Pietro Vertechi

Yes, right now if you push directly to a `StructArray` with the wrong type, the behavior is undefined. There is a `StructArrays.append!!` method that does "mutate or expand" (along the...

If all the fields are `isbits` then materializing a new struct does not allocate and is a reasonable thing to do. If your `StructArray` is large, OTOH, you should probably...

Benchmarking on a cheap operation (like `sum`) does not reveal a difference: ```julia julia> using StructArrays, BenchmarkTools julia> s = StructArray(a = rand(1000), b = rand(1000)); julia> @btime sum(LazyRow($s, i).a...

Thanks for the PR! In general I'm hoping to reduce the number of generated functions. I used to have generated `getindex` but it turns out it was not needed. Here...

Somehow, if it is a small number of tweaks, I would prefer to pass them as options via the julia API, but maybe there are cases where that doesn't make...

Don't really have the bandwidth for this at the moment, but I'd be happy to review a PR for this. The suggested approach seems reasonable.

Seems reasonable to me. For tuples and named tuples the conversion would happen field by field regardless when calling `setindex!` and company on the component arrays. It may be best...

I am also realizing that things are problematic for array types where the result of indexing is some custom type that has a reference to data from the original array,...

Thanks for the PR, I've left a review with some suggestions. I'll be traveling for the holidays so I might be slow in responding for the next couple of weeks.

I also think it's a bit odd that this should live here. IMO, the idea behind `ShiftArrays.circshift` versus `Base.circshift` (or `fftshift` and co.) is to return a lazy object with...