Provide friendly_type?

[hadley]

It'd be nice to be able to write:

abort("`{.arg levels} must be a character vector, not a {.friendly levels}")

And know that the user is going to get something that makes sense to them.

[gaborcsardi]

Re the implementation, in general {.class ...} only does styling and not other evaluation and it also does not have information about the frame where the error was thrown. But we could have functions that return some cli::cli_vec() objects, that have a desired style, after the appropriate processing. We could add a class argument to cli_vec() to make it easier to return a vector with a cli class.

So the syntax would be just a regular function call:

abort("`{.arg levels} must be a character vector, not a {friendly_type(levels)}")

These functions also do not need to live in cli, btw. For the generic ones it makes sense to live there, but I imagine that some of the would be specific to a package, etc.

[hadley]

Oh yeah, this would be fine as a function too. I think this particular one belongs in cli because it's so needed in so many places.

[hadley]

Collection of places we've defined this before:

• ggplot2:::obj_desc (and soon to be copied in to forcats)
• rlang:::friendly_type_of
• R7::obj_desc

Somewhat related is vec_ptype_full(), but it's a precise description of the type, not something designed to be user friendly.

I suspect we'll want at least two versions, one for scalars, and one for vectors.