Re: [PATCH] Destructor support for C++11 thread_local variables

[Jason Merrill <jason at redhat dot com>]

On 10/11/2012 06:58 PM, Roland McGrath wrote:
> Is there any order requirement?  And is there a specific rationale for it
> other than the aforementioned "principle of least astonishment"?

The rule was basically copied from the rule for variables with static 
storage duration.

"It is implementation-defined whether the dynamic initialization of a 
non-local variable with static storage duration is done before the first 
statement of main. If the initialization is deferred to some point in 
time after the first statement of main, it shall occur before the first 
odr-use (3.2) of any function or variable defined in the same 
translation unit as the variable to be initialized."

and

"It is implementation-defined whether the dynamic initialization of a 
non-local variable with thread storage duration is done before the first 
statement of the initial function of the thread. If the initialization 
is deferred to some point in time after the first statement of the 
initial function of the thread, it shall occur
before the first odr-use (3.2) of any variable with thread storage 
duration defined in the same translation unit as the variable to be 
initialized."

I believe the rule for statics was there so that code in the TU that 
relies on side-effects of the initializations will work.

On 10/12/2012 02:15 AM, Jakub Jelinek wrote:
> Though, I still wonder how is that implementable.

The way I implemented it?  :)

> Because a TU could have
> say 3 thread_local vars with ctors/dtors, and the middle one of them could
> call in the constructor some function that accesses thread_local with
> ctors/dtors of some other TU, then you have just partially initialized the
> current TU and need to initialize another TU

Yes, and then we come back and finish initializing the first TU when 
that one is done.

> (and it is unclear in what
> order they need to be destructed).

Each variable has its own cleanup which is queued when its 
initialization is complete, just like variables with static storage 
duration.

Jason