instance counter

[Moshbear dot Net]

Given
class Fred {
public:
Fred() {
ctr = ++xctr;
}
....
private:
unsigned ctr;
static unsigned xctr; // initialized via boost::thread::run_once
....
};
,
what would be the best way to make the increment-and-read of xctr
atomic so that it is thread-safe?
I'm using xctr to track instance number of class so that I can use
operator<(const Fred&) to compare by instance-creation time.

 

[Chris M. Thomasson]

Given
class Fred {
public:
Fred() {
ctr = ++xctr;
}
...
private:
unsigned ctr;
static unsigned xctr; // initialized via boost::thread::run_once
...
};
,
what would be the best way to make the increment-and-read of xctr
atomic so that it is thread-safe?

Perhaps an atomic fetch-and-add operation, or a statically initialized
PThread mutex.

I'm using xctr to track instance number of class so that I can use
operator<(const Fred&) to compare by instance-creation time.

What happens when that "version counter" wraps around?

;^)

 

[Moshbear dot Net]

Perhaps an atomic fetch-and-add operation, or a statically initialized
PThread mutex.

Is there a portable way of doing
atomic {
++xctr;
ctr = xctr;
}

Also, it looks like I'm going to have to search the boost docs for
mutex-locked statically-scoped counters.
However, I believe that somebody has already done this and can offer
guidance as to which boost class is
preferred (or c++0x/tr1 class, given that this is c.l.c++

What happens when that "version counter" wraps around?

I'd be using ::uint64_t in actual code. This was just to show the
concept.

 

[Chris M. Thomasson]

std::atomic<int> xctr = 0;

[...]

Want it NOW!

http://www.stdthread.co.uk

:^)

 

[Moshbear dot Net]

On 2011-06-11 12:58:12 -1000, Moshbear dot Net said:

std::atomic<int> xctr = 0;

[...]

Want it NOW!

http://www.stdthread.co.uk

Any F/OSS versions? boost::interprocess::detail has atomic operations,
but it looks a bit ugly to use in end-user code.
And <atomic> is c++0x (existence in: g++ 4.5 and higher, likely in VC
1x (not in VC10)).