(unsigned)char again

[utzoo!henry]

I should have been clearer about a couple of points in my recent
notes about (unsigned)c in net.bugs.v7.  The interpretation of casts
as semantically equivalent to assignment to a temporary is straight
from Dennis Ritchie (as of Santa Monica USENIX two months ago) and
is therefore the Official Word.  To my mind this supersedes Hans
Spiller's paper.  Dennis also does not seem to agree with Hans's
idea of always making char signed on all machines;  what Dennis said
in my presence was about what I'd always thought his view was, i.e the
signed nature of char on the 11 is a lamentable historical accident
that should not be blindly imitated.

[Dennis, I apologize if I am misrepresenting your views.  I'm working
from notes made in the evening about conversations during the day.]

Dennis was not sure whether the definitive interpretation of casts
had made it into the 3.0 compiler.  Hans's paper indicates it didn't.

Who's right?  I dunno.  Making casts systematically equivalent to
assignment seems a good idea to me;  it gives them an exact definition
for the first time and eliminates needing two sets of conversion rules.
I do wish my compiler had unsigned char, though.  Whether char should
be considered signed in future implementations is less clear.  My own
view is that they ought to have been unsigned from the beginning, and
given that they weren't, any code which relies on EITHER interpretation
is wrong and unportable.

If there is interest in further discussion of this, please mail to me
and I'll summarize for the network.

					Henry Spencer
					decvax!utzoo!henry