.data pronoun speedup

[teunbrand]

This PR aims to fix #5730.

Briefly, using the .data pronoun caused the plot to slow down by a non-trivial amount of time. This PR rectifies this by considering the pronoun as 'non-data'. To explain a bit further, here follows the diagnosis.

unrowname() throws errors with pronouns:

x <- rlang::eval_tidy(quote(.data), mtcars)
print(x)
#> <pronoun>

ggplot2:::unrowname(x)
#> Error in `ggplot2:::unrowname()`:
#> ! Can only remove rownames from <data.frame> and <matrix> objects.

^{Created on 2024-02-29 with reprex v2.1.0}

And unrowname() is used to compare to see if a part of an expression evaluates to the plot data here:

ggplot2/R/aes.R

Lines 437 to 440 in bc3e401

	tryCatch({
	data_eval <- eval_tidy(x[[2]], data, env)
	identical(unrowname(data_eval), unrowname(data))
	}, error = function(err) FALSE)

As this is in a tryCatch() block, the errors that unrowname() throws for pronouns are silenced. However, formatting the error message is still relatively expensive.

In this PR, we circumvent this problem by recognising that the .data pronoun is not the same as the regular plot data, thereby bypassing the check in the lines above.

As a benchmark, we see that typical evaluation and with pronoun are in the same range with this PR:

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

df <- mtcars
df$name <- rownames(mtcars)

p <- ggplot(df) +
  geom_point() +
  geom_text() +
  facet_grid(gear ~ cyl)

standard <- p + aes(x = mpg, y = disp, label = name, colour = factor(cyl))
pronoun  <- p + aes(
  x = .data[["mpg"]], 
  y = .data[["disp"]], 
  label = .data[["name"]],
  colour = factor(.data[["cyl"]])
)

bench::mark(
  ggplot_build(standard),
  ggplot_build(pronoun),
  check = FALSE, min_iterations = 10
)
#> # A tibble: 2 × 6
#>   expression                  min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>             <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 ggplot_build(standard)   67.4ms   68.4ms      14.7    6.17MB    14.7 
#> 2 ggplot_build(pronoun)      68ms   69.4ms      14.4    1.96MB     9.61

^{Created on 2024-02-29 with reprex v2.1.0}

For reference, this is the same benchmark result, but on the current main branch:

#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 2 × 6
#>   expression                  min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>             <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 ggplot_build(standard)   67.5ms   71.4ms     13.8     6.17MB     6.91
#> 2 ggplot_build(pronoun)     175ms  180.1ms      5.52    4.48MB     5.52

[teunbrand]

Couldn't find anything in rlang like is_pronoun() so we only escape .data here and not .env. I don't think .env is very common.

[thomasp85]

LGTM

This is perhaps the most efficient diff in terms of character changed to performance gained :-)

[aphalo]

Thanks! You were extremely fast at fixing this!
Best wishes,
Pedro.