hardcode choices in match.arg

[thomasp85]

During performance benchmarking and profiling of plumber2 I stumbled upon the fact that the majority of time in toJSON() during formatting of smaller data is spend on match.arg() calls. The speed of these calls can be improved 3x by hardcoding the choices vector and avoid the lookup it otherwise performs in the calling environment.

This PR simply hardcodes all the values from the function formals into the relevant match.arg calls

[jeroen]

Hmm I'm a bit surprised if duplicating the values gives significant improvement. Can you include a quick example? Are talking about more than like 1 millisecond?

[thomasp85]

foo <- function(x = letters) {
  match.arg(x)
}
bar <- function(x = letters) {
  match.arg(x, letters)
}
bench::mark(foo(), bar())
#> # A tibble: 2 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 foo()        2.13µs   2.42µs   293471.    1.92KB     29.4
#> 2 bar()      614.91ns  697.1ns  1090702.        0B      0

^{Created on 2025-12-09 with reprex v2.1.1}

It is not a lot, granted, but there are quite a few calls in there and they stack up. As said, for a simple plumber2 api the json conversion is the biggest sink and the match.arg calls are the biggest part of that. As the change is cosmetic and the improvements are real I figured we might as well have it

[jeroen]

We are really talking about microseconds here... Even if we have a nested json object with hundreds of asJSON calls, this won't end up to a single millisecond ? I can hardly imagine this is a noticeable difference in plumber2... not sure if this really justifies duplicating this code...

[thomasp85]

fair - I agree it is small gains but since there were no downsides to this I thought I'd give it a go