matthew green writes:
   >i admit.  i'm completely stumped by anyone wants to put documentation
   >into wired kernel text.  the dynamic nature is completely irrelevant.
   >any subsystem that creates sysctl nodes should document them.  that's
   >where the documentation is.  it's not like people distribute binaries
   >without documentation.
   
   But where do we draw the line?  At a handful of sysctl nodes or by reams
   of PCI, PCMCIA, CARDBUS, EISA, ISAPNP, MCA, and USB vendor and device ID
   strings?


the only sort of "verbose" documentation in the kernel that i like
is the SCSIVERBOSE-style ones -- they report real actual errors to
me in human terms.  none of the above really count in this class..

it would be nice if we had some way to generate manuals with the
various driver verbose information... that's probably not very hard
even.  i think i'll add that to my todo...


in addition to the above list, individual drivers often have no
option to not include all their local verboseness.  eg, all the hme
attach routines that print "Sun Happy Meal Ethernet".  why?  isn't
that in hme(4)?


i guess the major difference between all these instances and sysctl
is that sysctl is a software concept -- something we created and
maintain unlike hardware.  (i'm not sure how relevant that
distinction is to kernel text as documentation.)