WISH: `drop` = FALSE by default for `[.data.frame` about wishlist-for-r

[TimTaylor]

To clarify - the "wish" is for when there is an explicit i. I know there are alternatives but I'd still like the default behaviour to be consistent and always return an object of the same type (data frame) irrespective of the length of j. Not, for example

rows <- 1:3
class(mtcars[rows, 1L])
#> [1] "numeric"
class(mtcars[rows, 1:2])
#> [1] "data.frame"

[gaborcsardi]

@chainsawriot Then people's packages start behaving differently for each user and/or script. The option already didn't work for stringsAsFactors, it is not going to work for this, either.

[karoliskoncevicius]

I noticed you only mentioned [.data.frame. Since data.frames are a hybrid between lists and matrices we can just use the list substitution [i], instead of the matrix way [,i]. So:

iris[1]
iris["Species"]
iris[c(1,2,3)]

All preserve data.frame without the need for drop=FALSE.

[ltuijnder]

As a package developer, I often have dynamic input in the j-slot that is user controlled. Forgetting the drop=FALSE can lead to nasty bugs as often the code afterward assumes the resulting object is still a data.frame.

[karoliskoncevicius]

@ltuijnder as mentioned above: using df[j] instead of df[,j] should solve this problem for you.

[InductiveStep]

This recently confused me as I was writing code iterating over the rows of data frames and sometimes the data frame only had one column.

Here's an example of what went wrong:

> data.frame(x = 42)[1,]
[1] 42

I didn't want it to lose its data.frame-ness. tibbles have sensible defaults:

> tibble(x = 42)[1,]
# A tibble: 1 × 1
      x
  <dbl>
1    42

[JosiahParry]

This became a fairly divisive topic on Mastodon. This behavior is not intuitive or user friendly. https://fosstodon.org/@josi/111096750116664099

[karoliskoncevicius]

This issue should be about matrices and arrays preserving dimensionality (and hence matrix/array objects) after [ selection. As it affects these objects first. And data.frame matrix-based selection would then follow too. Changing the behaviour only for data.frames would introduce an inconsistency between data.frame and matrix objects.

[chainsawriot]

As a possible interim solution is to make drop of current x[i, j, ... , drop = TRUE] to getOption("default.drop", TRUE), i.e. x[i, j, ... , drop = getOption("default.drop", TRUE)].

It can then be set per session (or per user, if one puts that in the initialization script): options(default.drop = FALSE).