Deprecate .data in data-expressions

[lionel-]

To make programming with data-masking and programming with selections completely distinct.

[lionel-]

Do we need an all_of() variant that expects exactly 1 variable?

[dblodgett-usgs]

Out of curiosity, was the fact that this pattern has been recommended in this vignette https://dplyr.tidyverse.org/articles/programming.html for quite some time taken into account when this change was made? It seems like it's causing a lot of people to go changing a lot of code ... is it really needed?

[hadley]

@dblodgett-usgs yes

[dblodgett-usgs]

Well ok then, thanks I guess. It appears the documentation has actually changed at that vignette?

I see you committed a change ~20 days ago but it only deployed to github pages two days ago. Reading that now, it does make more sense... it just seemed strange because I checked that vignette and it hadn't changed.

[chrarnold]

I'd like to know what the best way is regarding this change when it comes to notes during R CMD check? I recently added .data$ to eliminate the (many, many) notes during package checking, but with the newest package versions, look like I need to revert back. Couldnt find it anywhere or I oversaw? Thanks!

[hadley]

The warning message should tell you what to do:

out <- mtcars |> dplyr::select(.data$cyl)
#> Warning: Use of .data in tidyselect expressions was deprecated in tidyselect 1.2.0.
#> ℹ Please use `"cyl"` instead of `.data$cyl`

^{Created on 2022-12-08 with reprex v2.0.2}

[chrarnold]

Sorry maybe I was not clear, the warning message is absolutely clear. However, when building packages, R CMD check prints the almighty famous "no visible binding for global variable" note for each instance when not using the .data$ prefix. Before this change, I found the .data$ construct to be the best to avoid this note, I didnt like defining dozens of global variables. Is there any updated recommendation on this in light of this change?

[dblodgett-usgs]

@chrarnold -- this article was updated recently and has a pretty thorough summary of the details on this. https://dplyr.tidyverse.org/articles/programming.html#data-masking In short, use .data$foo for things that do not use tidyselect and "foo" for things that do use tidyselect. Which function is throwing the error should be a guide but for me it was basically select/rename use "foo" and filter/mutate/group_by use .data$foo. See DOI-USGS/nhdplusTools#306 for my clean up issue on this if it helps.

[lionel-]

@chrarnold If you use dplyr::select("cyl") (with quotes) instead of dplyr::select(.data$cyl), there will be no warnings.

[chrarnold]

Thanks @lionel- and @dblodgett-usgs , I didnt know dplyr::select("cyl") can be used, does this work only for newer versions of dplyr or this was already supported for older versions? For the package, I have to find an implementation that is as generically working as possible without having to require particular package versions etc if possible.

@dblodgett-usgs In a tidyselect context, wouldnt the R CMD check notes still be there then? I really want to get rid of these notes, having hundreds of them bloats up the output so much

[lionel-]

This was also supported in old versions of dplyr.

@dblodgett-usgs In a tidyselect context, wouldnt the R CMD check notes still be there then?

No because in tidyselect contexts you now use quoted names like "foo" as @dblodgett-usgs mentioned.

[chrarnold]

Thanks, I have not read the "new" quoted recommendation anywhere else to be honest, also not in the quoted (and updated) https://dplyr.tidyverse.org/articles/programming.html#data-masking article. I'll update my code correspondingly!

[lionel-]

@chrarnold To be clear, in data-masking (as in the link you just shared), a string represents a string, not a column. So you can't do things like 1 + "cyl" because you'd be adding a number to a string.

It's only for column selection that you can refer to a column with a string.

[shahronak47]

@lionel- Then what is the recommended way to replace .data in mutate ? Since .data is deprecated we can't use mtcars %>% mutate(new = 1 + .data$cyl) or mtcars %>% mutate(new = 1 + .data[["cyl"]])

[lionel-]

.data is not deprecated in mutate().

It's only been deprecated in tidy selections, e.g. in contexts like select(...) or pivot_longer(cols = ).