Michael Abbott

In a chat now lost, I was persuaded by someone that a fairly common situation is to have (say) 90% of samples almost at the minimum, and 10% a bit...

Here's an attempt... with https://github.com/JuliaCI/BenchmarkTools.jl/issues/237#issuecomment-922498262 's suggestion of ~~95th~~ 99th percentile instead of max. * I think putting the times first, and in order, makes them easier to find. *...

It was indeed printing things twice, sorry, now fixed. (~~Weirdly Revise isn't working for me, for the first time ever.~~ my fault!)

Thanks for taking a close look. To explain my thinking a bit: * The GC times are below "2 allocations..." partly to give the 1st line breathing space, and partly...

Notice in that example that `min 970.874 μs, mean 25.000 ms, max 97.087 ms` is quite revealing. Combined with the histogram of many points at the minimum, something is going...

Yes I don't think I've ever learned anything from `max`. It's possible that other people do, if they care about real-time things, or network time-outs?

A high percentile might be neat, we should try it out. One nice thing about printing `min, 50th, mean, 95th` or whatever is that all numbers displayed are total times....

One further step might be to put the number of samples etc. under the histogram -- it still refers directly to the data, but is out of the way. It's...

The code is a mess BTW, sorry. Lots to tidy up. > `median 4.607 μs (½)` I don't love this, maybe there's a better way. I hoped there would be...

> I was talking about the positioning of the (½), not the symbol used. Sure. I meant that needing to provide a key at all seems a bit awkward, I...