Key ordering vs. ordering of original columns with gather() - r

Question

Does key ordering depend on whether I first list the columns to gather vs. those not to gather?
This is my data.frame:
library(tidyr)
wide_df <- data.frame(c("a", "b"), c("oh", "ah"), c("bla", "ble"), stringsAsFactors = FALSE)
colnames(wide_df) <- c("first", "second", "third")
wide_df
first second third
1 a oh bla
2 b ah ble
First I gather all columns in a specific order, and my ordering is respected in the key listing as second, first, although the columns are actually ordered as first, second:
long_01_df <- gather(wide_df, my_key, my_value, second, first, third)
long_01_df
my_key my_value
1 second oh
2 second ah
3 first a
4 first b
5 third bla
6 third ble
Then I decide to exclude one column from gathering:
long_02_df <- gather(wide_df, my_key, my_value, second, first, -third)
long_02_df
third my_key my_value
1 bla second oh
2 ble second ah
3 bla first a
4 ble first b
The keys are again ordered as second, first. Then I code it like this, believing to be doing the exact same thing:
long_03_df <- gather(wide_df, my_key, my_value, -third, second, first)
long_03_df
And I get the keys ordered according to the real column order in the original data.frame:
third my_key my_value
1 bla first a
2 ble first b
3 bla second oh
4 ble second ah
This behavior does not even change, when I call the function with factor_key = TRUE. What I am missing?

Answer 1

Summary
The reason for this is that you can not mix negative and positive indices. (You also should not: it simply makes no sense.) If you do that, gather() will ignore some of the indices.
Detailed answer
Also for standard indexing you are not allowed to mix positive and negative indices:
x <- 1:10
x[c(4, -2)]
## Error in x[c(4, -2)] : only 0's may be mixed with negative subscripts
It makes sense that this is the case: Indexing with 4 tells R to only keep the fourth element. There is no need to tell it explicitly to throw away the second element in addition.
According to the documentation of gather(), selecting columns works the same way as in dplyr's select(). So let's play with that. I'll work with a subset of mtcars:
mtcars <- mtcars[1:2, 1:5]
mtcars
## mpg cyl disp hp drat
## Mazda RX4 21.0 6 160 110 3.90
## Mazda RX4 Wag 21.0 6 160 110 3.90
You can use positive and negative indexing with select():
select(mtcars, mpg, cyl)
## mpg cyl
## Mazda RX4 21 6
## Mazda RX4 Wag 21 6
select(mtcars, -mpg, -cyl)
## disp hp drat
## Mazda RX4 160 110 3.9
## Mazda RX4 Wag 160 110 3.9
Also for select(), mixing positive and negative indices makes no sense. But instead of throwing an error, select() seems to ignore all indices that have a different sign than the first one:
select(mtcars, mpg, -hp, cyl)
## mpg cyl
## Mazda RX4 21 6
## Mazda RX4 Wag 21 6
select(mtcars, -mpg, hp, -cyl)
## disp hp drat
## Mazda RX4 160 110 3.9
## Mazda RX4 Wag 160 110 3.9
As you can see, the results are exactly the same as before.
For your examples with gather(), you use these two lines:
long_02_df <- gather(wide_df, my_key, my_value, second, first, -third)
long_03_df <- gather(wide_df, my_key, my_value, -third, second, first)
According to what I've shown above, these lines are identical to:
long_02_df <- gather(wide_df, my_key, my_value, second, first)
long_03_df <- gather(wide_df, my_key, my_value, -third)
Note that there is nothing in the second line that would indicate your preferred ordering of the keys. It only says that third should be omitted.