On Monday, October 27, 2014 10:45:40 PM UTC, Nils Bruin wrote: > > In general I expect that cached_methods should preprocess their arguments > a bit before doing a key lookup anyway (such as do the required coercions > on arguments). Most of the time, functions with arguments should *not* be > cached, because their cache quickly becomes a memory leak. >
I tend to disagree, IMHO this should just be handled better by the caching framework. For example, have a strong LRU cache of a certain size, and expire older results into a weakly referenced cache. With optional arguments to tune the LRU size etc. But the most important thing is that caching must be 100% reliable, and that means input must be really equal (at most up to some unique isomorphism), and not just "equal enough". So either we make __eq__ much stricter, or we write our own associative container that doesn't use __eq__ to compare inputs. The _cache_key() convention in #16316 is a baby step in that direction, except that it does it only for caching and doesn't provide a generally usable associative container. Eg. you'd essentially have to write your own hash table for Buchberger just for padics. It seems that there could be three versions of associative containers (dict, set) 1) Python version. Randomly wrong if you add elements from different parents. 2) A version coercing to a common parent (similar to Sequence) The first two are insertion-order dependent on same parent with overly loose == comparison. 3) A version that treats elements from different parents as distinct before even calling ==, and with a hook to override == with a stricter comparison. This is what caching needs to use. -- 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 http://groups.google.com/group/sage-devel. For more options, visit https://groups.google.com/d/optout.
