Set width as aesthetic in geom_col()/geom_bar().

[teunbrand]

This PR aims to fix #3142.

Motivation is given here: #3142 (comment). To recap: yes variable width bars are bad, but we allowed them implicitly anyway and are jumping through hoops to clunkily discourage it.
This PR removes the hack that recognises width as a parameter, and implements it as an aesthetic instead.

[thomasp85]

LGTM

[katossky]

Not sure whether I should comment here or on the issue :

• this feature seems to solve the width issue but not the height one
• more importantly, it does not work as I expect

# devtools::install_github("https://github.com/tidyverse/ggplot2")
ggplot2::diamonds |>
    dplyr::summarise(n=dplyr::n(), avgPrice=mean(price), .by=clarity) |>
    ggplot(aes(x=clarity, width=n, y=avgPrice)) +
    geom_col(fill=NA, color='black')

produces :

I checked that I have the correct version of geom_bar and indeed :

> geom_bar
function (mapping = NULL, data = NULL, stat = "count", position = "stack", 
    ..., just = 0.5, na.rm = FALSE, orientation = NA, show.legend = NA, 
    inherit.aes = TRUE) ...

... does not have the width parameter anymore.

[teunbrand]

this feature seems to solve the width issue but not the height one

Can I ask what height issue exactly?

more importantly, it does not work as I expect

Running that code with the CRAN version or even older versions such as 3.4.4 yields the same plot, so I'm reasonably sure that this PR didn't affect that outcome.

[katossky]

height issue : geom_col does work with numeric x and discrete y ; it should maybe accept height in that case ?

[katossky]

For the other issue, I did not say that you break anything ! Just that the result does not fit my expectations.

If :

ggplot2::diamonds |>
    dplyr::summarise(n=dplyr::n(), avgPrice=mean(price), .by=clarity) |>
    ggplot(aes(x=clarity, y=avgPrice)) +
    geom_col(fill=NA, color='black')

... produces :

... then I do not expect the columns to get crumble together when I add a width aesthetic.

[teunbrand]

Just that the result does not fit my expectations

The n you compute ranges in the thousands, whereas the discrete x-axis places only takes 8. The labels crammed in the middle makes sense to me as the scale has to accommodate the thousands.

geom_col does work with numeric x and discrete y ; it should maybe accept height in that case ?

It does, and there is no need for height as width already controls the size of horizontal bars in the same way it does for vertical bars.

library(ggplot2)
ggplot(mtcars, aes(mpg, rownames(mtcars))) +
  geom_col(width = 0.5)

^{Created on 2024-05-23 with reprex v2.1.0}

[katossky]

So it was my understanding all along, sorry for that.

width indeed works with numeric x and discrete y (thank you) but I wrongly expected height there in alignment with geom_tile.

I was not expecting either width (a priori something quantitative) to share the units with x (a priori discrete) ! But I guess it may makes sense for consistency ?

ggplot2::diamonds |>
    dplyr::summarise(n=dplyr::n(), avgPrice=mean(price), .by=clarity) |>
    ggplot(aes(x=clarity, y=avgPrice, width = n / max(n))) +
    geom_col(fill=NA, color='black')

... indeed yields :

Last thing, in the preivous graph, I expected equal column spacing, irrespective of the width, as the only meaningful case I have in mind for variable-width column charts is this :

... but maybe I was wrong here again and that is not what you or anyone had in mind.

Thank you for your patience and clarification. Should I forward these comments somewhere where they are more useful ? Maybe am I not the only confused one ?