[Leaplist] good read -- a few good men--er, reads--er, distros--er, whatev--er

[Bryan J. Smith]

On Thu, 2008-03-20 at 22:05 -0400, Justin M. Keyes wrote:
> I've always wondered about that myself. Every time I try to get more
> cutting-edge than Debian "testing" (i.e., Debian sid, Ubuntu, or even
> Fedora, in my experience ;)

First off, Fedora is neither based nor has any lineage at all to Debian,
it's not comparable.  Debian has greatly influenced Fedora (and I
regularly praise Red Hat for taking "the best attitudes" from Debian
ex-post-facto), but there is absolutely no lineage.  In fact, the most
remotely thing you could say is that Debian "follows" Fedora and doesn't
"trailblaze" like Fedora every 2-4 releases.

On that note, secondly, every 2-4 releases of Fedora have purposeful,
well announced breakage.  Debian doesn't do that.  Fedora "trailblazes"
new features being considered for a future Red Hat Enterprise Linux
(RHEL) product in 12-24 months.  Debian often doesn't adopt them in
"STABLE" until the RHEL product has them (if not later).  I assume
that's what you're referring to.

In response to people demonically saying, at the expense of Fedora, "our
distro waits until its stable," I've more recently and bluntly called
this "riding the coattails of Fedora."  I.e., Fedora does the
integration of its own pacakges, identifies possible regressions and
works with the community to accommodate them, and after that is when
other distros adopt them.

I've recognized this reality of Red Hat's "community distros" ever since
Alan Cox personally responded to my "An Open Letter to Red Hat" on Red
Hat Linux 7[.0].  Everything Red Hat does is extremely calculated, and
always has been since Red Hat Linux 4.0, especially when it comes to
GCC.  That's why they have the longest-running C/C++ ABI/API
compatibility of any distro vendor, despite demonizations to the
contrary.

Most recently Red Hat discarded GCC 4.2 and had already backported
anything of use to 4.1, forging ahead with GCC 4.3 for Fedora 9 and the
likely RHEL 6.  Too many distros "just ship latest."  Red Hat jumped
over GCC 2.8 directly to EGCS 1.1.2 (later GCC 2.91.66) and "held back"
and away from GCC 2.95.x (which had varying levels of ABI/API breakage).
They are commonly demonized on GCC 2.96, and even though it was an
official GCC release (because Red Hat was the maintainer) due to a
single developers inaccurate posting (./ did the rest, which means he
basically slit his own throat), the real reason for the breakage was the
"arm twisting" of ANSI C++ compliance.  Something Cygnus (now Red Hat)
had repeatedly tried to "enforce" with EGCS prior.

It's also the reason why object binary compatibility goes back the
farthest with Red Hat.  I'm still running 2.91.66/2.96 libstdc++ 2/3
applications from Red Hat Linux 6/7 on not just RHEL 5 but even Fedora
8.  Any distro that used 2.7.x, 2.8.x, EGCS 1.1.2/2.91.66 and 2.95.x
basically has 4 incompatible ABI/APIs prior to GCC 2.96+.

> , I get bit. So I'm quite happy with
> Debian, despite annoyances like the fact that pgadmin has been stuck
> at 1.4 in the debian main repos for way too long :( Yes, I can live
> without compiz/beryl until it's blessed by apt.

I don't understand what you mean by "bless by Apt"?  Debian is the
reason, not Apt.  There are so many things attributed to Apt which are
really Debian.  Saying such is just pushing the same non-sense that
Ubuntu and other advocates speak.

There is no Beryl, it's been Compiz-Fusion for awhile now.  Compiz
extremely stable, and has been for a long while.

The issue isn't the end-user software, it's the performance v. kernel
coherency hacks.  The "GPL" drivers, like those from Intel, just discard
performance altogether, not offering the kernel coherency hacks that are
used in Windows (which is why Intel performance on Linux sucks compared
to Windows, and Intel** already sucks pretty hard on Windows** ;).
ATI's reliability has improved significantly, and AMD has opened
virtually everything now (although the open source road will never get a
full kernel driver with similar capability as their closed source, which
has Intel, Microsoft, nVidia, SGI, etc... IP), but they still trail
nVidia's nearly 9 year history in a solid, cross-platform object with
its kernel coherency hacks.

AIGLX makes it even more stable from a hardware standpoint, because 3D
is only used as it is needed -- unlike full, "always libGL" window
managers like Xgl.  This approach on Fedora also makes even old Intel
855G GPUs useful for basic things, although the more objects using 3D,
the more it slows down.

[ **NOTE:  AMD/ATI's new 790G chipset GPU is 7-11x faster at most 3D
titles than Intel's new X48 GMA.  The 790G GPU also has a
"Crossfire" (dual-GPU) option by adding a single Radeon 3450 (IIRC?)
card in its PCIe x16 slot (which is really a fusion of HyperTransport
technologies), giving it quite capable 1024x768 performance at today's
gaming titles without full screen anti-aliasing (FSAA).  One
semiconductor analyst believes the mobile 790G + dual and triple-core
Phenom (lower cost), despite the "raw" CPU/SSE performance, will cause
Intel to lose up to 50% of the mobile market in 2008.  Unlike Intel, the
790G is manufactured at a 55nm process (well below most Intel chipsets),
and the new 45nm dual/triple-core Phenom for portables coming out of
Dresden and IBM's foundry will make the cost-performance a no-brainer. ]



-- 
Bryan J  Smith              Professional, Technical Annoyance
mailto:b.j.smith at ieee.org  http://www.linkedin.com/in/bjsmith
-------------------------------------------------------------
           Fission Power:  An Inconvenient Solution