Preliminary analysis of standard utilities
==========================================

Here is my analysis of the tasks to upgrade the utilities.  IMO,
all of these are enhancements that do not have to be done for the 1.8.0
release. 


h5dump
------

Should work OK without modification: default ignores creation order
if available

New options:

1. display creation order in DDL
2. sort by creation order, if available


h5repack
--------

Should work OK without modification, but will not preserve creation
order.

New feature:

1. if creation order indexed in source file, create the objects in
the same creation order in the output.  Create creation order index
in output.

New options:

1. don't make creation order index
2. make creation order index even if input didn't have one.

h5diff
------

Should work OK.

New option:

1. compare and report if creation order is different. E.g., /foo/bar is
in both files, but was the 2nd object in /foo in one, and the 3rd in the
other file.


Other tools, e.g., HDFview
--------------------------

Current code should work, default will ignore creation order.
Probably will want new options similar to above, TBD.