Adaptive rendering doesn't fully work

[jennybc]

I am following the latlon example in Printing vectors nicely in tibbles.
In particular, I want to do adaptive rendering.

In my real example, I have a character vector of ugly strings that mean almost nothing to humans (Google Drive file ids, but think: Git commit SHAs or similar).
If there is any sort of horizontal space shortage, this column should be heavily truncated. However I don't, in general, want to truncate this object. I don't want to truncate in a format method. Only inside a tibble.

I'll use the toy foo class to demonstrate.

library(tidyverse)
library(vctrs)
library(pillar)

new_foo <- function(x = character()) {
  vec_assert(x, character())
  new_vctr(x, class = "foo")
}

data <- tibble(
  month = month.name[1:3],
  sentences = new_foo(stringr::sentences[1:3]),
  blah = stringr::sentences[4:6]
)
data
#> # A tibble: 3 x 3
#>   month   sentences                                   blah                      
#>   <chr>   <foo>                                       <chr>                     
#> 1 January The birch canoe slid on the smooth planks.  These days a chicken leg …
#> 2 Februa… Glue the sheet to the dark blue background. Rice is often served in r…
#> 3 March   It's easy to tell the depth of a well.      The juice of lemons makes…
print(data, width = 60)
#> # A tibble: 3 x 3
#>   month  sentences                                   blah   
#>   <chr>  <foo>                                       <chr>  
#> 1 Janua… The birch canoe slid on the smooth planks.  These …
#> 2 Febru… Glue the sheet to the dark blue background. Rice i…
#> 3 March  It's easy to tell the depth of a well.      The ju…

Here we see that the foo column is hogging all the space, because it never gets truncated, because I haven't implemented any methods that would allow this.

If I implement a pillar_shaft method as the vignette shows, I do, in fact get truncation. However the horizontal space gained isn't redistributed to the other columns.

pillar_shaft.foo <- function(x, ...) {
  full <- format(x)
  trunc <- format(paste0(substr(x, 1, 7), cli::symbol$continue))
  pillar::new_pillar_shaft(
    list(full = full, trunc = trunc),
    width = pillar::get_max_extent(full),
    min_width = pillar::get_max_extent(trunc),
    class = "pillar_shaft_foo"
  )
}

format.pillar_shaft_foo <- function(x, width, ...) {
  if (pillar::get_max_extent(x$full) <= width) {
    ornament <- x$full
  } else {
    ornament <- x$trunc
  }

  pillar::new_ornament(ornament, align = "left")
}

data
#> # A tibble: 3 x 3
#>   month    sentences                             blah                              
#>   <chr>    <foo>                                 <chr>                             
#> 1 January  The bir…                              These days a chicken leg is a rar…
#> 2 February Glue th…                              Rice is often served in round bow…
#> 3 March    It's ea…                              The juice of lemons makes fine pu…
print(data, width = 60)
#> # A tibble: 3 x 3
#>   month   sentences                   blah                  
#>   <chr>   <foo>                       <chr>                 
#> 1 January The bir…                    These days a chicken …
#> 2 Februa… Glue th…                    Rice is often served …
#> 3 March   It's ea…                    The juice of lemons m…

Having poked around a bit, it seems like the logic in colonnade_distribute_space_df() isn't really taking advantage of the fact that the foo column only has width 8 if it is truncated at all.

Is there any way for me to do this?

If we can't display all columns fully, then foo should use its truncated form. And all the freed horizontal space should be available to the remaining columns.

[krlmlr]

Thanks. The issue here seems to be that any unused width is not rebalanced. I'm not sure if this even was the case with pillar 1.4.7 (needs a compatible tibble version from around that date).

---

author: kirill
date: 2021-06-27
output: "reprex::reprex\_document"
title: mid-guppy_reprex.R

library(tidyverse)
library(vctrs)
#> 
#> Attaching package: 'vctrs'
#> The following object is masked from 'package:dplyr':
#> 
#>     data_frame
#> The following object is masked from 'package:tibble':
#> 
#>     data_frame
library(pillar)
#> 
#> Attaching package: 'pillar'
#> The following object is masked from 'package:dplyr':
#> 
#>     dim_desc

new_foo <- function(x = character()) {
  vec_assert(x, character())
  new_vctr(x, class = "foo")
}

data <- tibble(
  month = month.name[1:3],
  sentences = new_foo(stringr::sentences[1:3]),
  blah = stringr::sentences[4:6]
)

pillar_shaft.foo <- function(x, ...) {
  full <- format(x)
  pillar::new_pillar_shaft(
    list(x = x, full = full),
    width = pillar::get_max_extent(full),
    min_width = 8,
    class = "pillar_shaft_foo"
  )
}

format.pillar_shaft_foo <- function(x, width, ...) {
  if (pillar::get_max_extent(x$full) <= width) {
    ornament <- x$full
  } else {
    ornament <- format(paste0(fansi::substr2_sgr(x$full, 1, width, type = "width") , cli::symbol$continue))
  }
  
  pillar::new_ornament(ornament, align = "left")
}

data
#> # A tibble: 3 x 3
#>   month    sentences                            blah                            
#>   <chr>    <foo>                                <chr>                           
#> 1 January  The birch canoe slid on the smooth p… These days a chicken leg is a r…
#> 2 February Glue the sheet to the dark blue back… Rice is often served in round b…
#> 3 March    It's easy to tell the depth of a wel… The juice of lemons makes fine …
print(data[-1], width = 60)
#> # A tibble: 3 x 2
#>   sentences                       blah                      
#>   <foo>                           <chr>                     
#> 1 The birch canoe slid on the smo… These days a chicken leg …
#> 2 Glue the sheet to the dark blue… Rice is often served in r…
#> 3 It's easy to tell the depth of … The juice of lemons makes…

^{Created on 2021-06-27 by the reprex package (v2.0.0)}

One solution would be to measure the actual width and then rebalance each tier so that columns are shown next to each other and still use the available width.

[krlmlr]

This seems consistent with your observations.

Also, in the original example the consumed width seems too wide, I'm not sure why width = 80 gives a width of 83 characters.

[krlmlr]

re width = 83: I counted #> 🤦

I think I have a good idea how to achieve this. In the first iteration, we rebalance each tier individually; later we could consider rebalancing across tiers to minimize surprise.

[krlmlr]

Very soon:

library(tidyverse)
library(vctrs)
library(pillar)

new_foo <- function(x = character()) {
  vec_assert(x, character())
  new_vctr(x, class = "foo")
}

data <- tibble(
  month = month.name[1:3],
  sentences = new_foo(stringr::sentences[1:3]),
  blah = stringr::sentences[4:6]
)

pillar_shaft.foo <- function(x, ...) {
  full <- format(x)
  trunc <- format(paste0(substr(x, 1, 7), cli::symbol$continue))
  pillar::new_pillar_shaft(
    list(full = full, trunc = trunc),
    width = pillar::get_max_extent(full),
    min_width = pillar::get_max_extent(trunc),
    class = "pillar_shaft_foo"
  )
}

format.pillar_shaft_foo <- function(x, width, ...) {
  if (pillar::get_max_extent(x$full) <= width) {
    ornament <- x$full
  } else {
    ornament <- x$trunc
  }

  pillar::new_ornament(ornament, align = "left")
}

data
#> # A tibble: 3 × 3
#>   month    sentences                                   blah                     
#>   <chr>    <foo>                                       <chr>                    
#> 1 January  The birch canoe slid on the smooth planks.  These days a chicken leg…
#> 2 February Glue the sheet to the dark blue background. Rice is often served in …
#> 3 March    It's easy to tell the depth of a well.      The juice of lemons make…
options(width = 50)
print(data)
#> # A tibble: 3 × 3
#>   month    sentences blah                         
#>   <chr>    <foo>     <chr>                        
#> 1 January  The bir…  These days a chicken leg is …
#> 2 February Glue th…  Rice is often served in roun…
#> 3 March    It's ea…  The juice of lemons makes fi…

^{Created on 2021-07-25 by the reprex package (v2.0.0.9000)}

This took way way longer than anticipated, but it's good for the codebase and for usability.

[jennybc]

Thanks, I look forward to taking advantage of this! In my life, at least (Drive ids, Git SHAs), I can think of many potential vctrs classes that are suitable for this sort of treatment.

This took way way longer than anticipated, but it's good for the codebase and for usability.

I thought it was really hard, so am pleasantly surprised to already have a fix.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.