Bug#278081: gcc-3.3: REGRESSION: Doesn't follow precedence

[Adam Majer <adamm@galacticasoftware.com>]

Package: gcc-3.3
Version: 1:3.3.5-1
Severity: important

IMHO, this bug should be grave since it has a potential for
breaking a lot of software, at least anything that 
depends on things like stack operations. That is

stack<int> A;
A.push(5);
A.push(2);

// do a subtraction of 5-2 and push resuls onto stack
A.push( -A.pop() + A.pop());

What it does, it actually A.pop() - A.pop() which is very wrong.

I can reproduce this problem on both gcc 3.3 (default) and gcc 3.4. 
gcc 2.95 in woody compiles this code correctly, hence this is a regression.

You can rearrange -X+Y, as well as -X()+Y or -X+Y(), but you cannot do
this for -X()+Y() unless you can guarantee that X() doesn't depend on
Y() and vice-versa.

The following test case will also illustrate this problem. Hopefully
this regression doesn't translate to stable...

>>>>>>>>>>>>>    TEST CASE   >>>>>>>>>>>>>>

#include <stdio.h>
double A(){
	static int div;
	double r = 10.0/++div+div;
	printf( "A() returns: %f iteration %d\n", r, div );
	return r;
}

int main()
{
	/* EXECUTES A()-A() INSTEAD OF -A()+A() */
	printf( "-4.0 is %f\n", -A() + A());
}

<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
						 

-- System Information:
Debian Release: 3.1
  APT prefers unstable
  APT policy: (500, 'unstable'), (1, 'experimental')
Architecture: i386 (i686)
Kernel: Linux 2.6.8-1-k7
Locale: LANG=C, LC_CTYPE=C (charmap=ANSI_X3.4-1968)

Versions of packages gcc-3.3 depends on:
ii  binutils                    2.15-4       The GNU assembler, linker and bina
ii  cpp-3.3                     1:3.3.5-1    The GNU C preprocessor
ii  gcc-3.3-base                1:3.3.5-1    The GNU Compiler Collection (base 
ii  libc6                       2.3.2.ds1-18 GNU C Library: Shared libraries an
ii  libgcc1                     1:3.4.2-3    GCC support library

-- no debconf information