On Friday 24 August 2007 14:12, Randall Thompson wrote:
If the consensus is that
there should not be any connection between the primary header and
extensions.
perhaps this should be stated in the revised standard. For example,
state that it
should be possible to add extensions without requiring any changes to the
primary header. This would of course deprecate the EXTEND keyword.

It's a complex issue with no easy answer. In many interferometric FITS
files, there are binary tables which sometimes cross reference each other.
Without a unique FITS extension identifier, this may be tricky. I just wanted
the voice the potential danger.

I don't see any implications on EXTEND as it does neither explicitly refer
to the number of extensions nor to an individual extension but only indicates
the general format of them. As said before, I have no problems with making
it optional (or even deprecate it) assuming that it is obvious that no other
kind of extension can exist. I still believe that it was justified and useful
during the transition period but that has long passed.

Preben