/, >>, %, and other ugliness...

[Scott Bigham]

In article <11529 at bellcore.bellcore.com> sjs at ctt.bellcore.com (Stan Switzer) writes:
>Well, I really hate to open this can of worms yet again, but in my
>experience whenever (for integer i) "i%4" and "i>>2" differ...

Errr... in my experience, i%4 and i>>2 will almost always differ.  They're
not even doing the same thing.

>, what you
>REALLY want is what "i>>2" does anyway (assuming 2's complement).

I'll assume (possibly incorrectly) that you meant i/4 when you said i%4
and answer below.

>Which is to say that you want the MODULUS operation rather than the
>REMAINDER operation.

Errr... The modulus operation IS the remainder operation.

>  I have seen MANY cases where this is true, and
>have NEVER seen a case where it was false.

(See my assumption above) Consider:

double r;
int i=5;
.
.
.
 r=i/4;

What I want is not what i>>2 does, nor, for that matter, what i/4 does.
What I want is what i/4 -should- do; ie. return r=1.25.  Yes, I know I can 
say r=(double)(i/4).  The irony here is that C can't make this simple
conversion, and yet will gladly convert a char * into a FILE * when I don't
even want it to.

>The PC/RT implements ... i%n to always
>return a value (0..n-1) for positive "n" (dunno about negative "n").

Then you're lucky.  All the C's I've ever used thought (-1)%5==-1.

>Please, do not argue that integer -1/2 should be -1 so that the sign
>is the same as -1.0/2.0 or because "by the laws of arithmetic" (-1)/2
>equals -(1/2).  These are pointless arguments that have no useful
>consequences as far as correct programs are concerned.

By definition, the greatest integer of x is the largest integer n such
that n<=x; then integer -1/2 is -1.  Do programs that use correct
math not count as correct programs?

>Anyone wishing to take up the other side of the argument must find an
>example of an actual situation where having -1/2 yield -1 is useful.

Actually, I'd MUCH rather have -1/2 yield -0.5...

>1) [0..3] represents [north, east, south, and west].  With the modular
>% operator, "dir = (dir-1)%4" means "turn left".  I first came across
>this in a program that drew Hilbert curves.

Be careful; as I mentioned above, some C's don't get this right.  This has
bitten me several times.

>2) Bit extraction:  To get the n'th bit from the current (char) pointer
>"p" (0 bit is low) use "bit = (p[i/BITSPERCHAR]>>(i%BITSPERCHAR)) & 1"
>This comes up in rasterization code often enough.  The usual solution
>is to jimmy it so that you avoid negative "i" (or just use >>3 and &7
>instead).

Yes, if you're doing bit manipulation, then bitwise operations are the natural
choice.  If you're doing arithmetic, than arithmetic operations are the
natural choice.

>You should never have to code "i>>3" when you mean "i/8".

Exactly.  At least three people have said this already.  If you mean i/8,
-say- i/8.

>  You should
>also never have to worry that a 45 cycle divide instruction is going
>to be used so that in case the quotient is negative you'll get the
>answer that someone thought you wanted instead of the one you probably
>really want anyway...

Errr... when I say i/-4, I want i/-4.  You seem to think I want something
different.

>Stan Switzer  sjs at ctt.bellcore.com

						sbigham
-- 
Scott Bigham                         "The opinions expressed above represent
Internet sbigham at dukeac.ac.duke.edu   me and everyone that agrees with me.
USENET   sbigham at dukeac.UUCP          If that includes Duke University,
...!mcnc!ecsgate!dukeac!sbigham       I'll be amazed."