2.3.51 tulip broken

[Jeff Garzik]

On Thu, 16 Mar 2000, Jim Morris wrote:
> Well, yes and no.  Thinks like filesystem drivers certainly belong in
> the kernel.  But I am not convinced that specific hardware device
> drivers should be developed as part of the kernel source tree.  I'm not
> saying they cannot be shipped with the kernel source - but I don't see
> why the primary development of the drivers has to go on in the kernel
> tree, and not as a separate development project.

Kernel infrastructure changes, and drivers need to track that.  You
simply MUST expect changes across different -stable- versions of the
kernel drivers.


> > I personally hate installing drivers off a floppy.  It's much easier if it
> > just works without having to change something on your system.
> 
> I'm not saying a Linux distribution should not include all the latest
> and greatest device drivers on the CD-ROM or FTP image.  I'm just saying
> that maybe it would be acceptable to have more of a split between
> "Linux" itself, the kernel proper, and independent device drivers.

This becomes easier after the 2.3 changes, but a driver simply must
driver kernel versions.  There is no way around that.  Older drivers
might continue to work with newer kernels, but if they are not updated,
they do not work well.


> Maybe so.  If nothing else, I don't think Linus should refuse to put the
> latest and greatest version of a specific device driver in, just because
> the official maintainer of that peice of software didn't send him
> incremental patches.

The whole reason for incremental patches is that it makes it easier to
find and resolve bugs and other issues.  In development kernels, you
must also update (and test) for infrastructure changes.


> Its like when I do a development project.  I may make a LOT of changes
> from one relesae to the next, and I may or may not commit those changes
> to CVS.  When a release is ready though, I commit it all, and apply a
> release tag.  At that point, I give the release to QA for testing - and
> if QA checks off on it, I cut a release for the customer.  I don't give
> QA or the customer all of the incremental changes between released
> versions of the program!

Think about about the kernel is developed.  The odd series is
development, and would never go to a customer.  The even series is
stability, and does go to customers.  And you can be SURE that both the
customer and the developer will appreciate incremental changes when it
turns out that a stable bug fix is not really stable.

Changes are easier to isolate with incremental changes.  One big update
means things work, or they don't -- with the fallback (before the
update) typically being a far inferior, and more buggy version of the
software.

	Jeff





-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-tulip-request@beowulf.org