Text 41237, 155 rader
Skriven 2006-10-13 21:23:02 av Michiel van der Vlist (2:280/5555)
  Kommentar till text 41213 av mark lewis (1:3634/12.0)
Ärende: smart compilers (was: pi)
=================================
Hello mark!

On Friday October 13 2006 11:37, you wrote to me:

 MvdV>> I have done quit a bit in Pascal programming, but that was a
 MvdV>> long time ago. Before the  IBM PC (clones) dominated the
 MvdV>> market. My compiler went to the museum together with my home
 MvdV>> brew 8 bit system running 6809 FLEX. Haven't written a line of
 MvdV>> Pascal in over fifteen years. So maybe someone will point out
 MvdV>> that it will not complie at all because something is missing...

 ml> don't look at "the language"... look at the code... t'is a mismash of
 ml> at least three...

I would perhaps have been better if you had chosen a specific language. That
way we would not have to make assumptions on how exactly it interprets this
"code". Or in what order it parses...

 ml>>   if (2+2==5) and (3+3==7) then print "EGADS!";

 ml>>   if (2+2==4) or (3+3==7) then print "whew!";

 ml>>   if (2+2==5) or (3+3==6) then print "Whew!";

 ml>>   if (2+2==4) and (3+3==6) then print "WHEW!";

 MvdV>> All these examples are variations of one ande the same thing. A
 MvdV>> constant between the if and the then. The constant is either
 MvdV>> TRUE or it is FALSE. If it is true the print statement will be
 MvdV>> compiled and executed when the object code is run. If it is
 MvdV>> FALSEe, the smart compiler will skip it so nothing will ever be
 MvdV>> printed.

 ml> ahhh... you see one constant where i see three... a large one made of
 ml> two seperate smaller ones... the large one being made of the "and" and
 ml> "or" between the two smaller ones...

An expression made up of only constants is itself a constant. I was referring
to the (boolean) constant that is the result of evaluating the expression
between the if and the then,


 MvdV>> You want me to do the leg work? Ok...

 MvdV>> EGADS! : skipped
 MvdV>> whew!  : printed
 MvdV>> Whew!  : printed
 MvdV>> WHEW!  : printed

 ml> hey! you didn't take the short-circut like you seemed to be doing
 ml> before... otherwise, you'd have skipped the "Whew!" because the
 ml> evaluation of the first was false and short-circut boolean evaluation
 ml> would have jumped out right there assuming the whole thing was
 ml> false...

Skipping after the first term would have been wrong as it is followed by an or
and what is after the or can change the FALSE into TRUE.

Mind you: in my original example I was just saying that if what is between the
If end the then evaluates to FALSE, then it is irrelevant what follow the then
and it can be skipped.

You are now narrowing it down to the individual terms in the expression bewteen
the if and then but then things become a little bit more complicated. I never
meant to imply that applying short circuit boolen evaluation *that* way is a
good way to handle things.

 ml>> now, try these...

 ml>>   a=2; b=2; c=5; d=3; e=3; f=7;
 ml>>   if (a+b==c) or (d+e==f) then print "egads!";

 MvdV>> Ah, that is a bit more tough. If it is a *very* smart
 MvdV>> optimising compiler it will be aware that although the terms in
 MvdV>> the if statement are written as variables, the actual values
 MvdV>> are already known at compile time and so the compiler will
 MvdV>> handle it the same as when the actual values were substituted:

 ml> yeah, but we're not studying compiler design so please... don't let
 ml> that bother your noggin', eh? each of those variables could have been
 ml> grabbed by some randomizing function... but i had to have some way of
 ml> presenting the scenerios to you...

Well, then it would have been better to say so instead of just presenting them
as hard coded in the source....

 MvdV>> if (2+2==5) or (3+3==7) then print "egads!"

 MvdV>> The compiler will skip it and nothing will be printed

 MvdV>> If it is a less smart compiler, it will treat the terms in the
 MvdV>> if statement as true variables - like the ones that are only
 MvdV>> entered at run time - and generate the complete object code
 MvdV>> needed to handle the general case.

 ml> yes...

Ok, lets us take it from there. a..f are true variables.

 MvdV>> Still, when the object code is run, it will evaluate, the
 MvdV>> condition as FALSE (2+2 is not equal to 5 and 3+3 inot equal to
 MvdV>> 7) and it will not look what is after the then and just skip to
 MvdV>> the statement following the print "egads!"

 ml> and yes...

So far so good.

 ml>>   if (a+b==c) and (d+e==f) then print "EGADS!";

 MvdV>> In this example, when the object code is run, is will not
 MvdV>> evaluate the (d+e==f) part as the condition has already failed
 MvdV>> on the first part.  Whatever is after the and can never make
 MvdV>> the condition come TRUE, so no need to waste time evaluating
 MvdV>> it.

 ml> you're sure? there's several assumptions being made... one is that the
 ml> first statement is on the left ;)

A reasonable assumption. In fact I do not know of any language using this
algebraic notation that does not evaluate left to right. If you ment to use an
exotic language that does it different and still uses this standard algebraic
notation, you should have said so in advance...

 ml>>   c=4;
 ml>>   if (a+b==c) or (d+e==f) then print "whew!";

 ml>>   c=5; f=6;
 ml>>   if (a+b==c) or (d+e==f) then print "Whew!";

 ml>>   c=4; f=6;
 ml>>   if (a+b==c) and (d+e==f) then print "WHEW!";

 MvdV>> I will leave it to the reader to work out the rest, I think I
 MvdV>> have explained the basics.

 ml> yes, the basics as well as some shortcuts that can bite at not so good
 ml> times...

I don't think I have used any such shortcuts. Maybe you misunderstood what I
was trying to convey.

 ml> witness bill gates' numerous demonstrations of various windows
 ml> platforms in front of hundreds of people :0 :) :) :)  witness,
 ml> also, how many bugs have lead to some very severe security holes in
 ml> those same windows platforms...

I am not Bill gates. My programmes, at least the ones I wrote while having full
control over the results do not have holes like that.


Cheers, Michiel

--- GoldED+/W32-MSVC 1.1.5-b20060315
 * Origin: http://www.vlist.eu (2:280/5555)