Lightweight wrapper for floating point arrays with NaNs

[simonster]

It'd be nice to have a subtype of AbstractDataArray that wraps a regular floating point array and treats NaN values as NA. I think this should generally be faster than indexing the BitArray that holds the list of NA values, and with #4 and some minor API changes this should eventually give us nansum etc. with equivalent performance to manual loops.

[johnmyleswhite]

This would make @tshort happy I suspect. It's definitely worth adding, but I'm a little worried about how it would fit into the system -- would you always use FloatDataArray instead of DataArray when working with floats?

[simonster]

I'm not sure, although it seems safer to use a DataArray even when working with floats. With a DataArray you could potentially have values that are NaN but not NA, and thus should contaminate the result when NAs are ignored. This wrapper wouldn't be able to express that. On the other hand such cases probably arise rarely in practice. How does R treat NaN vs. NA?

[tshort]

In R, NA and NaN have a different bit pattern (both of which are NaN's). You could do the same in Julia.

NA's and NaN's can get mixed up if you mix operations that include both.NA + NaN could be a different answer than NaN + NA. I don't think you can avoid those issues. Anyway, I think it's rare to have conflicts between NA and NaN.