Allow vec_cast of numeric to units

[hughjonesd]

I've been working on integrating support for units into hughjonesd/santoku. I sometimes want to determine whether a units vector x is positive or not. For x > 0 to work, I need to either promote x to a numeric or promote 0 to the relevant unit. My ideal would be x > vec_cast(0, x), since this keeps the code generic. At the moment, this fails. Would it be reasonable to allow casting a numeric type to a non-mixed units variable?

[Enchufa2]

Implementing such a method would mean that pivoting a data frame or binding rows could lead to silent binds between units and non-units vectors, and this is not something we want.

But how is this x > vec_cast(0, x) more generic than simply as.numeric(x) > 0?

[hughjonesd]

as.numeric(x) will often convert to NA, when it should error, e.g. as.numeric("foo"). My current workaround is to write my own strict_as_numeric() function.

[Enchufa2]

I see. I still think that this is the best approach here. units treats plain numeric vectors as unitless quantities, and it has been the case for a number of years now. The reason is that you are able to do things like 2*set_units(1, m), which seems natural. Casting unitless to any other unit cannot be allowed, and thus this behaviour with comparisons. As I already said, this casting would allow row bindings with units and non-units objects with dplyr, and this breaks the unitless assumption.

Let us know if we could do something else to help you with this, but otherwise, we should close this.

[hughjonesd]

Yeah, I understand. I wonder if rlang would benefit from a stricter as_numeric() function.

[Enchufa2]

Perfect, thanks for the discussion!