make preserve = "single" more robust in position_dodge().

[clauswilke]

This is an attempt at fixing #2801. Not sure this is ready to merge yet. One problem I see is that the violins are always moved to the left of the reference point, even if they should be moved to the right. Not sure how position_dodge() could know how to do this right.

Also, does a similar problem occur in position_dodge2()?

library(ggplot2)
p <- ggplot(mtcars, aes(factor(cyl), mpg))

# effect on geom_boxplot()
p + geom_boxplot(aes(fill = factor(vs)), position = "dodge")

p + geom_boxplot(aes(fill = factor(vs)), position = position_dodge(preserve = "single"))

# effect on geom_violin()
p + geom_violin(aes(fill = factor(vs)), position = "dodge")

p + geom_violin(aes(fill = factor(vs)), position = position_dodge(preserve = "single"))

Created on 2018-08-08 by the reprex package (v0.2.0).

[karawoo]

One problem I see is that the violins are always moved to the left of the reference point, even if they should be moved to the right. Not sure how position_dodge() could know how to do this right.

Yeah this has been reported for bars in the past (#2076) and I don't see a way to fix it in position_dodge().

Also, does a similar problem occur in position_dodge2()?

position_dodge2() doesn't work at all for violins or dot plots, at the time I wrote it we decided to punt on making it compatible with those geoms.

[hadley]

I'd rather not make any more position_dodge() changes without thoroughly thinking them through. This might be a project for @thomasp85 to take on — IIRC position_dodge() and position_dodge2() are quite slow, and I think the algorithms are more elegantly expressed in C/C++.

[clauswilke]

That's fine. Should I leave this PR open as a reminder or close? #2801 remains open.

[hadley]

Close, I think

[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/