Ryan Gordon's Linux Game Porting Guidance

[bison]

What's the do-while loop in the macro on slide 17 for?

[mrugiero]

What's the do-while loop in the macro on slide 17 for?

Macros can't declare blocks of code in other way, I don't recall why, so it's a common hack to make a do-while(0) loop instead. As it's while(0), it only runs the first time.

[smitty3268]

What's the do-while loop in the macro on slide 17 for?

It's a semi-standard hack to allow more safely replacing existing functions with multi-line macros.

In particular, it ensures that the macro gets compiled as a single statement.

Take this code, for example:

if (COND)
foo();
else
bar();

if you replace foo(); with your MACRO(); call, it has to be exactly 1 statement in order to compile correctly. if your macro becomes 2 statements, the if statement will only hit the 1st one, the rest gets run unconditionally, and then the else below will break the compile for not matching up with an if statement anymore.

if your code always uses brackets, then i don't think it serves any purpose.

[Grogan]

Ryan Gordon is a hero.

I'm not a game developer (or much of a programmer at all) but I enjoyed the slides and comments, it was interesting and made good sense. I especially liked his do's and don'ts.

[bison]

Macros can't declare blocks of code in other way, I don't recall why, so it's a common hack to make a do-while(0) loop instead. As it's while(0), it only runs the first time.

That must be a really old C thing, before my time. (!)

I compiled it with

gcc -Wall -pedantic -std=c89 test.c

and it worked without the loop.

[mrugiero]

That must be a really old C thing, before my time. (!)

I compiled it with

gcc -Wall -pedantic -std=c89 test.c

and it worked without the loop.

That it compiles doesn't mean it works as expected. Someone else explained why it's done like that, they need to make sure the compiler treats it as a single statement.

[bison]

That it compiles doesn't mean it works as expected. Someone else explained why it's done like that, they need to make sure the compiler treats it as a single statement.

It worked as expected. I retained the block, sans loop construct.

[mrugiero]

It worked as expected. I retained the block, sans loop construct.

In how many compilers did you test it?

[stqn]

It worked as expected. I retained the block, sans loop construct.

AFAIK it shouldn?t compile, because you can?t write a block followed by a semicolon.

[smitty3268]

It worked as expected. I retained the block, sans loop construct.

If you retain the block without the loop, it will work, but you have to exclude the semicolon and it looks weird. It's very hard not to screw that up eventually, because someone will come along and "fix" the missing semicolon.

e.g.

//works
if (COND)
MACRO()
else
bar();

//fails because you can't declare if(COND) { ... }; else { } as valid.
if (COND)
MACRO();
else
bar();


Anyway, as discussed, the do-while(0) loop is only needed to help ensure corner conditions are satisfied. There are lots of ways the macro will work without it. It's only necessary to prevent some rare things from ever becoming an issue, especially since when that happens it can be fairly hidden from view and hard to track down. So it's just better to be safe than sorry as a general rule of thumb.

And there's no reason not to do it. The loop will always run exactly once, so it's trivial for compilers to optimize out any extra overhead. It's just a way to hack around a rough patch in the syntax of the language.