Re: ISIS and heasoft 6.6

On Wed, Dec 03, 2008 at 10:02 -0500, John E. Davis wrote:
> John Houck <houck_at_email.domain.hidden> > I don't think it's possible to build xspec local models without
> > a complete xspec installation.
>
> How do those with xspec local models handle the updates?

Because xspec local models are compiled to produce a library,
not an executable like isis, they don't to resolve every symbol
provided by the xspec libraries and they don't need to
explicitly search other shared libraries to find arbitrary
symbols mentoned in the model.dat file.

Isis has to do both of these things.

> Is the xspec local model interface stable?

It has expanded several times over the past 10 years, with the
introduction of new parameters to fortran routines, then with
support for C and then C++ implementations.

In this instance, the problem is with extern "C" linkage to
C++ models new with heasoft-6.6.  xspec is supposed to use a
certain naming convention for these symbols, but this
convention was not followed in the heasoft-6.6 release.

Because the new models don't have the names that isis expects,
isis cannot find them, and the xspec module fails to load them.

I could work around this by coding these symbols as a special
case, but this approach would eventually break the ability
to auto-generate code to support new xspec releases.

As long as heasoft follows a consistent naming convention this
process can be automated.  The current xspec module in isis
relies on this naming convention.

Another issue is that users will expect to build their xspec
local models using the method documented in the xspec manual.
That method relies on xspec internals that aren't easily
separated from the model library and will not work unless a
full xspec installation is available.

> If so, couldn't isis provide the necessary symbols?
>

No.

Thanks,
-John