Re: reminder: GC phone mtg

[Dave Stampf <drs@bnl.gov>]

Hi,

I had planned on calling into the meeting but other events overwhelmed
me. I'd like to give a quick status report on the Objectivity port to
this group tho...

I went to Objectivity for two weeks 8/31 - 9/11. The purpose was to 1)
solve the Sun/Linux interaction problem, 2) get the shared libraries
done, and 3) look at the compatability of our port with Red Hat 5.1 and
the egcs compilers. I wouldn't qualify the trip as a success, but alot
was learned.

First of all, a couple of really important people left objectivity
before I arrived - Rick Spicklemyer who was the head of their
development group and the person who worked closely with me on the
previous two trips left for another startup, and a few days before I got
there, Dragan Milanovich who was the head of the QC group and taught how
to use Objy/C++ also left. At the same time, his second in command is on
sabatical. So virtually all of the people I was supposed to work with on
this trip were gone.

The results of the work:

1) The problem with the Linux/Sun failure to interoperate is *not* the
fault of the linux port. Every architecture uses an array of machine
descriptions which spell out byte ordering etc. Linux was the last added
to the list and the Sun version (at least) does not check to see that it
has walked off the end of the array when it is told that architecture
#24 wishes to communicate with it. Crash. See ComoSci 101. It should
work in the 5.1 release, sometime in October/November.

2) We delayed this until after 3)

3) Redhat 5.1 is proving to be a real pain. However, for this port, what
matters is that the default configuration of compilers for 5.1 is gcc
2.7.2.3 (because 2.8.x has trouble compiling parts of Linux - gulp) and
egcs for c++ version 1.02. While I was there, egcs 1.1 was released.
Initially, my feeling (and Objectivity's feeling which has not changed)
is that they want to run with a commercially released version of Linux.
This point of view is bolstered by the fact that the new egcs 1.1 is
incompatible with all other binaries on the Linux release. To switch to
that compiler demands that your recompile all of your existing libraries
*and* re-acquire any other libraries that you may be using (motif? etc.)
It could be quite chaotic to play the usual Linux game of upgrading
compilers/kernels on a daily basis.

Anyway..... the egcs 1.02 compiler isn't ready for prime time. It failed
to compile only one line of the Objectivity source - it didn't like
having a default function argument be the use of the new operator. We
were easily able to program around that. The code produced by ddl
however was another story. Error messages such as "foo not implemented
yet" on single line functions where common. These single lines were
really fairly complex code involving macro expansion, template
invocation and use of -> operator. We may have been able to work around
any one ddl output, but doing it for the general case and feeling
confident in the results was seen to be not worth doing. For kicks, I
loaded the egcs 1.1 compiler and had similar (tho more subtle) compiler
bugs. I have the relevant parts of the code here and I will report the
bug to egcs.

3) We didn't attempt this, but I've been told that they will do it for
u. Part 2 consumed all of our time trying to chase down compiler bugs
and one very bad Objectivity bug (an array being declared different
sizes in different functions which ended up corrupting part of the
stack) and I think we want to keep looking ahead to 5.1/egcs -
especially since so many people have expressed an interest -- a demand
actually -- to use the egcs compiler in the near future. I hope that
sometime soon (January?) Redhat 5.2 will be available with a compiler
that will handle Objectivity.

So - what part of this is the good news?

Well, it seems that Objectivity has a formal relationship now with
Redhat which to me means that they want this as a product. Also, the
first two times I went there, I did my work apart from the main source
tree. This time, the linux code is part of the main source tree --
another indication that they plan on supporting this port. Also, it is
really quite easy to configure in a new machine now, so I won't have to
spend + days configuring. Also, this time, they assigned themselves some
work to do before I return.

I also have them seriously considering permitting the source code to be
located here on a secure machine so that we can continue to work on the
port in a timely fashion. (I've spend 7 weeks in the past 5 months there
which can't continue!) This change in attitude was helped by the request
from Cern to have a source license - shared with the HEP community in
general. Many feel that it will happen and letting the source out of
their company a few months early won't hurt. Keep your fingers crossed.

That's about it. Sorry to have missed the call. Please don't construe
anything that I've said here as a negative feeling about Objectivity.
They have been enormously accomodating under some pretty nasty
circumstances earlier in my visit and we share a mutual respect at this
time. Our work is very interesting and important to them and I'd like to
stay in their good graces!

Cheers,

Dave

Doug Olson wrote:
> 
> 17 Sept. 98 Phone / Web Meeting
> 10:30am PDT, 12:30pm CDT, 1:30pm EDT
> 
> Dial-in number - (516) 344-8261
> 
> 
> 
>     ---------------------------------------------------------------
> 
>                    Name: Douglas Olson.vcf
>     Part 1.2       Type: unspecified type (application/octet-stream)
>                Encoding: 7bit

-- 
David R. Stampf     | "Beyond a certain point, there is no return. This 
Brookhaven Nat. Lab.| point has to be reached." 
Upton, NY 11973     |   
1-516-344-4148      | Franz Kafka