(j3.2006) J3/15-244

[John Reid]

Malcolm Cohen wrote:
> John Reid writes:
> <<<
> I don't like J3/15-244 at all. It is tearing into the TR before it has even
> been published. I think all that is needed is to delete "Otherwise, it is
> assigned the value -1." from the description of COUNT.
>>>>
>
> Not so.  The description of EVENT_QUERY as an atomic subroutine is
> fundamentally flawed.  It does not have an ATOM argument, and does not do
> anything "atomically" with respect to the execution of other atomic
> subroutines.  Describing it as an atomic subroutine thus involves several
> internal contradictions.

But the count of EVENT is atomic and is being referenced. It is surely 
important that the value returned is obtained atomically somewhere in 
the sequence of changes made by EVENT POST and EVENT WAIT statements. It 
is really very like ATOMIC_REF.

> The cleanest way to fix this is to simply describe the required properties
> of EVENT_QUERY directly, as 15-244 attempts, and not try to shoehorn it into
> the Atomic Subroutine category which it manifestly does not match.

This is not so. The first sentence of 8.2 of the TS says "An atomic 
subroutine is an intrinsic subroutine that performs an action on its 
ATOM argument or the count of its EVENT argument atomically."

Cheers,

John.
>
> This is not "tearing into" anything, it is fixing errors.  There is no
> technical effect from describing it correctly (viz not as an atomic
> subroutine) beyond avoiding contradictions in the description.  And since we
> have two different statements of what value is assigned to COUNT, the
> specific one in EVENT_QUERY is obviously the one which is "truest" to the
> soon-to-be-published TS.
>
> Cheers,
>