Re: Complaint about recent LINUX_VERSION_CODEs

[Brian Gerst]

Theodore Ts'o wrote:
> 
>    Date:        Thu, 10 Aug 2000 10:58:04 -0400
>    From: Brian Gerst <[e-mail]>
> 
>    The only logical reason to not use KERNEL_VERSION is if you want to
>    support 2.0.x (which should be deprecated, IMHO anyways).
> 
> There *are* still people using 2.0 in production environments.  Without
> naming names, let's just say that some important parts of the
> Internet/web infrastructure are still using Linux 2.0 systems, and that
> if you've surfed the web lately, you've probably used a 2.0 system
> without knowing it.
> 
> As Alan Cox has pointed out 2.0 probably had the most stable VM,
> compared to either 2.2 or 2.4.....  For folks who don't need the sexy
> new features of the newer kernels, sometimes staying on the older
> kernels for stability's sake makes good operational sense.

On the other hand, is it really necessary to carry 2.0's baggage on
forever in the current kernel in the form of #ifdef's sprinkled
throughout the source code (some of which are not trivial)?  For a good
example of what I'm trying to say, try to make sense of the Advansys
driver...  There's cruft dating back to 1.2 in there (including
alternate PCI configuration functions), and it's a rat's nest of
ifdefs.  That's the most extreme case, but there are other places in the
kernel that could use some cleanup as well.  The line needs to be drawn
somewhere or else these thing are going to continue to collect.

I also would like to point out that many of the drivers in 2.4 with 2.0
conditional code, wouldn't compile on 2.0 anyways, because of changes
that are unmarked.  The driver maintainers need to go through and
evaluate what versions they want to support, and clean up their drivers
accordingly.

--

				Brian Gerst