What is "shape", exactly?

[MichaelChirico]

vctrs/R/ptype-abbr-full.R

Lines 21 to 22 in 78d9f2b

	#' @param suffix_shape If `TRUE` (the default), append the shape of
	#' the vector.

I'm trying to write a vec_ptype_abbr() method for an S4 class and wondering if it's possible to define a method for determining its shape, but the ?vec_ptype_abbr documentation is not helping. Could this be better defined, possibly including a link for reference to another method?

[DavisVaughan]

I generally think about shape as meaningful for arrays. The way I think about it is "every dimension except the first one". So for a 2x3x4 array, the shape is [3, 4] and the size is 2.

> vctrs::vec_ptype_abbr(array(dim = c(2,3,4)))
[1] "lgl[,3,4]"
> vctrs::vec_size(array(dim = c(2,3,4)))
[1] 2

[DavisVaughan]

vec_ptype_shape gets at this

vctrs/R/ptype-abbr-full.R

Lines 133 to 144 in 78d9f2b

	vec_ptype_shape <- function(x) {
	dim <- dim2(x)
	if (length(dim) == 1) {
	if (is_null(dim(x))) {
	""
	} else {
	"[1d]"
	}
	} else {
	paste0("[,", paste(dim[-1], collapse = ","), "]")
	}
	}

We do seem to report [1d] for the case of 1-d arrays, but I think that's just trying to convey "an array, but it has no shape"

[DavisVaughan]

The other place shape comes into play sometimes is broadcasting vctrs:::shape_broadcast() which performs numpy style broadcasting on the shape of the array, used in vec_ptype2() for computing array common types

# the 1s are "broadcast" up to the other number 1->4, 1->3, see also numpy docs about broadcasting
dim(vctrs:::vec_ptype2(
  array(dim = c(0, 1, 3)), 
  array(dim = c(0, 4, 1))
))
[1] 0 4 3

Notably the shape is part of the ptype, but the size is not. That's why shape is a separate concept for us

[MichaelChirico]

Thanks for the explanation!

Is this solid enough to merit adding to the documentation? I saw vec_ptype_shape() but noted it's not exported / registered S3, so I couldn't (shouldn't) use it.

Your example of vec_ptype2() ~basically makes sense for arrays, but I'm not sure I understand well enough to apply it to more general pairs of objects, but also not sure it matters for my use case.

For my purposes (for now?), I'm only interested in improving how these list(<S4>) columns get printed in a tbl, so I'm just adding my own "shape suffix" to the end.

(feel free to close if you think there's nothing actionable left here)