IMO, the energy spent on this would be better spent making programs not
   access kernel memory directly.


savecore has extra problems over normal kvm programs - it wants to copy
the kernel program to the crash directory as well as the core file, so
it needs to know where to copy it from.

ideally it should use /dev/ksyms instead of /netbsd for name-lookup but
right now that would require at least some interface change -- the -N
option is used for both namelist and copying.


i agree with the sentiment on making programs use well-define interfaces
but how does that related to a @booted_kernel magic string?  isn't it
just a handy thing for people who want it, ignored by the rest of us,
like the rest of the magic strings?  i'm sure i can think of really
horrible uses of those too, but we don't reject them for that.





.mrg.