Re: Suspicious reentrant semaphore

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

Jano wrote in message ...
>Hello everybody,
>Leaving aside why I have done that (yes, poor design, I'm removing it
>anyway ;-), I would like to know if my package is right or really is
>there some race condition. Please comment on it. I know that
>Current_Task shouldn't be called inside entries (in fact Gnat doesn't
>allow it, I tried :-P) but doesn't complain about it being used in
>barriers. Anyways, I have a (maybe wrong) feel about that being related
>to the deadlock.


Why don't you use the attribute defined for this purpose? E'Caller
returns the task-id of the caller.

Claw uses a lock much like this one, and it uses E'Caller, not
Current_Task (which may not be the caller).

Here's the body of the "Get" entry from Claw:

        entry Get when True is
            use type Ada.Task_Identification.Task_Id;
        begin
            if Use_Count = 0 then
                Owner_Task_Id := Get'Caller;
                Use_Count := 1;
            elsif Owner_Task_Id = Get'Caller then
                Use_Count := Use_Count + 1;
            else
                requeue Wait_for_Free with abort; -- Let any timeouts
expire.
            end if;
        end Get;

And Wait_for_Free looks like:

        entry Wait_for_Free when Use_Count = 0 is
        begin
            Owner_Task_Id := Wait_for_Free'Caller;
            Use_Count := 1;
        end Wait_for_Free;

           Randy Brukardt.