Content in aes is warned But the proposed treatment is problematic (Apologize for the title)

[Breeze-Hu]

Hi, Teun!

I'm encountering the following when putting ggplot into a custom function:

Simply put, since the column names for x and y in aes change each time the structures aes(x = data[[1]],y = data[[2]]) are used, ggplot gives a warning. But when dealing with it based on ggplot's warnings, an error occurs straight away.

library(ggplot2)

set.seed(123)
data <- data.frame(a1 = rnorm(100),b1 = rnorm(100))  
#  Again there will be various combinations of scenarios: c(c2,d2), c(a3,e3) ...

ggplot(data = data, aes(x = data[[1]], y = data[[2]]))+
  geom_point()

This is able to draw the image, but a warning pops up:

Warning messages:
1: Use of data[[1]] is discouraged.
ℹ Use .data[[1]] instead.
2: Use of data[[2]] is discouraged.
ℹ Use .data[[2]] instead.

Doing as suggested in the warning message, using .data[[1]], etc., unfortunately runs the error.

ggplot(data = data, aes(x =.data[[1]], y = .data[[2]]))+
  geom_point()

Error in geom_point():
! Problem while computing aesthetics.
ℹ Error occurred in the 1st layer.
Caused by error in .data[[1]]:
! Must subset the data pronoun with a string, not the number 1.
Run rlang::last_trace() to see where the error occurred.

My question is: should the first behavior be warned (banned in a potential sense)? Or is there a chance that the latter could be realized by making code changes?

Best wishes,
Hu

[teunbrand]

Yeah using numeric indexing for selecting rows/columns is not preferred for aes().
I think the warning does a good job of steering people away from this behaviour, no?
It is fine to numerically index column names though.

library(ggplot2)

data <- mtcars

colnms <- colnames(data)

aes(x = .data[[colnms[1]]], y = .data[[colnms[2]]])
#> Aesthetic mapping: 
#> * `x` -> `.data[["mpg"]]`
#> * `y` -> `.data[["cyl"]]`

^{Created on 2025-06-27 with reprex v2.1.1}

[Breeze-Hu]

ohhh Thanks! This one was indeed a method I didn't know about.

And I noticed your related discussion in this location: https://stackoverflow.com/questions/70213108/ggplot2-warning-while-ploting-barplot-use-of-traini-is-discouraged-use

[Breeze-Hu]

Using names indexes is indeed the more elegant case.
But as in the case I mentioned, the column names changes in the loop, and that's when it's more convenient to use a positional index.
Anyway, the problem is solved! Thank you. 👍

[Breeze-Hu]

Warning messages:
1: Use of data[[1]] is discouraged.
ℹ Use .data[[1]] instead.

Alternatively, directing users to names is indeed a good job. What's puzzling is the content of the warning itself.
As in the question, I thought that by changing data[[1]] directly to .data[[1]], the code would run, however it reported an error. I have a feeling a lot of people are affected by this.

[teunbrand]

Yeah I can see that the order of the messages is somewhat confusion, but I'm not sure a lot of people are affected by this and thus would merit detecting numeric indices and throwing a tailored warning. In the current setting, you just have to go through 2 steps of correction.

[Breeze-Hu]

Yes. Modifying my own code is not complicated after the guidance you gave me. : )

[Breeze-Hu]

1: Use of dt[[1]] is discouraged.
ℹ Assuming the dataframe used is df, use .data[[colnames(df)[1]]] instead of df[[1]].

Would using such a case avoid introducing a lot of code to handle complex targeted warnings, while also providing more clarity for the user? The first line is a hint of what's out there right now. The second line is a can-always-be-used example.
Of course, I agree that this solution seems to partially forgo uniformity and neatness for utility.

[teunbrand]

The issue I'd have with changing this is that the current code we have for detecting such cases is already pretty complicated (and as such tricky to maintain) and piling on even more edge cases exacerbates that. So uniformity and neatness in terms of these messages is already out of the window, but I'm not looking to make it worse.