_impure_ptr invalid after task_delete() called?

[Chris Johns]

Joel Sherrill wrote:
> 
> It's more of the former I think.
> 

Ok. I will add something to the task delete when I update the create
extension doco with the boolean return (PR228) unless we can resolve it.
Should task variables also have doco added ?

> Chris .. is it possible this is similar to the problem we were seeing
> where when POSIX was enabled, it looked like the extensions were
> getting run twice on the same task?  Or was that a ghost?

I do not think so. I need to see what was happening with this. The POSIX
task delete should check the class of the object before acting on it. I
do not think it does, but I am not sure.

> 
> The order is supposed to be defined and documented. In general, the
> extensions are run forward, but the delete order is backwards to be
> the opposite of create order.  That way you should be able to depend
> upon the same resources being there for create and delete.
> 

Sure, but is the extact list and where user or driver extensions are
relative to RTEMS ones documented, eg ASR, newlib, classic API, POSIX,
ITRON, task variables, stack monitor, capture engine, etc ?

What about situations where a static extension is added by a user, or
created by a driver, or some other part of a BSP create, eg post driver
phase etc ?

It would seem the task variable extension is being added before the
newlib extension. I feel being able to access libc as Peter wishes is
something that should be supported.

-- 
 Chris Johns, cjohns at cybertec.com.au