David Widmann

Nice! IMO both approaches try to address a different problem (which we should both solve in some way). The approach with `return` (or some other similar model-based approach) tries to...

You're absolutely right - I assumed `_setval!` just operates on vectors, I guess due to the new indexing behaviour in MCMCChains#dev (IIRC `chains[i]` just returns an array (with labelled axes)).

It seems the main issue is that currently it is not clearly defined if `missings` is based on the name of the arguments or name of random variables in the...

I don't think 2 is needed. I assume one can call `_evaluate` not with the `model` but with `merge_missings(model, __model__)` where `merge_missings` adds the `missings` of `__model__` that begin with...

I know this Github feature, we use(d) it over at JuliaGaussianProcesses :slightly_smiling_face: When we started using bors, it didn't exist. At that time there were often many PRs at the...

Another nice thing about bors is that one can specify explicitly when a PR should be tested and CI tests be run.

> This can be done via GitHub too, right? Either via git commit --allow-empty -m 'Trigger CI' or via the web interface. How can this be done in the web...

This PR is included in https://github.com/TuringLang/DynamicPPL.jl/pull/406 as well. That PR is not merged yet since some package was holding back Setfield.

> the conditioned model doesn't need to allocate y at all. The model explicitly states though that `y` should be allocated: `y = similar(...)`. Turing does not change your model...

It's nice to see what already can be done - but I strongly feel like nobody should have to code models in that way. Users should not use any of...