options() in an extension package

[EvaMaeRey]

What are best practices for using options or similar.

[teunbrand]

I don't really like using the options. It feels like some global state that departs from the grammar.
That said, probably use getOption() explicitly in the function arguments, so that users can see that in the usage section.
Example:

my_function <- function(x, opt = getOption("mypackage.opt", default = NULL)) {
  opt <- opt %||% compute_default(x)
  ...
}

[yjunechoe]

I've never noticed this but that's a nice design! Ex:

> ggplot2::scale_fill_discrete
function (..., type = getOption("ggplot2.discrete.fill")) 
{
    type <- type %||% scale_fill_hue
    <...>
}

[yjunechoe]

I think the answer also depends heavily on whether the package plans to simply read ("get") options, or also write ("set") options to different values in different contexts (aside from initializing them).

I keep bookmarked some advice for taming the side-effect behavior of options() in:

• R Packages: https://r-pkgs.org/code.html#sec-code-r-landscape
• Tidy design principles: https://design.tidyverse.org/side-effects.html

Edit: for the record, I also don't really like using options() other than for trivial behaviors that specifically target the user experience (vs. change a package's feature) like the verbosity of messages/warnings.

[EvaMaeRey]

One place where I felt really grateful for options, and where it also felt like it made sense and was justified is options(lubridate.week.start = 1) so that wday will switch from Sunday to Monday. I think a really global switch needs to be possible for this kind of thing.

That said, my use case might not be as compelling. Something like...

options(ggregion.geo.ref.data = northcarolina_counties, ggregion.required.aes = "fips|county.name"

Then exported functions geom_region() and friends would be able to just work in the northcarolina county context, I think.

[EvaMaeRey]

Nope... rnaturalearth is the culprit.

[EvaMaeRey]

set_region_country_rnaturalearth <- function(scale = "small"){

rnaturalearth::ne_countries(
scale = scale, returnclass = "sf") |>
select(country_name = sovereignt, iso3c = iso_a3) |>
mutate(country_name = ifelse(country_name == "United States of America", "United States", country_name)) |>
mutate(iso3c = ifelse(country_name == "France", "FRA", iso3c)) |>
options(sf2stat.ref_data = _,
sf2stat.required_aes = "country_name|iso3c")

}

[EvaMaeRey]

Annotation layer feature...

[EvaMaeRey]

Similar 'stamp' code below, but with fill scale informing about regions.

[EvaMaeRey]

I like the possibility of using this options strategy in modeling context too. Where you might want to define the model of interest, for example y~x+cat 😺 and lm, and then plot feature incrementally, model fit across x, fitted values, residuals, etc. where each is a layer.