sizeof feature

[utzoo!decvax!ucbvax!ihnss!mhtsa!research!dmr]

been officially unsigned.  I believe this was documented with the distribution,
although it is not in the C book.  Unfortunately (some would say fortunately,
I gather),  I have not yet prevailed on the proprietors of any PCC
compiler, including that for the Vax, to go along.  The reasoning was that it
is easy, as utzoo!henry found, to declare objects whose size, on a 16-bit
machine, looks negative when interpreted as an int.  I felt that the least
painful way to deal with this was to make the expression unsigned,
rather than ignoring the issue altogether or making sizeof either
long or short depending on its operand.

In practice, the change caused us trouble only in the one case pointed
out by utah-cs!lepreau, namely
	(long) -sizeof(...)
It may look canonical, but as I recall it occurred only in one place
(though that place was init.c!)

Finally, the code sequence produced for a[i][j] %= k  is indeed wrong
in the distributed v7.  I can no longer point to the offending line,
but I'm pretty sure it is an obvious lack of a 'B1' in the code
tables for %=.
		Dennis Ritchie