xml-general mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From James Duncan Davidson <james.david...@eng.sun.com>
Subject Re: [spinnaker] Announce
Date Mon, 10 Jul 2000 10:29:40 GMT
on 7/9/00 11:25 PM, GOMEZ Henri at hgomez@slib.fr wrote:

> Seems this announce turn in a IBM vs SUN position war.

I'm not making this a war. Other people will try. I made a mistake in my
initial posting that gives an opening for them.

> Please gentlemen, don't use OpenSource projects as a battle ground.

I don't want to. I want to produce a good parser for the needs that I see.

> In my early days in computer programming, my teacher says that the compiler
> must be adapted to programmers and not the reverse. Did we have to change our
> coding style to adapt to HotSpot JVM ?

Not really. Hotspot was designed to optimize naturally written code. Code
that isn't all tangled up with optimizations. By taking out the need to go
to great lengths to eek out performance, it should be more natural for
programmers to adapt to.

In addition, Hotspot does really neat things like producing machine code
optimized for the exact CPU that you have running. Say G4 Altivec optimized,
or P3 optimized. Or whatever CPU you have.

The only people that have to rewrite their code are those that heavily
optimized the hell out of their code for 1.1. The HotSpot VM people were
*extremely* clear about this at *every* single HotSpot presentation and
write up that they've done for the last *three* years. If you weren't
listening, it was probably because you weren't interested, but you should
have been.

> Did the Java community will have to be splitted in two teams. One using and
> coding for Sun's JVM (à la hotspot) and others using IBM's JVM ? Did you
> forgot (at Sun & IBM) the 'Write Once Run Anywhere' credo ?

Not at all. I can't speak for the Xerces team, but having worked on projects
that needed performance under JDK 1.1 (Java WebServer for example), its
natural to heavily optimize. This isn't a IBM vs. Sun thing here -- it's a
more modern VM vs. a more primitive VM. If anybody here thinks that JDK 1.1,
as shipped from Sun and what the IBM and other VMs are based on, was at the
cutting edge of performance abilities, I've got seashore land in Oklahoma
for you. (Though I have to say that IBM and Apple have done some pretty nice
work with JITs on 1.1 of late, still doesn't mean that its not time to move
on. :)

Programmers do the best with what they've got, yes. IBM VMs have been at 1.1
for a very long time. It looks like (though I don't know, I don't have
inside info) that they are jumping right to 1.3 and skipping 1.2. That's
great. I hope that they move faster with it because quite frankly 1.1 is
pre-historic to work on and deal with (even though I do it day in and out
with my Mac laptop :).

And I understand the Xerces team optimizing for what they ship product on
day in and day out. I never said "Bad Xerces Team, Bad" -- I said that
because of that, it has problems on more modern VMs. That's a problem for
me. A problem that I'd like to fix in time for when the world moves away
from 1.1.

Things change. Deal with it.

> If Xerces code is complicated, Ok, help Xerces team to polish it. There is
> many parts of OpenSource code around the world which is hard to read but
> it WORKS. All our End users are much more concerned by response time than
> by the beauty of underlying software coding.

And I'm not saying abandon Xerces dammit. I'm saying I want to take an
engineering look at how to do something that learns from Xerces. I want
users to continue using Xerces.

Half the reason why the Sun vs. IBM thing is a problem here is because some
third parties aren't listening to me... See next para.
 
> We are many annoyed to see more and more Apache Group projects started and
> lead by commercial companies. Did Spinnaker have to be another one ?

It's not. It was started by an ASF member who happens to work at Sun (and
make the mistake of talking about his team where he works -- bad me, I
should know better). They (people who work at Sun) didn't like the idea at
first. They were skeptical. Some still are. I'm sure that everybody at Sun
higher than my manager is probably shaking their head saying "dammit, what
is James doing *this* time".

If you want to continue to paint this as IBM vs. Sun. Fine. It ain't Sun vs.
IBM no matter how much you want it to be. And it definitly ain't me vs. IBM.
I said it in my initial post, I'll say it again. They've done impressive
things with Xerces, esp. on 1.1 VMs. It's admirable. It's good.

Just for the record, if you want to look at what I've done at Apache for the
last year, you'll see that I've not been a corporate shill. For example, The
Rules for Revolutionaries paper was written so that Craig McClanahan (before
he worked for Sun) could get the space he needed to start Catalina without
interference from the Sun team which, quite frankly, most of the members
wanted to kill. But, I'm pretty sure that this lil' nugget of info will fall
on deaf ears and that you are convinced that I'm a Sun robot. <sigh>

.duncan


Mime
View raw message