David Widmann

Because my impression is that downstream developers are annoyed by breaking releases of such a core part of the ecosystem (IIRC there are issues with complaints and last time we...

> Why shouldn't we fix the issue at hand in the next breaking release? I'm not against changing it in the next breaking release (I still think that it's somewhat...

> what is the preferred solution to this problem? My preferred solution at the time being is to not do anything - there doesn't seem to be an obvious bug...

How is #86 related to this PR?

> I don't understand why Document-CI is failing, FLoop is not in BangBang's [deps](https://github.com/JuliaFolds2/BangBang.jl/blob/master/Project.toml). MLUtils is a docs dependency and depends on FLoops - however FLoops does not support BangBang...

MLUtils is used in a single example: https://github.com/TuringLang/DynamicPPL.jl/blob/abcf5843856235cf8ee391b67bcde2750cb2aa56/docs/src/tutorials/prob-interface.md?plain=1#L112-L141 Since it's a quite heavy dependency, I wonder if we could avoid it.

I think that's a good idea. The package is stable, mature, and widely used, so I think that should also be reflected in the version number.

We would have to define `PDSparseMat` in the main package without restricting the type parameter to be able to export it - which is definitely possible but indicates that this...

https://github.com/JuliaStats/PDMats.jl/pull/175 should fix these major issues with `PDSparseMat`, so it seems with this PR there's not a strong argument anymore for removing `PDSparseMat`.

In general, credible intervals are no confidence intervals and don't satisfy any coverage guarantees.