Start using look-up tables instead of using complex conditionals?

After a certain point, a block of conditional statements can become quite difficult to read, maintain, and further extend.

To see what I mean, consider this example from {insight}:

transform_fun <- "exp"

if (transform_fun == "identity") {
  out <- list(transformation = function(x) x, inverse = function(x) x)
} else if (transform_fun == "log") {
  out <- list(transformation = log, inverse = exp)
} else if (transform_fun %in% c("log1p", "log(x+1)")) {
  out <- list(transformation = log1p, inverse = expm1)
} else if (transform_fun == "log10") {
  out <- list(transformation = log10, inverse = function(x) NA)
} else if (transform_fun == "log2") {
  out <- list(transformation = log2, inverse = function(x) NA)
} else if (transform_fun == "exp") {
  out <- list(transformation = exp, inverse = log)
} else if (transform_fun == "sqrt") {
  out <- list(transformation = sqrt, inverse = function(x) x^2)
} else if (transform_fun == "power") {
  out <- list(transformation = function(x) x^2, inverse = sqrt)
} else if (transform_fun == "expm1") {
  out <- list(transformation = expm1, inverse = log1p)
} else if (transform_fun == "log-log") {
  out <- list(
    transformation = function(x) log(log(x)),
    inverse = function(x) exp(exp(x))
  )
}

The alternative here is to create a look-up table, which is much easier to read, and importantly, extend - we just need to add another row for every new transformation:

df <- tibble::tribble(
  ~transform_fun, ~out,
  "identity",     list(transformation = function(x) x, inverse = function(x) x),
  "log",          list(transformation = log, inverse = exp),
  "log1p",        list(transformation = log1p, inverse = expm1),
  "log(x+1)",     list(transformation = log1p, inverse = expm1),
  "log10",        list(transformation = log10, inverse = function(x) NA),
  "log2",         list(transformation = log2, inverse = function(x) NA),
  "exp",          list(transformation = exp, inverse = log),
  "sqrt",         list(transformation = sqrt, inverse = function(x) x^2),
  "power",        list(transformation = function(x) x^2, inverse = sqrt),
  "expm1",        list(transformation = expm1, inverse = log1p),
  "log-log",      list(transformation = function(x) log(log(x)), inverse = function(x) exp(exp(x)))
)

These two approaches, of course, yield the same result:

identical(
  out, 
  df$out[df$transform_fun == transform_fun][[1L]]
)
#> [1] TRUE

^{Created on 2022-12-21 with reprex v2.0.2}

The only complication this introduces is making sure that this data frame is available at build time, which requires collation order (one can use #' @include to this easily).

Should we start using such look-up tables where relevant?

Start by reviewing the linked R/get_transformation.R example and the surrounding project code to identify which conditional blocks are intended to change. The issue does not define a target set of files, so completion would first require agreement on scope, then preserving existing transformation behavior while replacing the selected conditionals with lookup tables.