061297: bugs in vfs

Hello,

061297 vfs seems to work alot better! I have been able to get
through several iterations of compile/builds without losing
any blocks, though I seems to have managed to stay within
a single freelist sector, which is fine with me :) I'lll get
to unpacking the other tars as well and fill the thing up.

I did notice however that the response of the system as a whole
seems to be more sluggish. Startup is somewhat slower and response
in different vtys is, well, 'sticky'. It's kinda hard to explain..
I'll try to quantify it more this wk.

I had a few comments about the src tree:
  srv/mach/fd, srv/mach/scsi, bin/init needs -lsrv in LDFLAGS.
  srv/mach/joystick stuffers from coming from a dosfs; libjoystick.c
and libjoystick.h get truncated filenames so the makefile dies.
  ls -l output format isin't conducive to pipelining, and I personally
find it hard to read.

I was wondering what negative results might come from doing something
like
  mount fs/root:vsta /
all the time, to construct a union mount in / of the dos stuff and the
vsta stuff. The intent is to establish a Plan9ish consistency,
with /lib, /bin, /include etc. and writing the src tree to assume
these are the locations of stuff it needs. This mount would eliminate
a depth dependency that exists in the src tree (all the -I/-L../../../
stuff) and make 'mkall install' work regardless of where the src tree
resided, and more easily support more architectures.

  I haven't run into any trouble with this mount even though
it hides whatever bin/lib/include dos junk is beneath it, but I
have very little DOSstuff :)

Sincerely,

Eric Dorman
edorman@tanya.ucsd.edu
Received on Mon Jul 14 10:17:34 1997