Fix ordering in plyr compats...

[thomasp85]

Fixes #3313 and #3315

This fixes the differences in ordering between plyr and the replacement functions. Some notes

• the plyr ordering in ddply() is based on a pretty weird heuristic but I think I've nailed it
• the difference in join_keys() is the result of differences in rbind_dfs() and plyr::rbind.fill(). I honestly think the current ggplot2 approach is more correct, but I've made Date handling even with plyr to unbreak revdeps... If that fixes everything we should leave it at that instead of pursuing other parity with other weird vector formats

I probably don't have a head for more work this week, so bear with me if there are any problems with the PR

[yutannihilation]

It seems this test fails:

ggplot2/tests/testthat/test-layer.r

Lines 59 to 64 in 291cd7e

	test_that("if an aes is mapped to a function that returns NULL, it is removed", {
	df <- data_frame(x = 1:10)
	null <- function(...) NULL
	p <- cdata(ggplot(df, aes(x, null())))
	expect_identical(names(p[[1]]), c("PANEL", "x", "group"))
	})

Error:

names(p[[1]]) not identical to c("PANEL", "x", "group").
2/3 mismatches
x[1]: "x"
y[1]: "PANEL"

x[2]: "PANEL"
y[2]: "x"

This ordering was tweaked in #3013, so I guess you just need to revert this.

https://github.com/tidyverse/ggplot2/pull/3013/files#diff-12165842aeed0c5cb8c55e833132b58c

[clauswilke]

Is this the kind of global assignment that should be moved to .onLoad()?

[thomasp85]

Oh, yeah. Good catch 🙂

[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/