Re: [boost] BOOST_ENUM proposal

From: Sebastian Redl (sebastian.redl_at_[hidden])
Date: 2005-12-10 07:28:59

Yuval Ronen wrote:

>One more thing: using static data inside a function usually introduces 
>thread-safety issues. I'm not sure if boost::array suffers from this, 
>but I think it worth a check. If there is a problem, you might want to 
>switch to native C++ arrays, or get some help from Jason's singleton...
>  
>
First, we have to remember that boost::array pretends to be a POD type 
by not having any constructors or destructors. This means that a 
compiler might allocate the static in preset storage, thus removing all 
thread safety issues. I don't know if it's required to do so, though.
If it doesn't, then the constructor for boost::array must be called and 
there are two scenarios.
1) The compiler generates a first_entry flag and sets it immediately 
upon entering. In this case, a thread might attempt to use the array 
before it's finished constructing. Access violations for the strings 
would be the result.
2) The compiler generates a statics_constructed that gets set after all 
constructors have finished. In this case, two threads might both try to 
construct the array. The trivial nature of boost::array means that there 
are no issues with this.

I don't think any sane compiler would go for option 1, but I can't know 
for sure.

Can anybody with the standard resolve the POD issue? If the compiler is 
required to allocate the array in preset storage, there are definitely 
no issues.

Sebastian Redl