Nathan J. Williams wrote:

> Please note that while adding sysctl support or another interface for
> accessing live data is fine, it is not fine to *remove* the support
> used to access crash dumps. This means that they must continue to be
> able to grovel kernel memory, even if it's not the primary mode of
> operation.

... which suggests they need either quite formalized regression testing,
or to be used in that mode often enough to ensure they don't rot--that's
the other side of the 'migrate to sysctl' equation, having to actively
maintain two separate interfaces.

-Chap