keep_units() with non-units x input

[d-morrison]

If keep_units() is called with a non-units x value, would it be reasonable to just pass through that value to the normal function and not apply units at the end? For example:

units::keep_units(rexp, n = 100, x = 3)
#> Error in UseMethod("units"): no applicable method for 'units' applied to an object of class "c('double', 'numeric')"

^{Created on 2025-02-13 with reprex v2.1.1}

This would be helpful in a function I'm writing where I don't know in advance if the user will provide an input with units or not. As a workaround I can check for units on the input and only call keep_units() when there are units to keep, but built-in behavior would be great!

[d-morrison]

PS - I'm now realizing I picked a particularly unfortunate example, because in the exponential distribution, the units of the rate parameter need to be inverted for the output, if they are present, so I'll need more than keep_units() for that particular case anyway. Nevertheless, pass-through behavior for non-units still seems potentially useful?

[Enchufa2]

keep_units is a pretty simple convenience wrapper

> units::keep_units
function(FUN, x, ..., unit=units(x)) {
  set_units(do.call(FUN, list(x, ...)), unit, mode="standard")
}
<bytecode: 0x55bcfb8018d0>
<environment: namespace:units>

that, as you can see, calls the function and sets the units of the result equal to the one of the first argument. So I'm not sure how we should handle a general case in which there is an indeterminate number of arguments, and where the user may want to keep the unit of an arbritrary argument, with potentially units of different types in different arguments, and also non-units objects here and there. :) IMHO such special cases should be treated individually.

[d-morrison]

Thanks for your reply! At the risk of overstaying my welcome, how about something like this:

keep_units <- function(FUN, x, ..., unit=units(x)) {
  if (inherits(x, "units")) {
    set_units(do.call(FUN, list(x, ...)), unit, mode = "standard")
  } else {
    warning("`x` does not have units.") # maybe not necessary to warn?
    do.call(FUN, list(x, ...))
  }
}

It could also throw a warning in the else clause. Still pretty simple, but handles the test case above. Understandable if out of scope; apologies if unhelpful!

[Enchufa2]

It is simple enough, and I do think that the warning is appropriate. Would you open a PR with this?

[d-morrison]

Thanks! Will do.

[Enchufa2]

One question about this use case. Wouldn't you want to do something like this?

lims <- set_units(c(1, 3), "s")
keep_units(runif, 3, lims[1], lims[2], unit=units(lims))
#> Units: [s]
#> [1] 2.738217 2.520332 2.669949

This currently works. But it wouldn't with your approach.

[Enchufa2]

Or

rate <- set_units(3, "1/min")
keep_units(rexp, 3, rate, unit=units(1/rate))
#> Units: [min]
#> [1] 0.70244335 0.38413054 0.05673904

[d-morrison]

One question about this use case. Wouldn't you want to do something like this?

lims <- set_units(c(1, 3), "s")
keep_units(runif, 3, lims[1], lims[2], unit=units(lims))
#> Units: [s]
#> [1] 2.738217 2.520332 2.669949
This currently works. But it wouldn't with your approach.

Good point; in my use case, I'm currently getting around the "first-argument defines units" issue by naming the n argument:

lims <- set_units(c(1, 3), "s")
keep_units(runif, n = 3, lims[1], lims[2])

I hadn't noticed the unit argument in keep_units(); I'm thinking about whether I can add a bit more logic to handle that better. I set up a draft PR (#394), but I'm still working on it; I'll convert to full PR asap (might be a few days before I can circle back to it)