[OpenAFS-devel] fs sysname

Ted McCabe ted@MIT.EDU
Thu, 18 Oct 2001 15:01:41 -0400


At 7:21 PM -0400 10/14/01, Derrick J Brashear wrote:
>I prefer the former but could be pushed the other way. Since Ted McCabe
>implemented this originally, perhaps he has some comment?

At the time I made the change, I was unaware of the documentation 
that Harald references - I would have suggested changing such at the 
time had I been aware.

It is my preference to simply update the documentation and not add an 
additional pioctl.  The justification I see for not adding a pioctl 
is perhaps due to lack of imagination: I can't envision a situation 
where it would provide a tangible benefit over simply 
modifying/extending the definition of the existing pioctl.

The downsides I see in redefining the pioctl, as has been implemented, are:
1) If one tries to set a list on a client that doesn't support lists, 
the caller of the pioctl can only determine that the list isn't set 
by using the pioctl to get the current list, not by some error on the 
initial call.  I do see this as simpler than trying some new pioctl 
specifically for lists and falling back to the old if the new pioctl 
isn't supported, but only marginally so.
2) Some automated agent that uses the output of fs expects the 
sysname to be a single element and gets a list - Harald's suggestion 
makes that easier to resolve.  Since the agent should expect to 
possibly need updating if the client involved gets an updated kernel 
mod, I don't think this is an issue.

So my current thought is that unless there is a tangible benefit I'm 
not aware of, the easiest solution is simply to update the 
documentation and be done with it.

    --Ted