Improve performance of render_html()

[mgirlich]

Summary

Rendering slightly bigger tables (e.g. 50k rows) as HTML is very slow. The two main culprits are create_body_component_h() and {htmltools} (in particular tagWrite() but also creating the tags themself).

Two small benchmarks

diamonds_1000 <- ggplot2::diamonds |>
  head(1000) |>
  gt::gt()
diamonds <- ggplot2::diamonds |>
  gt::gt()

prepare <- function(data) {
  data <- gt:::build_data(data = data, context = "html")
  data <- gt:::add_css_styles(data = data)
  gt:::create_body_component_h(data = data)
}

bench::mark(
  diamonds_1000 = prepare(diamonds_1000),
  diamonds = prepare(diamonds),
  check = FALSE
)

# CRAN
# # A tibble: 2 × 13
#   expression        min median `itr/sec` mem_alloc `gc/sec` n_itr
#   <bch:expr>    <bch:t> <bch:>     <dbl> <bch:byt>    <dbl> <int>
# 1 diamonds_1000   1.14s  1.14s    0.878     5.37MB     2.63     1
# 2 diamonds       56.42s 56.42s    0.0177  175.21MB     2.11     1

# This PR
# # A tibble: 2 × 13
#   expression         min   median `itr/sec` mem_alloc `gc/sec` n_itr
#   <bch:expr>    <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl> <int>
# 1 diamonds_1000 146.87ms 148.81ms     6.33     7.72MB     6.33     4
# 2 diamonds         7.46s    7.46s     0.134  301.76MB     6.03     1

Checklist

• I understand and agree to the Code of Conduct.
• I have listed any major changes in the NEWS.
• I have added testthat unit tests to tests/testthat for any new functionality.

Notes

create_body_component_h() is quite huge. I think it would make sense to split parts of it into separate functions, e.g. a function for the group headings, the summaries, the data itself. I didn't do this as I didn't want to change the style too much.

The PR is probably quite a bit easier to understand by looking at the commits one by one. Otherwise the changes are hard to follow.

There are a couple of places in render_row_data() that deal with spaces and creating HTML. This feels quite tedious and error prone. It would be great to use {htmltools} for such a task but this would require to increase the performance of {htmltools} much more.

I also opened an issue in {htmltools} regarding the performance

[CLAassistant]

All committers have signed the CLA.

[mgirlich]

Using vectors instead of lists is much nicer to work with and also faster.

[mgirlich]

Accessing fields from a list is much faster than fields of a data frame

[rich-iannone]

@mgirlich This is great work and I really appreciate it!

The only problem I can see is that adding vtcrs as a dependency would bring the total number to 21, which is something that CRAN flags during updates (and ultimately won't, to my knowledge, accept). However, the tibble dependency could be removed with some minor, extra work. If you'd be willing to replace the two instances where tibble is used, then the vtcrs addition will be okay.

From a quick search, tibble functions are used here:

• https://github.com/rstudio/gt/blob/9c6a46c84c05a111bf9b62e8f07222d2dd79029b/R/utils_render_rtf.R#L415C21-L415C21
• https://github.com/rstudio/gt/blob/9c6a46c84c05a111bf9b62e8f07222d2dd79029b/R/z_utils_render_footnotes.R#L357C5-L357C5

[rich-iannone]

lgtm!