PSARC 2007/601 FastTrack timeout 12/05/2007 - "spec.txt" added to the materials directory

[Garrett D'Amore]

Artem Kachitchkine wrote:
>
> Like I said, interface stability only applies to interfaces. /devices 
> is not an interface.

If /devices is "not an interface", even an ON-private one, then lots of 
things that use it are doomed. :-)  I never understood that that 
/devices wasn't suitable for use within ON.  (I do understand that some 
day in the future it could change, so its not appropriate for unbundled 
software.)

On the other hand, stating that /devices isn't an interface seems, to 
me, akin to putting ones head in the sand.  Its been around, unchanged 
for so long, with the same basic relationship, that I think its almost a 
"de-facto" interface.  I can probably name half a dozen separate 
projects that rely on it.  And its exposed to administrators.  If the 
devfs team really believes that nothing should use those device nodes in 
/devices, then we should just go ahead and get rid of them (or make the 
ddi-internal-pathname only components...) and force everyone to use 
/dev.  Of course, then fallout from such a change would be giganormous.  
Every cfgadm plugin, libdevinfo, etc. relies on /devices paths, and the 
fact that readlink() on /dev nodes is a way to determine the position in 
the devinfo tree.

I should probably have you review my sdcard cfgadm plugin, which does 
just this.... modeled after the sata cfgadm plugin, of course.

    -- Garrett