Why does a macro fail if passed constructed symbols as arguments?

This code works produces a nice chart:

using DataFrames, StatsPlots

df = DataFrame(A = rand(20), C = repeat([1,2,3, 4], inner = 5))

plt_dens1 = @df df density(
    group = :C,
    title = "This works"

If the Symbol arguments are constructed as below, code that is similar does not work. It generates an error message starting with:

"ERROR: MethodError: no method matching _extract_group_attributes(::Symbol, ::Symbol)" is generated."

asymb = Symbol("A")
csymb = Symbol("C")

plt_dens2 = @df df density(
    group = csymb,
    title = "This doesn't work"

Question: Does this mean Julia does not respect referential transparency?

Fortunately this code generates a chart just like the first one:

plt_dens3 = @df df density(
    df[:, Symbol("A")],
    group = df[:, Symbol("C")],
    title = "This also works"

However, it is a little more cumbersome.

Question: Why doesn’t the code for plt_dens2 work and is there a simple way to fix the problem?