Sort within x prior to stacking

[thomasp85]

This fixes #1552. collide now sorts the data by x and then by negated group. The negation is performed to ensure that the stacking order follows the default legend order of the fill.

[thomasp85]

To continue @karawoo's example, with this fix you get:

library("ggplot2")

## Create data
set.seed(40)
dat <- data.frame(x = rep(c(1:10), 3),
                  var = rep(c("a", "b", "c"), 10),
                  y = round(runif(30, 1, 5)))

## Plot
ggplot(dat, aes(x = x, y = y, fill = var)) +
  geom_area(position = "stack")

It is still not able to match stacking order if the scale has been manually modified:

ggplot(dat, aes(x = x, y = y, fill = var)) +
  geom_area(position = "stack") + 
  scale_fill_manual(breaks = c('c', 'a', 'b'), values = c('forestgreen', 'steelblue', 'firebrick'))

[thomasp85]

This is very related to #1691 so I'll try to solve it within the same PR with @hadley's blessing.

[thomasp85]

Continuing @hadley's example 00eedd1 allows you to mix positive and negative y-values when stacking so you get:

library(ggplot2)
library(tibble)

df <- frame_data(
  ~x, ~g, ~y,
  1,  1,  1,
  1,  2,  -1,
  1,  3,  1,
  2,  1,  2,
  2,  2,  -3
)

ggplot(df, aes(x, y, fill= factor(g))) + 
  geom_bar(stat = "identity")

Stacking order is as close to the default legend order as possible, but can get mixed up if one category jumps between negative and positive y-values

[hadley]

This looks good. I think it needs a few unit tests (so we don't accidentally break again in the future), and a section in position_stack() docs that describes the algorithm. Changing the order of the bars is such a common question that it's probably also worth mentioning in the docs for geom_bar().

[hadley]

Do we need a similar tweak for position_fill()?

[thomasp85]

I can't really wrap my head around how position fill should behave with negative values...

[hadley]

@thomasp85 hmmmm, I was thinking it we'd apply the same principle and fill separately for positive and negative

[thomasp85]

that would mean that positive and negative values received different scaling at the same point on the x-axis... And what if there are only negative values for some x? what to draw?

[hadley]

LGTM - just needs a news bullet that makes it clear this is a big change (and we'll need to highlight in the blog post)

[thomasp85]

@hadley do you want stronger wording/more emphasis in the news?

[thomasp85]

I've squashed it and will merge (and close issues) if you approve the news

[hadley]

Should be position_stack()

[hadley]

Looks good (apart from the small formatting issue). We'll definitely highlight this in the final release notes, and in the blog post.

[thomasp85]

How do you handle roxygenize? once, prior to release, or for every PR?

[hadley]

Every PR.

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/