layer_bars does not work if `x` and `y` are not initially specified in `ggvis` call

[ttzhou]

The following does not work:

library(magrittr)
library(ggvis)

mtcars %>%
ggvis() %>%
group_by(gear) %>%
layer_bars(
    x = ~factor(cyl),
    y = ~mpg,
    fill = ~factor(gear),
    width = 1
) %>% 
scale_ordinal('x')

Error is: Error in UseMethod("prop_label") : no applicable method for 'prop_label' applied to an object of class "NULL"

But the following does:

library(magrittr)
library(ggvis)

mtcars %>%
ggvis( #specifying the x and y here now, instead of later
    x = ~factor(cyl),
    y = ~mpg
) %>%
group_by(gear) %>%
layer_bars(
    fill = ~factor(gear),
    width = 1
) %>% 
scale_ordinal('x')

Even the following works:

library(magrittr)
library(ggvis)

mtcars %>%
ggvis(
    x = ~hp,
    y = ~mpg
) %>%
group_by(gear) %>%
layer_bars(
    x = ~factor(cyl),
    y = ~mpg,
    fill = ~factor(gear),
    width = 1
) %>% 
scale_ordinal('x')```

i.e. layer_bars works only if x and y are initially set in the initial ggvis call.

The source code for layer_bars seems to ask for 'x.update', which I'm guessing only works if x is initially defined.

This does not seem to be the case for layer_points, i.e.

library(magrittr)
library(ggvis)

mtcars %>%
ggvis() %>%
layer_points(
    x = ~factor(gear),
    y = ~mpg
) %>% 
scale_ordinal('x')

works fine.

In my opinion, this should be made consistent... if some marks allow for empty initial values for props, I think all of them should?

Also: thank you, to the entire team, for making such a great contribution to the data community. I know this isn't high on your priority list right now, but there are definitely people out there who love ggvis and are hoping for further development.

[ttzhou]

In layer_bars:

function (vis, ..., stack = TRUE, width = NULL) 
{
    new_props <- merge_props(cur_props(vis), props(...))
    check_unsupported_props(new_props, c("x", "y", "x2", "y2"), 
        c("enter", "exit", "hover"), "layer_bars")
    x_var <- find_prop_var(new_props, "x.update")
    discrete_x <- prop_countable(cur_data(vis), new_props$x.update)
    vis <- set_scale_label(vis, "x", prop_label(**cur_props**(vis)$x.update))
    ...
}

Would setting cur_props(vis) to new_props(vis) rectify this?