2.6.3-mm4

[Mike Fedyk]

Jean Delvare wrote:
>>You have to be kidding me.  Are you saying that with your patches to 
>>libsensors it won't support 2.6.3 style sensor sysfs names?
> 
> 
> Yes I am.  This isn't a fundamentally new problem.  Each Linux 2.6 has
> had its matching libsensors version that was not backward compatible
> (with earlier 2.6 kernels; it's still fully compatible with 2.4).
> 
> Keeping newer versions of libsensors compatible with all early 2.6
> kernel versions would make it incredibly more complex, with no
> significant benefit IMHO.
> 
> The facts are that the sysfs interface to i2c chips is just stabilizing.
> Greg's original naming scheme had many drawbacks (also we can't blame
> him for that, since nobody objected before me), as I have been
> explaining in a previous post on LKML. Also, many chip drivers did not
> comply with it in early 2.6 kernels, and new chip drivers wouldn't fit
> in it anyway.
> 
> The new interface is required if we want to write a chip-independant
> libsensors someday. I estimate that this is worth breaking the
> compatibility. The impacted kernels (later 2.5 and earlier 2.6) will
> obviously not be used anymore within a short time anyway.
> 
> I invite you to read my original post here:
> http://marc.theaimsgroup.com/?l=linux-kernel&m=107715308608606
> 
> I explain the problems of the current interface and my goals with the
> new one. If you can think of a better solution to the problem, please
> speak up.

Working from the premise that there is a current (old-style with mostly 
chip dependent code), libsensors has 2.4 /proc support, and each 
specific release supports one of 2.6.[0123]...

I'm glad I'm not the maintainer of libsensors for a distribution.  Since 
you have effectively pushed the compatibility work to them.  Just think 
of angry customer complaints about this. :(

Since there is going to be an effective libsensors-new library with chip 
independent code, I suggest you put the compat code into the old library.

Mike