Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!watnot!watmum!rmariani
From: rmariani@watmum.UUCP
Newsgroups: comp.sys.amiga
Subject: Major bug in Aztec C V3.20
Message-ID: <716@watmum.UUCP>
Date: Thu, 4-Dec-86 00:21:18 EST
Article-I.D.: watmum.716
Posted: Thu Dec  4 00:21:18 1986
Date-Received: Thu, 4-Dec-86 04:42:23 EST
Distribution: net
Organization: U of Waterloo, Ontario
Lines: 58


Watch out for this one folks... it bit me pretty hard


In a for loop where the initialiation is constant

e.g.

	for (i=0;i<200;i++) { ... }

The good folks at Aztec do the initial check at compile time and
determine that the test can be made at the end of the loop thus
saving a branch.

The machine code looks like this


	set i=0

LOOP
	do the code

	increment i
	compare against 200
	branch if less than to the LOOP


This is swell... however it isn't smart enough, on a loop like this

	for (i=200;--i>=0;) { ... }

It generates code of the same form!!!!!  but that means that the
first time it enters the loop  i=200!!!!!  That's wrong!  They're
gonna have to be smart enough to take side effects in the conditional
into consideration.

If you want to see the exact problem just compile the following:


main()
{
	int i;

	for (i=20;--i>=0 ;)
		printf("%d\n",i);
}

And note the results (20 is printed, it shouldn't be), then compile again with
the -A and -T options and look at the assembler it generated

Is there an Aztec beta tester listening???  Please, please have them fix this
for version 3.40.  Let me know if you read this and you're someone who can do
something about it... tell me who to write to... anything...


	-Rico

	...watmath!watmum!rmariani