Yes, this is embarrassingly slow, so I solved your problem

stackoverflow.com

71 pointsnojito5 comments

This brings me joy -- a problem well expressed, and a quick and helpful response

I can imagine how nice it felt being that developer too, seeing this and being like... You know what? This could be better - and just knocking it out of the park

Small things that make a big difference like this are a high I chase

me not. he said to have implemented the new inset lookup with an optional hashmap. look at the PR commit. it uses a C++ ordered set and map to do the lookups, ie a red-black tree, not a hashmap. either the implementation is fucked up, or the API.

I admire your tenacity -- after a certain hour on a Friday, I try to avoid other peoples diffs :)

the map member used for the lookups is a tsl::hopscotch_map (https://github.com/Tessil/hopscotch-map), which is a proper hash map. so it seems to be the latter, that the API is wrong, but from what I can tell it is only a wrongly named class. i don't see where the API makes guarantees about iteration order, which is where the implementation difference would be noticeable (beyond performance for lookup).

Reply on news.ycombinator.com