`layer()$constructor` can be misleading

[yjunechoe]

This is a classic srcref problem affecting readability when printing $constructor to console

library(ggplot2)

p_expr <- parse(text = "
  p <- ggplot() +
    geom_point(size = 5)
")
eval(p_expr)

# Whole plot code displayed on print
p$layers[[1]]$constructor
#> p <- ggplot() +
#>     geom_point(size = 5)

# Stripping srcref shows the actual layer code
attr(p$layers[[1]]$constructor, "srcref") <- NULL
p$layers[[1]]$constructor
#> geom_point(size = 5)

A simple fix would be to apply this one-liner where the constructor property is resolved in layer():

ggplot2/R/layer.R

Line 183 in 7fb5760

fr_call <- layer_class$constructor %||% frame_call(call_env) %||% current_call()

Stripping the srcref of fr_call above would naturally resolve one other case where this happens (in layer_sf(), since it gets forwarded to layer())

ggplot2/R/layer-sf.R

Lines 25 to 28 in 7fb5760

	layer_class <- ggproto(NULL, LayerSf,
	constructor = frame_call(call_env),
	legend_key_type = legend_key_type
	)

If that seems reasonable, I can file a PR! :)

[teunbrand]

Hi June!

The reason the constructor call is kept is for the use in error messaging, so in the case you point out the stripped call would be more informative. Happy to review a PR!