Base R generic 'sort_by' for data.tables

[rikivillalba]

I wonder whether it is worth to implement sort_by.data.table (R 4.4.0) with internal forder instead of base order, and possibly an 'in_place' argument

Some improvement can be seen in my little test:

library(data.table)
set.seed(1234)
DT <- data.table(x=runif(1e7),y=runif(1e7))

# sort_by.data.frame with "forder" instead of "order"
sort_by2 <- function (x, y, ...) 
{
  if (inherits(y, "formula")) 
    y <- .formula2varlist(y, x)
  if (!is.list(y)) 
    y <- list(y)
  o <- do.call(data.table:::forder, c(unname(y), list(...)))
  x[o, , drop = FALSE]
}

microbenchmark::microbenchmark(times = 1, 
  sort_by2(DT, ~ x + y),
  sort_by(DT, ~ x + y)
)
#> Unit: seconds
#>                  expr      min       lq     mean   median       uq      max
#>  sort_by2(DT, ~x + y) 1.068101 1.068101 1.068101 1.068101 1.068101 1.068101
#>   sort_by(DT, ~x + y) 2.317193 2.317193 2.317193 2.317193 2.317193 2.317193

^{Created on 2024-12-12 with reprex v2.1.1}

[MichaelChirico]

Good idea. I wouldn't offer in_place (you can use setorder() for that, with a slightly different API).

It also looks like .formula2varlist() is key for allowing much more complicated formulas to WAI:

mtcars |> sort_by(~I(am + mpg))

It will be a lot harder to try and translate generic formulas into, e.g., a setorderv() call.