Whatever comes out of these discussion: I think it would be good to add a blogpost like document to the default Jupyter documentations where this system is explained for "kernel developers", so that other can use that to model their own system after it...

I can build one after this text here: IRkernel/IRdisplay#3 (comment) (second part)

[Edit: added a issue here. https://github.com/jupyter/jupyter_client/issues/2]

since the discussion is hidden now:

the default for results is now everything.

do you think i should do the same for jupyter.plot_mimetypes?

i think it only makes sense to list those which actually have a repr_*.recordedplot method (excluding repr_text.recordedplot which i only wrote so that print(r.plot.instance) doesn’t do something silly)

I think the problem here is that if the plotting option is just "setting an option like "jupyter.plot_mimetypes" to a list", we miss out any validation/"translation" to the normal display formatter like the ipython equivalent does. Or is there a way to validate options when it is set via options(...)?

we could use the settings package (section “Using the settings package as options manager for your package”):

# not exported
MYPKGOPTIONS <- options_manager(a = 1, b = 2)

#' <documentation of options here>
#' @export
pkg_options <- function(...) {
    # protect against the use of reserved words.
    stop_if_reserved(...)
    # validate stuff
    opts <- list(...)
    if (!is.null(names(opts))) for (opt in names(opts)) switch(opt,
        a = stopifnot(condition),
        b = stopifnot(condition))
    # set or get
    MYPKGOPTIONS(...)
}

so, any objections to pulling this in?

I think this looks good.