Custom formatting

library(pillar)

How to customize the printed output of a "tbl" subclass? This vignette shows the various customization options. Customizing the formatting of a vector class in a tibble is described in vignette("pillar", package = "vctrs"). An overview over the control and data flow is given in vignette("printing").

This vignette assumes that the reader is familiar with S3 classes, methods, and inheritance. The “S3” chapter of Hadley Wickham’s “Advanced R” is a good start.

To make use of pillar’s printing capabilities, create a class that inherits from "tbl", like tibble (classes "tbl_df" and "tbl"), dbplyr lazy tables ("tbl_lazy" and "tbl") and sf spatial data frames ("sf", "tbl_df" and "tbl"). Because we are presenting various customization options, we create a constructor for an example data frame with arbitrary subclass.

example_tbl <- function(class) {
  vctrs::new_data_frame(
    list(
      a = letters[1:3],
      b = data.frame(c = 1:3, d = 4:6 + 0.5)
    ),
    class = c(class, "tbl")
  )
}

The "default" class doesn’t have any customizations yet, and prints like a regular tibble.

example_tbl("default")
#> $a
#> [1] "a" "b" "c"
#> 
#> $b
#>   c   d
#> 1 1 4.5
#> 2 2 5.5
#> 3 3 6.5
#> 
#> attr(,"class")
#> [1] "default"    "tbl"        "data.frame"

Row IDs

By implementing the generic ctl_new_rowid_pillar(), printing of the row ID column can be customized. In order to print Roman instead of Arabic numerals, one could use utils::as.roman() to generate the corresponding sequence and build up a row ID pillar using new_pillar() and associated methods as has been introduced previously.

ctl_new_rowid_pillar.pillar_roman <- function(controller, x, width, ...) {
  out <- NextMethod()
  rowid <- utils::as.roman(seq_len(nrow(x)))
  width <- max(nchar(as.character(rowid)))
  new_pillar(
    list(
      title = out$title,
      type = out$type,
      data = pillar_component(
        new_pillar_shaft(list(row_ids = rowid),
          width = width,
          class = "pillar_rif_shaft"
        )
      )
    ),
    width = width
  )
}

example_tbl("pillar_roman")
#> # A data frame: 3 × 2
#>     a       b$c    $d
#>     <chr> <int> <dbl>
#> I   a         1   4.5
#> II  b         2   5.5
#> III c         3   6.5

Body

Tweak pillar composition

Pillars consist of components, see ?new_pillar_component for details. Extend or override the ctl_new_pillar() method to alter the appearance. The example below adds table rules of constant width to the output.

ctl_new_pillar.pillar_rule <- function(controller, x, width, ..., title = NULL) {
  out <- NextMethod()
  new_pillar(list(
    top_rule = new_pillar_component(list("========"), width = 8),
    title = out$title,
    type = out$type,
    mid_rule = new_pillar_component(list("--------"), width = 8),
    data = out$data,
    bottom_rule = new_pillar_component(list("========"), width = 8)
  ))
}

example_tbl("pillar_rule")
#> # A data frame: 3 × 2
#>   ======== ======== ========
#>   a             b$c       $d
#>   <chr>       <int>    <dbl>
#>   -------- -------- --------
#> 1 a               1      4.5
#> 2 b               2      5.5
#> 3 c               3      6.5
#>   ======== ======== ========

To make the width adaptive, we implement a "rule" class with a format() method that formats rules to prespecified widths.

rule <- function(char = "-") {
  stopifnot(nchar(char) == 1)
  structure(char, class = "rule")
}

format.rule <- function(x, width, ...) {
  paste(rep(x, width), collapse = "")
}

ctl_new_pillar.pillar_rule_adaptive <- function(controller, x, width, ..., title = NULL) {
  out <- NextMethod()
  if (is.null(out)) {
    return(NULL)
  }

  new_pillar(list(
    top_rule = new_pillar_component(list(rule("=")), width = 1),
    title = out$title,
    type = out$type,
    mid_rule = new_pillar_component(list(rule("-")), width = 1),
    data = out$data,
    bottom_rule = new_pillar_component(list(rule("=")), width = 1)
  ))
}

example_tbl("pillar_rule_adaptive")
#> # A data frame: 3 × 2
#>   =         =     =
#>   a       b$c    $d
#>   <chr> <int> <dbl>
#>   -         -     -
#> 1 a         1   4.5
#> 2 b         2   5.5
#> 3 c         3   6.5
#>   =         =     =

Tweak display of compound pillars

Compound pillars are created by ctl_new_pillar_list() for columns that contain a data frame, a matrix or an array. The default implementation also calls ctl_new_pillar() shown above. The (somewhat artificial) example hides all data frame columns in a column with the type "<hidden>".

ctl_new_pillar_list.hide_df <- function(controller, x, width, ..., title = NULL) {
  if (!is.data.frame(x)) {
    return(NextMethod())
  }

  if (width < 8) {
    return(NULL)
  }

  list(new_pillar(
    list(
      title = pillar_component(new_pillar_title(title)),
      type = new_pillar_component(list("<hidden>"), width = 8),
      data = new_pillar_component(list(""), width = 1)
    ),
    width = 8
  ))
}

example_tbl("hide_df")
#> # A data frame: 3 × 2
#>   <hidden>
#>           
#> 1 <hidden>
#> 2         
#> 3 <hidden>

Restyle body

Last but not least, it is also possible to completely alter the display of the body by overriding tbl_format_body(). The example below uses plain data frame output for a tibble.

tbl_format_body.oldskool <- function(x, setup, ...) {
  capture.output(print.data.frame(setup$df))
}

print(example_tbl("oldskool"), n = 2)
#> # A data frame: 3 × 2
#>   a b.c b.d
#> 1 a   1 4.5
#> 2 b   2 5.5
#> # … with 1 more row

Note that default printed output is computed in tbl_format_setup(), this takes a considerable amount of time. If you really need to change the output for the entire body, consider providing your own tbl_format_setup() method.