Jon Gjengset

Oh, you're right, I was thinking of `ops_f`, not `prefill_f`. And yes, looks like the assertion is inverted. Will fix that shortly!

Fix published in 0.4.

:p It's not clear to me that "real world" scenarios will always be prefilled. I think it makes sense to test both. A common use-case for example is to have...

One of these days when the benchmarks and implementations settle a bit, I'll also run the whole suite on a bunch of many-core machines I have hardware access to so...

Very true! It may be that we have to be smarter about garbage in the Rust case. Or maybe use some kind of arena/bump allocator for often-reused heap values.

That too!

The only place I can imagine that being relevant is [this allocation](https://github.com/jonhoo/flurry/blob/7b27a79cfa502c054583ad7ca52358e24306782b/src/map.rs#L1362) which may be dropped (without being shared) if it turns out we are aborting the operation early (like...

That y-axis had me pretty confused for a second. Lower is better is odd, but I guess it makes sense here. Is this with the newly released `flurry 0.3`? And...

What mix are you running with above? Are there any writes, or is it read-only following the population of the map?

I'd personally also be curious about a read-only workload. Mostly because I want to figure out exactly what it is that's causing flurry pain in these benchmarks. My _guess_ is...