Vignette: discuss optional string input

[teunbrand]

This PR aims to fix #6208.

If somebody has better advice for dealing with optional character-input, I'd be glad to adjust.

[yjunechoe]

Tangentially related, but I wonder whether rlang::data_sym() passing NULL through could resolve some of this headache:

rlang::data_sym(NULL)
#> NULL

Instead of its current behavior:

rlang::data_sym(NULL)
#> Error in `sym()`:
#> ! Can't convert `NULL` to a symbol.

Especially because I feel like this "optional character-input" problem has already been solved in tidyselect (with all_of()/any_of()), so it'd be nice if data_sym()/data_syms() could let us get something close to tidyselect semantics within tidyeval contexts (like aes()) by letting NULL pass through to mean an empty (NULL) selection. Ex:

df <- data.frame(x = 1)

# Tidyselect behavior
tidyselect_get <- function(df, col) {
  tidyselect::eval_select(tidyselect::all_of(col), df)
}
tidyselect_get(df, "x")
#> x 
#> 1
tidyselect_get(df, NULL)
#> named integer(0)

# Tidyeval with `data_sym()`
tidyeval_get <- function(df, col) {
  rlang::eval_tidy(rlang::data_sym(col), df)
}
tidyeval_get(df, "x")
#> [1] 1
tidyeval_get(df, NULL)
#> Error in `sym()`:
#> ! Can't convert `NULL` to a symbol.

# Tidyeval with modified `data_sym2()` that passes NULL through
data_sym2 <- \(x) if (!is.null(x)) rlang::data_sym(x)
tidyeval_get2 <- function(df, col) {
  rlang::eval_tidy(data_sym2(col), df)
}
tidyeval_get2(df, "x")
#> [1] 1
tidyeval_get2(df, NULL)
#> NULL

[teunbrand]

Thanks June, your comments make a lot of sense to me. I'd agree that this having data_sym() accept NULL would be the most elegant. I don't think ggplot2 should provide a user-exposed wrapper for this, as we'd have to maintain this into perpetuity. Rather, we could petition {rlang} if they'd be interested in this change. That allows us to include this PR as current advice and update the advice if and once we can use data_sym(NULL).

[yjunechoe]

Rather, we could petition {rlang} if they'd be interested in this change.

Yeah, that's the intended build-up of my comment! Would be nice for the FR to be backed by a need in ggplot 😄

[thomasp85]

I'm personally partial to advocating embrace for people wanting to build ggplot2 wrapper functions. The issue noted in the original issue (name of column coming as a string) is easily solved:

plot_embrace <- function(dat,xvar,yvar,colorvar=NULL,shapevar=NULL) {
  myplot <- ggplot(dat,aes(x={{ xvar }},y={{ yvar }},color={{ colorvar }},shape={{ shapevar }})) +
    geom_point()
  return(myplot)
}

whichvar <- sample(c('sex','employment'),1)
plot_embrace(plotdat,weight,height,colorvar=age,shapevar=!!sym(whichvar))

[teunbrand]

I'm happy to recommend the !!sym() trick, but it'd be a different approach than this PR, so I'll create a new one.