Re: Fixing the lm-sensors/i2c mess
On Wed, Dec 31, 2003 at 11:11:26AM -0500, Nathanael Nerode wrote:
> Christoph Hellwig wrote:
> >There real problem is that the new modules changed the API in an
> >incompatible way. This could be easily fixed by adding the old methods
> >back, what about escalating this back to upstream? This affects a lot
> >more than just Debian.
> I think you are mostly wrong. The modules changed the API, yes, and the
> ABI as well. It appears that it cannot be trivially fixed by adding the
> old methods back since data structures have changed.
As the port who did that work in 2.5 before someone backporting it
without considering the implications I can tell you that yes, you can
at least make it API-compatible. ABI-compatiblity is a completly
> But if it's not acceptable, then one of the following needs to be done:
> * lm-sensors needs to be dropped for 2.4 kernels entirely (is that
> acceptable to you?)
I'm not an x86 users, so I'm the wrong person to ask, but I guess many
people would scream then. Still better than the current situation,
> * lm-sensors needs to document its incompatibility with ALL DEBIAN 2.4
> KERNELS. (Is that acceptable to you?)
That's the safe variant.
> >Please make sure the mod_inc_use_count / mod_dec_use_count methods
> >are still present in the i2c function vectors and at the same problem
> >as before. Then most of the problems will just magically disappear.
> Since you believe that this is so easy, why don't you write and submit a
> patch in this manner fixing everything? For that matter, why didn't you
> do so months ago? I notice that you haven't. The Debian maintainer
> doesn't appear to feel capable of doing so, nor does upstream -- I
> wouldn't dare to. You want to, feel free to do so!
As mentioned above I did the conversation in 2.5 and told explicitly
that this is not a good idea for 2.4. I'm not very much interested in
keeping the pieces now if the lm_sensors folks are so ignorant.
> >Adding the 'new' i2c code will just trade breakade with the lm_sensors
> >modules vs breakage with other i2c modules.
> Yep. The 'new' i2c patches, however, are supposed to contain fixes for
> all the known drivers in the kernel tree which use i2c; did you even
> *look* at the page? That doesn't sound like it will create breakage to me.
Yes. The lm_sensors CVS is a _pain_. During the 2.4 series I've done
a lot of work importing their fixes without importing they're new
breakages or reverting fixes that went into mainline, and in 2.5 I've
done some significant rewrite of the i2c code until Greg took over
maintainership because he has much more time. And the fixes they have
in their tree in addition to what went into 2.4.24-pre doesn't look
important to me.