Re: RFC: C++ PATCH to support dynamic initialization and destruction of C++11 and OpenMP TLS variables

[Jason Merrill <jason at redhat dot com>]

On 10/05/2012 10:38 AM, Jason Merrill wrote:
> On 10/05/2012 04:29 AM, Richard Guenther wrote:
> > Or if we have the extra indirection via a reference anyway, we could
> > have a pointer TLS variable (NULL initialized) that on the first access
> > will trap where in a trap handler we could then perform initialization
> > and setup of that pointer.
>
> Interesting idea.  But I don't think there's any way to determine from a
> SEGV handler which null pointer needs to be initialized, and in any case
> users might want to have their own SEGV handlers.

But then, with support from the kernel and dynamic loader we aren't 
limited to normal signal handlers; it ought to be possible to mark a 
page of thread_local variables as trapping so that the first reference 
invokes a designated initialization function.  I think that such a 
scheme ought to be link-compatible with this one so long as the 
initialization function is the same; references would just be direct 
rather than through the wrapper function.

It sounds like recent versions of MacOS X support something like this, 
though clang doesn't take advantage of it yet.

Jason