Michael Abbott

I had one for aligning dimensions, here: https://github.com/mcabbott/NamedPlus.jl/blob/master/src/permute.jl But have not thought in any detail about how aligning by keys should work. It's going to `cat` on some zeros, along...

OK now I read the xarray doc, a bit. It sounds a bit like whether you intersect or unite the keys ought to be per-dimension, not per function call. Maybe...

> I don't think the behaviour should be completely different depending on whether you passed keywords, but rather that passing no keywords should perform some default behaviour (e.g., intersect for...

This business of how to map between "natural" and "structural" representations was bugging me, but I now think it's actually very simple: Think of the forward pass as maps `W(fields)...

> to produce a CuArray sometimes I guess we never actually want to produce this, for converting natural to structural. We want just `pullback(collect, ...)[2]` without making the `[1]` at...

A variant of (B) is to make `@inplaceablethunk(1)` an error -- if the top-level expression isn't broadcasting then it shouldn't guess. Another nice thing to have would be a 2-argument...

If there was method `add!!(dx::AbstractArray, plus::Broadcasted)` then it could accept un-materialised broadcasts of just the RHS, just `2 .* real.(ȳ) .* x`, and turn them into `dx .+= ...`. Then...

Sounds good. It already takes on keyword `force=false` to disable over-writing, so maybe `skip=false` is the default too (all the data must end up in the container) while true keeps...

Here's what projection stores for `x`: ```julia julia> using SparseArrays, ChainRulesCore julia> x = sprand(3,3, 0.2) 3×3 SparseMatrixCSC{Float64, Int64} with 3 stored entries: 0.356584 ⋅ ⋅ ⋅ ⋅ ⋅ 0.249112...

That's unfortunate, I guess it's an argument for not using the same function to construct from an array and from a Table. We did not think of this problem when...