Am Do., 26. Sept. 2024 um 20:32 Uhr schrieb Daphne Preston-Kendal < d...@nonceword.org>:
> On 26 Sep 2024, at 20:27, Marc Nieper-Wißkirchen <marc.nie...@gmail.com> > wrote: > > > Given the example of a weak hash table, when an existing association is > updated, why do we want to add the inefficiency of replacing the whole > ephemeron/weak pair when a simple mutation would work? > > I don’t accept that it would be more inefficient in all cases. Not in all cases, sure, but in those where it would introduce unnecessary boxing. > Under the optimization I suggest, replacing a potentially-circular > ephemeron with a weak pair (whose value is of a type that can’t possibly > link back to the key) would probably be a performance win. But without > benchmarks we are speculating here, and this has nothing to do with my > original suggestion, which is about using them as weak boxes, not as weak > pairs. > What speaks against a proper "weak pair" type? > > Another example I have in mind is a linked list of weak pairs (say, as > part of a cache structure) modelling a list of weak keys. > > Don’t you also need a strong pair to maintain the link to the next weak > pair in that case? Otherwise, one weak pair being broken will cause the > entire rest of the list to be lost. > That's why I wrote "weak pair" and not "ephemeron". Ah, now I understand the confusion. I am using a "weak pair" in the sense of Chez, where the value is retained even if the key is broken. In any case, this is an example of a mutable data structure where it would be inefficient to wrap each weak box.
