DQ

I didn't recently run our benchmarks since I'm currently in the middle of the tree bins, but if I recall correctly the last time I checked (after the `Moved` optimization),...

Worth noting that bustle initializes the maps with a reasonably high initial capacity by default if I understand correctly, so there are likely not that many moves occurring that would...

I agree. Just to have some exemplary numbers, here's the insert-heavy workload after the version bump locally: ``` -- Contrie 25165824 operations across 1 thread(s) in 17.4047412s; time/op = 691ns...

> I'm curious whether the flurry model fundamentally requires more allocations (and thus garbage) than contrie, or whether it is just a matter of the code not being sufficiently carefully...

I'm in a similar situation at the moment I fear. I've had some forced free time due to Covid19, but now everything is heading back online, except more chaotic. I've...

Finally found some time to look at this. I was interested most in the "gigabytes of garbage" for a start, so I ran valgrind's DHAT on a small test with...

I'm also a bit confused that we spend so much time/memory allocating epoch `LocalHandle`s. I know that we have a collector tied to each map so we can't just use...

It is in TLS, yeah, but that's where the `epoch::pin()` calls go. So we can't have just one handle per map. But accessing different maps should be covered by checking...

I was only thinking so far as that the check would cover the case where you try to access with a guard from a wrong local. I don't think I...

> We'd have to store a HashMap in the thread local. Which is doable, but also a little awkward. Why would we want to have the entire map be local...