Here are the notes I had from the last meeting.  John Marrafino spoke on PAW
maintenance and support. Please feel free to chime in if I missed anything
or am incorrect anywhere - Gene

COMIS -source of a lot of trouble, first piece to break, broke on SGI now.

PAW - 4MB exe, not counting shared libraries.

Support by CERN essentially stopped. Unstable against new bug fixes (fixes
generate bugs). Complete turnover of people to LHC++.

COMIS -> IRIX 6. May work with n32, sure to break on N64.  Argument passing
machinery broke (can't _use_ Double Precision now, only floats).

There is a workaround for COMIS interpreter problems: A compiler that will
generate the dynamic link libraries for linking into PAW.

IEEE format for Zebra.  CERN claims will break in going to 64 bit in IRIX
(N64 for 64 bit data). CERN estimates this will take 1/2 FTE-year to port.
Are we going to have to support his anyway?

It was pointed out that under IRIX, in a system composed of linkable
executables built from different sources of libraries, suppliers of
libraries  probably must make them available in all  3 ABIs: O32, N32, N64.
Not a problem on digital Unix (one ABI).

Q: Isn't digital Unix 64 bit and PAW works on it?
A: Yes, though only using 64 address, not 64 data. Also, COMIS breaks
because pointers are internally stored as longs (32 bit). Kludge to get this
to
work was use code section in lower virtual address range so that upper 32
bits
are insignificant. Yuch.

Names are not very mnemonic and hence the code difficult to follow. Not well
commented either.

Q: Frank - What would you look for wrt support and maintenance:
A: KISS - vanilla programming.


Q: Is COMIS a hack? There is some concern about cint.
A: Yes.

In some forgotten context about commercial products, Qizhong pointed out
that the main drawback is they most certainly won't have the functionality
required by physicists and that draws one immediately into issues of
responsiveness to users requests for extensions/additions.