> If the information is useful, it is probably better to use sigaction
  > with the SA_SIGINFO flag set, which is defined by POSIX.
  >

  Definitely - it used to be that not all architectures in NetBSD had
  ucontext.h (amd64 didn't for a while, I know) but now it looks like
  there is one there that emulates the uncontext stuff with siginfo (I
  think)

And an autoconf test for setcontext, rather than _BSD44_....

  It looks like coda is broken in netbsd-current too... I am getting a
  ENODEV for some reason when venus tries to mount /coda.  It looks like
  something has come adrift and I need to chase it down.  Also, I am not
  entirely sure why (possibly because things are broken) but it seens
  coda has disappeared from pkgsrc which is a bit of a worry.

Phil removed the bits from pkgsrc since they were really crufty, with
the notion that they should be restored with modern coda.  Coda
sources can't cleanly install jointly exhaustive, mutually exclusive
sets of files (client-install and server-install overlap), so hackery
is needed to make separate client/server packages.  I'm inclined to
just have one package, and someone who wants to save the disk space of
the server programs can split them.  Arguably the right thing is to
fix coda itself to be able to do partial installs cleanly, and then it
would be easy to have separate packages (common, client, server).

Jan is working on making lwp work; HEAD lwp fails on 2.0 and 2.99.15.

I'm succesfully (modulo unrepairable conflicts which are almost
certainly not a NetBSD issue) running ~6 month old coda on 2.99.15.