> It doesn't. As Section 3.2.1 of the Brussels design doc says, the 
 > property names are prioritized as:
 > 
 > 1. first search public props
 > 2. if not found, try the wifi props

Aren't the WiFi properties public?

 > 3. if not found, try private props.
 > 
 > The idea is to treat Private props as experimental, second-class citizens.
 > A driver that wants to venture in this area should first make sure
 > there's no name-space collision with public props (which will be 
 > fully documented in the dladm(1m) man page). 

... and if a public property with the same name is subsequently added, it
becomes impossible to access the private property?

 > I'm not averse to it, but adding the link_ prefix for Private props (or
 > vice-versa) places an artificial burden on Public props - for example
 > many MII props are already prefixed with "link_". If we choose to make
 > these Public ethernet props (as Brussels has done) we'd now have
 > to strip off the "link_" prefix.

We don't need to use that prefix.  A prefix just seemed simpler to me than
adding a formal namespace abstraction -- and I don't know how you can
address the issue I raised above without some sort of namespace isolation.

 > In the general case, if a driver comes up with some prop like "link_foo"
 > that ends up being a good candidate for a public prop, we'd have
 > to go through the exercise of stripping off the link_ to promote it.

But are we really likely to promote things as-is?  I think it's more
likely that N different drivers will all add their own version of
link_foo, with slightly different names and semantics, and we'll need to
add a public version, unified version with a name and semantics that
accommodate all of those N private properties.

-- 
meem

Reply via email to