Good bye

Les Mikesell lesmikesell at
Fri Feb 1 17:56:43 UTC 2008

Chris Jones wrote:

>> When the same program works on one version of an operating system but 
>> not on another, there's nothing to ask.  The operating system has 
>> clearly failed to provide a usable interface.
> We don't want stable interfaces (not quite yet, maybe in time).

Freedom doesn't consist of telling other people what they want.  It is 
about everyone having choices.

> Windows 
> may have more stability in this regard, but the down side is you end up 
> stuck with bad interfaces that you cannot change due to 'compatibility' 
> issues. Users of these interfaces then start writing there own kludgy 
> workarounds. A mess.

Examples?  I have reasons to dislike windows, but they mostly go back to 
pre-win2ksp2 days.  I can't think of any particular driver-related 
issues that are a problem now.  Likewise the binary OS X drivers work 
just fine for me.

> Linux IMHO, has a different view. If an interface is broken don't try 
> and live with it but fix it. If this breaks things downstream so be it, 
> in the OSS world fixing things is not so hard. The problems are with 
> closed sourced entities where those changing the interface cannot fix 
> the closed source.
> Lets be specifc - Most of the issues you refer to (nivida, vmware) 
> relate to the kernel.

OK, let's be specific.  Start with firewire.  I can't afford to deal 
with firewire as it has been handled in fedora for the 1st 6 versions 
and haven't tried with the last 2 yet. Show me how having people who 
have no reason to care if a device driver works or not writing the code 
has worked out better than making it simple for vendors to provide drivers.

> Here developers may change some interface, but 
> such changes are internal to the kernel, like drivers. In this case 
> those changing the interface fix the driver so the end shot is all is 
> still OK. They cannot do this for closed source stuff, which is why such 
> things are disliked and distros like Fedora do not support them. A quote 
> from Linus I recently read
> "Asked by Zemlin why the Linux kernel does not have a stable device
> driver application binary interface, Torvalds said, "We really,
> really, really don't want one. 

Once again, I'm not particularly interested in someone telling me what I 

> The main reason most people want a
> stable ABI [application binary interface] is so that they can have
> their binary drivers and not have to give out source. They don't want
> to merge that source into the stable kernel or the standard kernel.""

And that part is simply self-serving refusal to cooperate with anyone 
else through standard interfaces.

> I for one agree and do not want the kernel developers to stop changing 
> (improving) things just to provide a more stable interface to external 
> closed source stuff drivers etc.

I might have agreed back when I thought Linus was inexperienced at OS 
design  and incapable of specifying an interface suitable for use over 
some length of time.  I don't think that is the case anymore.  And I 
might has agreed that source was important back when many companies were 
competing with incompatible processor designs and there was some change 
a device might need to be moved to a new/different system.  Now I just 
want an interface that works for the life of the box.

 > If those closed source providers want
> to support linux, they either keep up with the kernel developers (which 
> they could if they wanted, kernel release candidates are always 
> available before release) or they allow their drivers to be OSS'ed and 
> placed in the kernel, where they will be maintained 'for free'.

I can't afford to use something 'free' that doesn't work as well as 
something written by the vendor that knows the most about the device in 
question and has the most reason to care about it operating correctly.
There's nothing wrong with source availability and, other things being 
equal, it is a good thing if everyone chooses not to place any 
restrictions on what others can do with it. However it is no more 
necessary for me to have source to a device driver for my purposes than 
to have all the parts to assemble a jet before I would ride in one.

> One day kernel development might slow down, but I don't want that to 
> happen prematurely and for us to end up with bad interfaces.

How old must Linus be before you'll trust him with this decision?

   Les Mikesell
    lesmikesell at

More information about the fedora-list mailing list