Michael Abbott

> Did you start julia withj --startup=no ? No, but the times are pretty consistent. If Revise takes a long time to think about things, that's relevant time for actual...

> option of ChainRules.jl isn't great because: Good to lay these things out. To me these reasons don't seem to obviously counteract the extra hassle of having one more package...

> it seems it shouldn't Why should this be disallowed? If e.g. `hcat` is going to work, then it will inevitably sometimes produce duplicates. Lookup works by `findfirst` I think:...

> which (I assume) very often But just asserting this doesn't answer the question. Why shouldn't the labels attached to axes be categories or dates or some other information, which...

To be fair, that's a MethodError without this package, and the expected answer with it. More surprising is: ``` julia> reshape(rand(1,2), Int32(2)) 2-element Vector{Float64}: 0.23063257026910478 0.007514586985956306 julia> reshape(rand(2,3), :, Int32(2))...

I assumed the same. And while I didn't write this, if I had, it would have been accidental piracy from this assumption -- trying to allow UnitRanges in otherwise the...

Not sure how to fix this, but here is another example, and another related error: ```julia x = AxisArray(rand(3,4)) # no errors with x = rand(3,4) v = view(x, :,...

Will do. But first, one more puzzle which just bit me: why is the first of these 5x quicker? ``` V2 = reshape(V, 1,:) @time sum(BroadcastArray(*, V, V', V3), dims=(2,3));...

Thanks for taking a look, it is indeed strange. The biggest difference in an isolated getindex I can see is... far from a factor of 5: ``` B1 = BroadcastArray(*,...

Yes that could be what's happening. I was trying to dig into what `CartesianIndices` etc. give, but couldn't find any difference. Here's another attempt, but I don't know what this...