Comments on: Prefer prefix operators over postfix

By: ford prefix

ford prefix — Thu, 24 Apr 2008 22:54:53 +0000

The real reason to use prefix rather than postfix is because when you have a choice,
you should write exactly what you mean. When
you write postfix, you are communicating that
you need the old value after the increment. If you don’t need the value, why in the world wold you write it that way? If you don’t need the old value, use prefix.

Would you leave statements such as this:

{
int f = 0;
// …
f + 1;
f - 1;
// use f for real purpose
}

laying around in your code?

Say what you mean, mean what you say.

By: Bennett

Bennett — Fri, 29 Jun 2007 04:32:32 +0000

Good stuff, Sam. I just talk about what compilers might do, but you go the extra mile and find out. Nice to know that gcc optimises simple loops. But I also said:

In this case, of course, the compiler will probably realize that the value of the expression is not used, and can therefore avoid the copy.
…
Even if i is an integral type today, somebody might change it tomorrow to be some relatively heavyweight class type like an iterator: suddenly, you get a new object created and destroyed every time through the loop.

And don’t forget about visibility and consistency too, which in most cases are more important than efficiency anyway.

By: Sam Danielson

Sam Danielson — Thu, 28 Jun 2007 11:30:03 +0000

Here’s the objdump -d a.out for a postfix loop with a simple body. I compiled this with with gcc 4.1.2, with the -ggdb option.

The machine isn’t storing a copy of i, it just puts the i++ after the body of the loop.

for (i = 0; i
804835e: 8b 45 f4 mov 0xfffffff4(%ebp),%eax
8048361: 89 45 f8 mov %eax,0xfffffff8(%ebp)
8048364: 83 45 f4 01 addl $0×1,0xfffffff4(%ebp)
8048368: 83 7d f4 09 cmpl $0×9,0xfffffff4(%ebp)
804836c: 7e f0 jle 804835e