@df: Passing column name as string overflows stack

[ron-wolf]

I was trying to plot this CSV dataset using StatsPlots.@df. I wrote and ran the following code in a fresh Pluto notebook:

using Downloads, CSV, DataFrames, StatsPlots
lifesat = CSV.read(Downloads.download("https://github.com/ageron/data/raw/main/lifesat/lifesat.csv"), DataFrame)
@df lifesat plot(cols("GDP per capita (USD)"), cols("Life satisfaction"))

I expected this to work, as per the README, but I was instead greeted by a StackOverflowError. Running again line-by-line in the Julia REPL yielded this stack trace:

 [1] add_sym!(cols::Vector{Symbol}, s::Char, names::Tuple{Symbol, Symbol, Symbol}) (repeats 79984 times)
   @ StatsPlots ~/.julia/packages/StatsPlots/cStOe/src/df.jl:199

However, this isn't the whole story, as expanding the macro by itself with macroexpand yields the following code (cleaned up):

(d -> begin
  (var1, var2), names = StatsPlots.extract_columns_and_names(d, "GDP per capita (USD)", "Life satisfaction")
  StatsPlots.add_label([StatsPlots.compute_name(names, "GDP per capita (USD)"), StatsPlots.compute_name(names, "Life satisfaction")], plot, var1, var2)
end)(lifesat)

And sure enough, attempting to call that first function extract_columns_and_names with those arguments yields the same error. The culprit is the following line of code in that function:

StatsPlots.jl/src/df.jl

Line 222 in bb65b7c

selected_cols = add_sym!(Symbol[], syms, names)

Which delegates to add_sym!, which recurses indefinitely over the first character of the first string, as follows:

add_sym!(Symbol[], "GDP per capita (USD)", (:Country, Symbol("GDP per capita (USD)"), Symbol("Life satisfaction")))
add_sym!(Symbol[], 'G', (:Country, Symbol("GDP per capita (USD)"), Symbol("Life satisfaction")))
add_sym!(Symbol[], 'G', (:Country, Symbol("GDP per capita (USD)"), Symbol("Life satisfaction")))
add_sym!(Symbol[], 'G', (:Country, Symbol("GDP per capita (USD)"), Symbol("Life satisfaction")))
# ...

The solution is to special-case strings in the definition for add_sym!, making sure not to iterate over their characters individually, but instead to convert them to symbols first.

[ron-wolf]

Also, as a side note, I had expected StatsPlots to show the label of the DataFrame column by default, but it did not: