> The "raw partition" has always, for as long as I've used BSD,
> meant "the entire disk".  When disklabels appeared, this didn't
> change.  In order to make it work as "entire disk", you have to
> just ignore what the disklabel says are the start/size of that
> raw partition.

I wasn't getting all riled up about the definition of "raw partition", but
rather the revelation by cjs that 'sd0c' ignores the contents of its entry
in the disklabel. What little documentation I have found on the subject over
the years implies that partition 'c' must be left as the whole disk or you
will regret it. If that's no longer true, great, but once again I am left
a little flummoxed by the fact that it wasn't documented or FAQed enough
for me to have run into it before -- I generally read anything on these
topics that comes my way, precisely because it is so sparsely documented...

> Note that on all platforms, RAW_PART == c, except i386, where it is d.

Yep, that is gross, no question about it.

> The i386 is a historical nightmare inherited from Jolitz, in this regard.
> One of these days it might even get fixed if anyone were to follow through
> with their proposals on fixing it, but that's a flame war for another day :-)

On this score I most definitely agree with Curt, that if we are to fix it,
it should be done properly so we don't have to revisit the issue again for
a VERY long time.

> "Any disk driver."  src/sys/dev/scsipi/sd.c, sdstart() is a good spot
> to see this in action.

Thank you. Auditing the uses of RAW_PART in that file was most enlightening.

> (Note, that pointing to a data structure is sort of meaningless here...
> the name "raw partition" is pretty self-describing, if you ask me :-)

Sorry, I was really trying to say "greppable expression" -- if you tell me
just enough information for me to put an editor on the relevant lines of
code, that's the most I could ask for.

Todd Whitesel
toddpw @ best.com