alan@bk.dk (Alan Rooks) (04/14/88)
A short while ago, Chris Torek posted an informative summary of the value-
preserving/signed preserving question, and how it is in dpANS. He was of
the opinion that the sign-preserving rules were easier to learn, and the
"benefits" of value-preserving were marginal.
Doug Gwyn (I think) replied that sign-preserving rules weren't all that
simple, and besides, very few programs would be affected by using one or
the other model.
So it seems that one or the other is a little better/worse to use, and one
or the other is a little easier/harder to learn. But is this the issue? As
far as I can tell, using value-preserving rules changes the specification
of C in a fairly large way, not so much from K&R, which is out of date, but
from its successor C reference manuals. The first rule in the "usual
arithmetic conversions" section of the 3B2 System V Release 3 C Reference
Manual states
First, any operands of type |char| or |short| are converted to
|int|, and any operands of type |unsigned char| or |unsigned
short| are converted to |unsigned int| ...
This is clearly the sign-preserving rules, and will change in a major way
in dpANS (please correct me and ignore the rest if this is wrong - I don't
have a copy of the draft standard).
So if we have only a very slight change/improvement, is it worth changing
the language definition? It seems to me that if X3J11 isn't getting something
major for a change, then it shouldn't change! (and maybe it shouldn't
change even if they do get something major).
Of course, maybe I have too much of a UNIX-orientation. Do other non-Unix
compilers use (and define that they use) value-preserving rules?
I am interested in this because I am writing a translator for a C-like
language, which will translate to C. The C-like language will have the
same conversion rules as C (essentially), and thus my translator will
have to have the "usual arithmetic conversions" built in to it. I don't
want to have to change all of this later! (wishful thinking, I guess).
Alan Rooks - Bruel & Kjaer Copenhagen - ...!mcvax!bk!alan or ...!dkuug!bk!alangwyn@brl-smoke.ARPA (Doug Gwyn ) (04/15/88)
In article <36@bk.dk> alan@bk.dk (Alan Rooks) writes: >Do other non-Unix compilers use (and define that they use) value-preserving rules? I joined X3J11 after the original decision, but I was told that indeed some existing compilers used value-preserving rules. In any case it is certain that many variants of PCC got the rules wrong no matter which alternative was intended, so something needed to change in PCC at least. This is really a dead issue; even AT&T has given up on trying to get sign- preserving rules into the standard (although occasionally it is mentioned as an example of a "sacrifice" agreed to in order to reach consensus).