Matthew Dillon wrote:
> I haven't done anything. Well, keep messing around with it and report
> if the problem reoccurs. I suspect the bug is still there somewhere.
Sigh. You're right. FWIW I collected some data on the errors -- I hope
someone gets an idea from it:
First, the 'MAKEDEV all' target has more than forty lines in it, but all
of my errors occurred in just the first two lines:
all)
sh MAKEDEV std # standard
sh MAKEDEV fd0 fd1 # bdev, floppy disk
Each device in the 'std' group was created with perms of 600, which
happens to be correct for the first device (console). Now, if I
just do a mknod by hand the default perms come out 644, so I conclude
from this that MAKDEV did actually change the perms on *each* device
to 600 after creating it as 644. Does this sound right?
The errors were different in the second group, the floppies:
each device of the type fd0.1200 (for example) was created with the
wrong ownership (root:wheel instead of root:operator) but the perms
were correct.
This is one of the commands that creates these devices:
# User-readable and programmer-readable name sets
mknod ${name}${unit}.1720 c $chr $(($unit * 64 + 1)) \
root:operator
Note the change of ownership appears on the next line and it seems
to have been ignored. Dunno if that's just coincidence or not.
Wish I could find a way to create the problem reliably, but it
seems entirely random so far. Seems almost like sh goes insane
temporarily :o/