[Jason suggests on-the-fly initialization of dynamic bpf buffer limits]

... technically, sure.  But this was in context of repeated requests
(perry, smb) for dyamically computing a whole range of parameters.

One or two parameters, maybe the ad-hoc thing isnt so bad.  A
half-dozen or a dozen or more? No thanks, that way lies a morass of
ad-hoc ``have I been initialized yet?''  flags. Which leads
inevitably, to maintenance bugs, when someone updtes some
but not all of the constructors, or adds new uses  without an
``am I initialized yet? check.   YMMV, but thats my experience.


I happen to think Perry's right: we could do with an initialization
framework.