At Wed, 9 Apr 2003 18:38:44 +0000 (UTC), "David Laight" wrote:
> The problems start with naming the fs!

uh, really?

> The obvious thing would be to have:
> 	ufsv1	ufsv1 support only
> 	ufsv2	ufsv2 support only
> 	ufs	v1 and v2 support
> (done in a similar way to the current lfsv1/2 build)

Well, right now the boot blocks themselves are often called "*ffs*"
right?  (despite the fact that the code uses ufs* names internally.)

So, keep the existing ufs names.  add ufsv2 names which handle ufsv2.

then just add ufsv2 entries to boot blocks' file_system arrays as
appropriate.

I don't see why you need a single piece of object code that handles
both.  (sure, compile them from the same sources if you want w/
different #defines, but why insist on cramming them into the same .o?)


Obviously, machines that have an "ffs"-specific boot block (typically
w/ a file_system array referencing only ufs_* 8-), should get new
ufsv2 boot block if they're space constrained.


> I want to be able to build a single copy of libsa and libkern for
> multiple fs types.  Most of it isn't used and compiling all of it
> for every fs variant is a waste of disk space and cpu cycles.

You can already do that.  See e.g. the alpha boot blocks, the 'boot'
dir.  (same w/ lots of other machines.)

Yeah, compiling libsa/libkern lots of times is a bummer.  However,
that's completely orthogonal to the problem at hand.



cgd