On Tuesday, December 4, 2018 at 1:50:19 AM UTC-8, Jeroen Demeyer wrote: > > On 2018-12-03 17:21, Nils Bruin wrote: > > In order to > > make that possible, the coercion map (referenced strongly on the > > codomain -- it needs to be strongly referenced somewhere to keep it > > alive) must not hold a strong ref to the domain. > > I wonder if there is a way to somehow reference an object from a pair of > objects: have A and B reference C in such a way that, if either A or B > get deallocated, also C gets deallocated. >
Tripledict does that to some extent (with its keys): if one of the key parts gets deallocated, the weakref callback removes the strong reference to the value. Note that you will never ensure that C gets deallocated: if someone else is keeping a ref to C, it should be kept alive. The best you can hope for is a structure where the existence of both A and B ensures the continued existence of C, but as soon as one of A or B goes, then the existence of C is no longer assured. All the leaks I've seen in settings like this come from loops where A,B,C have relations as above, but somehow C ends up anchoring a reference to A and B. That's the big loophole with globally rooted references that are guarded by a weak reference callback: what would normally be cycles in garbage now are suddenly globally anchored data structures not eligible for collection. As we've seen again and again, that's a very difficult paradigm to program correctly with. Note that the weak ref to the domain on the "coercion maps" shouldn't be a performance issue. It's just there to ensure that it's straightforward to turn it into a full-blown map. In situations where the map is recovered from the coercion system, it's done by looking up via domain and codomain. No need to look it up on the map again. So a micro-optimization would be to short-cut some evaluation steps, if you think the domain is presently being actively recovered via the weakref. See https://trac.sagemath.org/ticket/14711 for very detailed discussions about the how and why of weakly referenced domains in the coercion system. -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To post to this group, send email to sage-devel@googlegroups.com. Visit this group at https://groups.google.com/group/sage-devel. For more options, visit https://groups.google.com/d/optout.
