Ignoring AsIs objects (again)

[teunbrand]

This PR is the 3rd attempt to fix #5142 and replaces #5290 and #5418.

Briefly, instead of assigning scale_*_identity() to <AsIs> objects, it ignores these in scale operations.

Benefits of this are:

• You can safely do aes(colour = I(...)) in one layers and do aes(colour = proper_column) in another layer, without having a clash of colour scales.
• Using x/y = I(...) effectively makes the position aesthetic in an relative unit ("npc" units), which can be very convenient for annotations and such.

The mechanism this PR employs is to hide/unhide <AsIs> column at strategic moments during plot building.
'Ignoring' packs <AsIs> columns into one .ignored data.frame column.
'Exposing' unpacks the .ignored data.frame column into separate columns again.

A quick comparison with the other PRs:

Compared to #5290:

• No need for scales to make a formal exemption for <AsIs> columns; plot building takes care of this.
• For that reason, this PR has a much smaller footprint and is less invasive.

Compared to #5418:

• Don't need bother with mappings to derive which columns should be ignored.
• Because the <AsIs> class is 'infective', the mechanism automatically propagates through reparameterization, which is desirable (e.g. from x/width to xmin/xmax).

A quick benchmark to show that hiding/exposing scales well with dataset size (nrow = 32, 234 and 53.940 for mtcars, mpg and diamonds respectively). All in all, I think this PR wouldn't even add 1ms to plot building in 90% of plots.

devtools::load_all("~/packages/ggplot2/")
#> ℹ Loading ggplot2

data <- list(
  transform(mtcars, cyl = I(cyl), disp  = I(disp)),
  transform(mpg,    cyl = I(cyl), displ = I(displ)),
  transform(diamonds, color = I(color), depth = I(depth))
)

bench::mark(
  "mtcars"   = ignore_data(data[[1]]),
  "mpg"      = ignore_data(data[[2]]),
  "diamonds" = ignore_data(data[[3]]),
  "all"      = ignore_data(data), 
  check = FALSE
)
#> # A tibble: 4 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 mtcars       38.7µs   41.3µs    22758.    3.84KB     4.55
#> 2 mpg          38.1µs   40.6µs    23882.        0B     4.78
#> 3 diamonds     39.5µs   42.2µs    23048.        0B     4.61
#> 4 all         107.9µs    114µs     8594.      304B     4.08

ignored <- ignore_data(data)

bench::mark(
  "mtcars"   = expose_data(ignored[[1]]),
  "mpg"      = expose_data(ignored[[2]]),
  "diamonds" = expose_data(ignored[[3]]),
  "all"      = expose_data(ignored),
  check = FALSE
)
#> # A tibble: 4 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 mtcars       11.2µs   12.4µs    77564.        0B     7.76
#> 2 mpg          11.3µs   12.2µs    76183.        0B     0   
#> 3 diamonds     11.2µs   12.4µs    78313.        0B     7.83
#> 4 all          25.6µs   27.9µs    34930.        0B     3.49

^{Created on 2023-10-12 with reprex v2.0.2}

[thomasp85]

Can you update this for review?

[teunbrand]

I've updated this PR with the latest suggestions

[thomasp85]

LGTM

[teunbrand]

Thanks for the review!