David Widmann

I'd actually prefer Runic, it seems to work fine in StatsFuns and PDMats and I've been bitten by JuliaFormatter 2 problems a few times. GLM uses JuliaFormatter though, so there...

Maybe let's merge https://github.com/JuliaStats/Distributions.jl/pull/1905 before changing the formatting.

> where I could reproduce the issue on [d1185830] SymbolicUtils v3.7.2 I detected the problem initially on SymbolicUtils 1.5.1, so it seems it has existed for quite some time.

Instead of dealing with subnormals, at least for the example here sampling in log space would be sufficient (see also https://github.com/JuliaStats/Distributions.jl/issues/1003#issuecomment-636450042, https://github.com/JuliaStats/Distributions.jl/issues/1003#issuecomment-556978582, and https://github.com/JuliaStats/Distributions.jl/issues/1810). For instance, with an `ExpGamma` version...

This is a known issue and was suggested in e.g. #1730 (also related: #1467). I think the type instability would be a quite unfortunate consequence of such a change. Maybe...

It's generally always preferable to use `nothing` instead of an endpoint of the untruncated distribution. The former allows to optimize to calculations with, and possibly even the returned type of,...

https://github.com/TuringLang/Bijectors.jl/pull/262 was supposed to fix it. But it seems there exists no DOCUMENTER_KEY at all in Bijectors?!

I added the key (and the matching deploy key). So now TagBot should be able to trigger the docs build.

Ah, so it's not relevant that `spec1` and `spec2` have completely distinct `color`s, both `visual`s are combined because potentially there could be a value that shows up in both of...

Naive question, but isn't this solved by SciMLSensitivity and typically the adjoint method used automatically when ADing through `solve(...)` (with possible manual adjustments) for a DE system? I'm not 100%...