The frequency weight trick seems like a cool one. You always have to code around frequency/probability/analytic weights, so you get two birds with one stone (It's not a perfect solution though, as if you need to e.g. regress two variables by different groups, then updating the mask for each value of the group ID is silly compared to just adding a conditional)
On a related note, I don't know why more people don't look at what others in stats have made (e.g. Stata's solution of having 32 missing values using the very last values of each numeric type, so you can bitmask them easily). If you really want to supersede Matlab, R, Numpy, Stata, etc. then don't discard everything they have made, learn about the good parts and use them (!!!).
Also, what's up with the comments here? At least 4/8 comments are basically saying "don't use regressions, use a technique that doesn't need missing value hacks..". Well guys that's NOT up to the language writer to decide. If you don't do linear regressions well and fast they you are dead in the water in terms of a basic statistical package. Same for the precision issue; many algorithms at some point require quad precision so just saying "screw it change the algo" just misses the point.
On a related note, I don't know why more people don't look at what others in stats have made (e.g. Stata's solution of having 32 missing values using the very last values of each numeric type, so you can bitmask them easily). If you really want to supersede Matlab, R, Numpy, Stata, etc. then don't discard everything they have made, learn about the good parts and use them (!!!).
Also, what's up with the comments here? At least 4/8 comments are basically saying "don't use regressions, use a technique that doesn't need missing value hacks..". Well guys that's NOT up to the language writer to decide. If you don't do linear regressions well and fast they you are dead in the water in terms of a basic statistical package. Same for the precision issue; many algorithms at some point require quad precision so just saying "screw it change the algo" just misses the point.