Re: Intended use of package Ada.Task_Attributes

["Randy Brukardt" <randy@rrsoftware.com>]

"Jerome Hugues" <hugues@merlin-lenchanteur.enst.fr> wrote in message
news:slrnd7kf55.5rk.hugues@merlin.enst.fr...
...
> If you read the implementation notes of GNAT (for GNU/Linux, see
> a-tasatt.adb), you'll understand why it is not something one should
> allow for any real-time system, unless your tool vendor provide strong
> guarantee for your target.

That's one specific implementation, and it certainly isn't the only way
these could be implemented. Certainly, I wouldn't use the Janus/Ada
implementation in a hard-real-time system (it being essentially what Dmitry
described. But I know of at least one compiler that allocates task
attributes directly in the TCBs. That of course means that the items that
can be task attributes are very limited, but certainly there is no run-time
penalty for having them. Those would be very appropriate for a hard
real-time system.

(Note that this is a perfectly legal implementation of Task_Attributes,
C.7.2(29) allows any limitation at all on the number and size of task
attributes.)

I would suppose that the Ravenscar designers didn't think Task_Attributes
were important enough to mandate a useful (for real-time) implementation.
Some of the restrictions (like this one) in Ravenscar are fairly mysterious.
(But I hasten to point out that real-time isn't my field, so I try not to
second-guess too much in that area...)

                            Randy.