Fix critical compound assign bug.

[HPCguy]

Without the fix on this branch,
the following code currently
adds 2 to the je ptr, instead of 8.

int *je = 0;
int n = 2;
je += n;

[jserv]

I wonder if compound-op bypasses literal const optimizations still make sense.

[HPCguy]

The old way was taking shortcuts while this technique is a "full parse" using the full standard AST rules. It is "complete" while the old way skipped steps. scripts/disasm can be used to look at the assembly code, so that you can verify.

[HPCguy]

int main()
{
int *je = 0;
je += 4 - 2;
return 0;
}
% amacc -si test.c
ENT 1
// int *je = 0
LEA je (-1)
PSH
IMM 0
SI
// je += 4 - 2
LEA je (-1)
PSH
LI
PSH
IMM 8 // 2 * sizeof (int)
ADD
SI

[HPCguy]

int main()
{
int n = 5;
n *= 4 - 2; // should be n = n * (4 - 2), not n = (n * 4) -2
return 0;
}
% amacc -si test.c
ENT 1
LEA n (-1)
PSH
IMM 5
SI
// n *= 4 - 2;
LEA n (-1)
PSH
LI
PSH
IMM 2 // correct!
MUL // correct!
SI

[jserv]

Thank @HPCguy for the effort!