On Mon, Feb 25, 2013 at 12:39:27PM +0000, Stefano Stabellini wrote:

On Sun, 24 Feb 2013, Greg Kroah-Hartman wrote:

quoted

You should not have unstable options in the kernel in the first place,
sorry.

 
With the premise that the removal of CONFIG_EXPERIMENTAL is not an issue
for me personally or my work, I am going to give you my 2 cents on the
matter, but feel free to ignore them :)

While I understand that CONFIG_EXPERIMENTAL has been abused, I feel that
rejecting everything that is not fully stable and with external
interfaces set in stones, might hinder the development of new features.

It's been this way for _years_ this isn't something new (the "you have
to get it right really quickly" problem).  See Documentation/ABI/ for
some words about how you can try to do this.

After all, given how fast the kernel is moving nowadays,

No faster than it has in the past.

maintaining a project out-of-tree until is completely ready for
production can be very expensive. Merging the project earlier and
completing the development upstream can bring better results.

Yes, but don't go changing user-visable apis when you do so.  That's
been a hard rule for a LONG time.

But in these cases one wouldn't want to "market" the feature as stable
yet, because it just isn't. If CONFIG_EXPERIMENTAL is going away, is
there anything in the kernel that can be used to tag a feature as "I
wouldn't use it in production if I were you"? Maybe just a comment in
the kconfig description?

I know this is hard, I've had my own problems with it in the past.  You
don't know if you get an api right until you have a lot of users.  See
our previous "discussions" about this topic on lkml if you are curious
as to the eventual outcome of threads like this:

	Yes, it's hard, but that's kernel programming.

Sorry,

greg k-h