On Mon, May 07, 2007 at 11:22:34AM -0400, Jeff Garzik wrote:

John W. Linville wrote:

quoted

On Mon, May 07, 2007 at 08:03:43AM -0400, Dan Williams wrote:

quoted

On Mon, 2007-05-07 at 11:41 +0100, Christoph Hellwig wrote:

quoted

quoted

Of course it's not anywhere near good shape.  Almost all items from my
review were completely ignored, and we have another totoally substandard
wireless driver with crappy thread handling, a huge number of broken 
private
ioctls and partially absymal codingstyle.

quoted

I've already updated libertas-2.6 git with a ton of updates for this.

In any case, lets push off any merge until 2.6.23 so the rest of the
comments can be dealt with:

Alright...Jeff, would you please pull the following branch for upstream 
ASAP:

git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-2.6.git revert-libertas

This is leading from behind :/  We don't need to blow about in the wind 
here.  If you reviewed the driver in depth -- which I assumed because of 
the trust placed in you as wireless maintainer -- then this situation 
really should not be happening.  You need to know the status of new 
drivers you are pushing upstream: what work is left to do, what has been 
done, what state the driver is in.

I view this request as a failure of the trust network :(

For my part, I _did_ review it.  Twice.  Once in the early days, and 
once when I pulled it into my netdev-2.6.git tree.  libertas needs the 
changes mentioned in this thread.  But the driver is in workable shape 
to be USED while being improved.  I strongly dislike people being cowed 
into not merging a driver for years, because the driver in question does 
not meet Christoph's idea of perfection.

It's a shame to expose new ABI bits that we expect to change to
mainline. Getting rid of obsolete interfaces is next to impossible so
the introduction of new interfaces really does warrant serious
consideration.

Can we come up with a scheme to keep the new ioctls introduced by this
driver from leaking over into distro-land before they get reworked?
Like preemptively adding a deprecation printk?

-- 
Mathematics is the supreme nostalgia of our time.