Oscar Smith

no. QOI is very specifically designed for 3 or 4 channel images. The result after QOI compression isn't that.

wow. oops.

because I closed the wrong one as a duplicate.

What's happening here is that `asech` has a branch cut at `im=0`, so the formula Julia is using (`asech(x) = acosh(inv(x))`) is getting tripped up because `inv(-3+0im)` is `-1/3 -...

This does appear to be a bug. We appear to be taking the wrong side of some branch cuts. That said, you probably shouldn't rely on finite precision math for...

yes.

can we get 32 core benchmarks as well? It would be very unfortunate if this was a regression for cpus with many cores. (it would be fine to limit GC...

Were these timings made on a version of Julia that includes the changes https://github.com/JuliaLang/julia/pull/45714 that make the GC time report correctly?

Were there any regressions? Can you post the before/after?

should this only happen for short string?