[opensolaris-summit] My comments (very subjective) on proposed Summit topics

[Philip Brown]

Putting my dusty "Driver developer" hat on for a second.. .:-)



On Wed, Sep 26, 2007 at 12:55:42AM -0700, Alan DuBoff wrote:
> ...
> I feel the testing needs to be evaluated in some cases. For instance, 
> testing a NIC driver to be loaded and unloaded thousands of times will 
> probably just tell you if there is a memory leak. There could be better 
> ways to test that, IMO. Loading and unloading a driver is not a common 
> practice on most systems.

Actually, that sounds like a GOOD TEST to me, that should be kept in!
I've occasionally had bugs in drivers of mine, that really only
 manifested themselves in a comprehensible way, at driver load time.

Very rare ones, to be sure. But then, that's a core component of driver
testing: working out all the "very rare" cases, and making them
bulletproof.