How to initialize and use a C11 _Atomic variable?

Question

Is the following code allowed?

_Atomic(unsigned int) a = 1;
if (a == 0) {

}

The C11 spec (n1570) says at 6.3.2.1p2:

if the lvalue has atomic type, the value has the non-atomic version of the type of the lvalue.

So this seems to say it's ok.

Perhaps this question is your answer: http://stackoverflow.com/questions/10668941/c11-grammar-ambiguity-between-atomic-type-specifier-and-qualifier — Tony The Lion, Mar 10 '13 at 01:02
ISO/IEC 9899:2011 §6.7.2.4 _Atomic type specifiers_ indicates that your declaration is probably OK. §6.7.4 _Type qualifiers_ includes an example: `_Atomic volatile int *p;` without the parentheses, so the parentheses are not always necessary. — Jonathan Leffler, Mar 10 '13 at 01:09
Well, it's more the reference to "a", whether that's an ok way to load the atomic value. — joey, Mar 10 '13 at 01:16
AFAICT, yes — but I've not actually tried it...have you got a C11 compiler to try it on? — Jonathan Leffler, Mar 10 '13 at 04:30

Jens Gustedt · Answer 1 · 2021-10-04T10:04:19.000

7

~~No, initialization like that is not ok. You'd have to use ATOMIC_VAR_INIT to initialize an atomic object. From C11 7.17.2.1:~~

The ATOMIC_VAR_INIT macro expands to a token sequence suitable for initializing an atomic object of a type that is initialization-compatible with value. An atomic object with automatic storage duration that is not explicitly initialized using ATOMIC_VAR_INIT is initially in an indeterminate state; however, the default (zero) initialization for objects with static or thread-local storage duration is guaranteed to produce a valid state.

Otherwise the object would be in a valid state, but "indeterminate" so you wouldn't know which value it has.

The state of this has changed with C17 which removed the requirement to initialize with ATOMIC_VAR_INIT. Now doing an initialization as presented in the question is ok and the right way to go.

As someone suggested, another possibility is still to do a dynamic initialization with atomic_init, but classic initialization is certainly to be preferred wherever you may.

edited Oct 04 '21 at 10:04

answered Mar 10 '13 at 07:49

Jens Gustedt

76,821
6
102
177

Thanks, I shouldn't have include the "= 1", I didn't want people to complain about using uninitialized values! (then I forgot to use the macro) – joey Mar 10 '13 at 23:09
And wouldn't `atomic_load` be necessary as well at `a == 0`? C++ always does it through `operator T()`. – Ciro Santilli OurBigBook.com Jun 17 '15 at 07:57
@CiroSantilli六四事件法轮功纳米比亚胡海峰, no, all operators work for atomic variables. They have sequential consistency per default. – Jens Gustedt Jun 17 '15 at 09:54
I couldn't find that very clearly in C11, what part describes it? – Ciro Santilli OurBigBook.com Jun 17 '15 at 10:03
1

@CiroSantilli六四事件法轮功纳米比亚胡海峰, it is 6.2.6.1 p 9 – Jens Gustedt Jun 17 '15 at 11:44
`ATOMIC_VAR_INIT` is deprecated in C17 and slated for removal in a future standard. Use `atomic_init()` instead. – Kevin Thibedeau Oct 03 '21 at 19:25
@KevinThibedeau, no `atomic_init()` is only suboptimal because its dynamic. Plain initialization of atomic objects is ok since C17. See my updated answer. – Jens Gustedt Oct 04 '21 at 10:05

How to initialize and use a C11 _Atomic variable?

1 Answers1