eoincondron

Minimal example `ds = rt.Dataset({'x': [0, 1, 2]})` `rt.Struct({'ds': ds}).save('tmp')` `rt.load_sds('tmp', name='ds', filter=[0, 1])` 

DatetimeNano.clip does not preserve type.

1

I guess they are other methods inherited from `FastArray` which also do not preserve type 

I noticed the when calculating `var` (and `std`) with `Accum2` that the value in the margin at the intersection (corresponding to the var of the entire dataset) is based on...

GroupBy median appears to preserve type for integer inputs when the result should be a float In [60]: cat = rt.Cat([0, 0, 1, 1]) In [61]: cat.median(rt.arange(4)) *key_0 col_0 ------...

Consider the example below  Perhaps also unexpected:  Should non-unique categories be supported?

Enable further signature compilation

3

Currently, when signatures are passed to the `jit` decorator, dispatcher compilation is disabled, meaning unseen signatures cannot be compiled when encountered for the first time (the behaviour that we get...