"Sean O'Rourke" <[EMAIL PROTECTED]> writes: > On Mon, 26 Aug 2002, Nicholas Clark wrote: > > All ways of doing deterministic destruction seem to have considerable > > overhead. > > One possible alternative would be to have file handles and other objects > with destructors that have to be called in a timely fashion keep > ref-counts. When the refcount drops to zero, they release whatever > critical resource they are holding. Then the GC comes by later to reclaim > the memory.
Dont know if this would be a good idea. If you put an deterministic destruction object into an compund object. Then the compound object must also be deterministic destructed, that meens also refcounted. Otherwise the compound object lays around until the next DOD run (which can be anytime) holding a reference an prohibiting the destruction of the deterministic destruction object. Next all objects holding references to the compound object must also be refcounted to ensure timely destruction. This continues the hole way down to the root set. So a big part of the memory will be refcounted. Finding all this references to refcounted objects is a series of DOD runs. As long as there is found at least one not refcounted object, a new DOD must be triggered to make sure that no further refrences exist. Only if the compound object the deterministic destruct object is added is already refcounted then its just an increase of one refcount. So there are two ways out. Do refcounting for all objects so there are no recursive DOD runs needed. AFAIK this is the perl5 way. But it would be a major effort and a big design change in parrot bringing refcounting into the core even as a Config time (or may be loadtime or runtime) option. The other way is live with the many DOD runs for getting the refcounts right, but then you may get arround with a one bit refcount, because you have to iterate over the children anyway. If there is a (interpreter-)global needs_more_DODs variable the recursive DODs can be defered till the next assignment or block exit. But this is exactly what Nicholas Clark proposed just using diffrent names (deterministic destruction flag and loose object counter(|flag?). bye boeboe -- Juergen Boemmels [EMAIL PROTECTED] Fachbereich Physik Tel: ++49-(0)631-205-2817 Universitaet Kaiserslautern Fax: ++49-(0)631-205-3906 PGP Key fingerprint = 9F 56 54 3D 45 C1 32 6F 23 F6 C7 2F 85 93 DD 47
