layers of abstraction, and how Gnome can win

[Brian Behlendorf <brian collab net>]

Note: though I am a vendor to Sun on the OpenOffice project and have been
looking at these issues for them under contract, I in no way speak for
Sun, OOo, Apache, or anything other than myself.

One of the subtle nuances in the "what is gnome office" thread that didn't
get explicitly discussed is one of layers of abstraction.  I believe this
is the fundamental difference between the Gnome and OO projects, and I'd
like to point out how by asking a few questions:

a) Is it the mission of the Gnome project to create the best free (as in
free-speech) desktop for Linux?  Or is it to create the best free desktop,
period?  I use FreeBSD and Gnome on my Vaio, and it's great - I've never
come across a bug in a Gnome app that struck me as being due to that app
not working on FreeBSD.  And clearly Sun and HP wouldn't have lept onto
the bandwagon if getting Gnome working well on Solaris and HP-UX wasn't
also not only permitted, but encouraged.  So I presume the answer is that
Gnome aims to be the best free desktop for all Unices... but what about
Windows?  I'm aware of the gtk-win32 port (yay!), but what about running
an entire Gnome desktop on top of a win32 kernel, replacing Explorer?  Or
on MacOS X/Darwin, replacing Aqua?

b) Is it Gnumeric's mission to be the best free spreadsheet out there,
whether you're using KDE or Gnome or even just a bare windowmanager?  If
someone contributed a series of patches to get Gnumeric to compile cleanly
and look nice under Win32 native widgets, would they be accepted?

As far as I can tell, OpenOffice aims to be (er, Sun aims it to be) a
truly cross-platform office suite.  Many, possibly most of the developers
in Germany are clearly developing in a Windows environment.  Like Mozilla,
they went through great pains over the last couple years (I've heard 10
years from one person) to create their cross-platform suite of portable
APIs they can write the core library of apps with, an environment they're
now pretty comfortable with, though it of course represents a substantial
learning curve to outsiders.  (That's not deterred patches from a couple
dozen outside contributors in the first two months of the project, some of
them fairly newsworthy bug fixes).  This affects everything - object
component system, UI, even things like string classes.

Today, OpenOffice at least has some hope that it can run consistantly
across Windows, Linux, and Solaris, the platforms that Sun developers
target.  It used to run on many more (a point of pride for the German
developers, I was told, was when a new processor or OS would be announced
and an OO port was available within days) but I believe they are focusing
on those three, leaving others to the community; there have already been
substantial patches offered for a freebsd native port, SGI, and
linuxppc.  But those all are platforms which do or should run Gnome, it
sounds like.

So the fundamental question is, if a company wants to develop a
cross-platform app, cares as much (if not more) about running on Windows
and non-Gnome-supported Unix-like OS's, as they do on platforms covered by
Gnome, what should they do?  Both Mozilla and OpenOffice are, by inertia,
choosing to maintain their cross-platform infrastructure in order to meet
this need, and any interfacing to Gtk or Bonobo will probably be in
the form of compatibility layers and bridges instead of core
architecture.  This is clearly not optimal - it bloats the apps
tremendously and makes for a whole lot of repeated work.

They could do what the Gnome hackers have asked - rip out anything that
resembles functionality already in the Gnome universe and replace it with
its Gnome equivalent.  I bet that would strip it down to a much smaller
size, improve performance, etc etc.  Or someone could pull a Galeon and do
it on them, though I suspect the code's complexity at this point would
dissuade anyone from trying.

But if they did that, if they really adopted Gnome as whole-hog and
single-mindedly as the office apps that came from within the existing
Gnome community have, then they'd lose the ability to run on Win32.  I
don't speak for Sun or the development team in Germany, but I suspect that
Win32 support is important enough that this would be a non-starter.

Why care about Win32?  And why am I writing about Win32 on a Gnome list
anyways?  Because I believe part of the reason why Sun (traditionally not
an end-user software company) launched the OpenOffice project and has
several dozen engineers dedicated to it is to commoditize the market for
office software in general.  Recognize that Microsoft makes over half of
its gross revenue from Office; and that due to the commoditization
of the OS thanks to Linux, Office is today a bigger monopolistic force on
the enterprise than Win32 is.  

I believe a similar mentality would be good for the Gnome project to
adopt; that bringing the Gnome api's and components to Win32 is not only
an interesting project worthy of support, but that when resources are
available, they are applied towards that goal, that the cause is advocated
whenever possible, and that engineers with Win32 experience are not made
to feel like second-class citizens.  Let's turn embrace-and-extend back on
Microsoft - let's get Gtk, Bonobo, and other infrastructural pieces solid
on Win32.  We'll have to do the same thing to counter the .NET proposal,
but that's not for awhile.

We've done this in Apache now - Apache 2 will run as well on Windows and
OS's like BeOS and OS/2 as it will on Unix thanks to abstracting all
system-level issues we care about into a library called "apr" (Apache
Portable Runtime, kinda like Mozilla's NSPR but specific to servers and
Apache-licensed).  This is partly because the core server
developers were happy to specify an API (posix-like), and companies like
IBM and a couple motivated developers were willing to do all the work
necessary under that API to translate it use the most appropriate native
calls, implementing new functionality as needed.  If Stallman and I can
resolve our BSD/GPL compatibility issues, then it's something Gnome could
use as well, though we don't touch UI issues at all.

Yes, I know the typical answer whenever an outsider lectures an open
source project on what they should do: "we await your patches". What I'm
asking here though is much more about policy and leadership than specific
code suggestions.  Maybe this is all actually happening and I don't
know; if so, great!  But let me tantalize you with a thought: Gnumeric
running on Windows, and Windows users suddenly able to use a desktop and
office suite much more interesting and free than what they can choose from
today.  This is how Gnome can win - not just over KDE, but over that other
OS.  

Thoughts?

	Brian