Rafael Schouten

You can always provide more specific methods than `AbstractDiskArray` for specific variable types. There may be a few boilerplate methods needed to avoid ambiguity, yes. What is required is a...

Ok yes that's no problem at all. Feel free to PR

Are these not still just NCDataset.jl using the macro? Do you get this with just DiskArrays and no other package? Also have you tried loading DiskArrays.il before anything else?

Only 1000 of these are from DiskArrays against Base (not just zip - all invalidations), if you just load DiskArrays and SnoopCompileCore and do `@snoop_invalidations`. Probably that is to be...

@charleskawczynski another thing to note is your invalidations are most likely cause from type instability where arrays are being inferred as `AbstractArray`, in which case the disk arrays methods need...

Probably need to do `modify(cache, dimarray)` first to use chunk caching in DiskArrays. It should help a bit. But otherwise I'm not sure how we can get Tables.jl sources to...

@felixcremer you said somewhere that YAX doesn't have this problem. Do you know where the YAX code is for chunked table loads? we can move it to the DD DiskArrays.jl...

Yes! It's working well, try it out. There are some examples in the tests. The trick to the macro is getting every argument into a separate variable so we can...

Yes, need to think about the best syntax. And how to get it. What are you thinking? We could always add our own type like `Sink` and do `CSV.read(table, Sink(DimStack;...

Yeah, probably thats best from a user perspective. Just slightly annoying to need an extension just for CSV.jl rather than for any table reader