bug: Class `meta` field should be expanded when converting ggplot class to a list

[schloerke]

Lines:

ggplot2/R/plot.R

Lines 278 to 285 in a2d0181

	#' @importFrom S7 convert
	S7::method(convert, list(from = class_ggplot, to = S7::class_list)) <-
	function(from, to) {
	S7::props(from)
	}
	S7::method(as.list, class_ggplot) <- function(x, ...) convert(x, S7::class_list)
	rm(`as.list`)

Reprex:

p <- ggplot(mpg, aes(cty, hwy)) + geom_point()
p$barret <- "Hello"
as.list(p)

Expecting: `list(..., barret = "Hello")

Received: list(..., meta = list(barret = "Hello"))

[teunbrand]

I agree with you in principle, especially if we want to advertise the meta field as a feature. I currently see the meta property as a temporary solution to bridge the gap between S3 and S7 until people write subclasses with their properties included. I'm trying to discourage its use in the docs as follows:

ggplot2/R/all-classes.R

Lines 121 to 122 in c495e9b

	#' @param meta A list for additional metadata. This will be deprecated in the
	#' future.

Is there a specific problem you'd like to use this for?

[schloerke]

No. But I needed it for GGally as it was expecting a flattened shape.

ggplot2 owns the as.list() method for class_ggplot.

I've fixed it in GGally with https://github.com/ggobi/ggally/blob/1a03dc5649ba386da076a7e20c2f7aad3cf3f9a8/R/ggmatrix.R#L207-L215

[teunbrand]

I think I'll close this issue because we don't want to encourage the S3 semantics of subassigning "properties" without actually building a true S7 property. Having the 'meta' property unfold in as.list() makes the S3 semantics slightly easier, so I think it'd be a step in the wrong direction. I'd consider it 'good friction' in the same sense that the painful way of creating secondary y-axes are 'good friction'.