panelFn with return(NULL) breaks trelliscope when used with rbokeh

[jrounds]

Hi,

My hobbies include staying up late finding bugs because my other hobby is writing code that hits corner cases... I was quite excited to see the potential of rbokeh mixed with trelliscope unfortunately nothing happened for me when I went to do it! I eventually figured this out, but sorry no quick fix from me tonight =) Though I think the work around is simply to always plot something.

When the leading panelFn call for a ddo returns NULL and all the rest return rbokeh objects no panels display. However, if the leading panelFn returns an rbokeh object and a subsequent panelFn returns NULL only the NULL panel fails to display as it should be. There is however a squawk from trelliscope while displaying the NULL of:
"Error in system.file(config, package = package) :
'package' must be of length 1"

R code to recreate this issue is this and it shows both the bad behavior and the good behavior:

data(iris)
idata = divide(iris,by="Species")
i = 1
idata = lapply(idata, function(kv){
    new.value = list()
    new.value$data = kv$value
    new.value$i = i
    i <<- i + 1
    kv$value = new.value
    return(kv)
})
idata = ddo(idata)
badPanelFn = function(v) {
    if(v$i == 1)
        return(NULL)
    p = figure() %>% ly_points(Sepal.Length, Sepal.Width, data=v$data, hover=list(Sepal.Length))
    return(p)
}
panelFn(idata[[1]]$value)
bad = vdbConn("/tmp/bad", autoYes=TRUE)
makeDisplay(idata, panelFn=badPanelFn, conn=bad, name="name")
view(conn=bad)

goodPanelFn = function(v){
    if(v$i == 2)
        return(NULL)
    p = figure() %>% ly_points(Sepal.Length, Sepal.Width, data=v$data, hover=list(Sepal.Length))
    return(p)
}

good = vdbConn("/tmp/good", autoYes=TRUE)
makeDisplay(idata, panelFn=goodPanelFn, conn=good, name="name")
view(conn=good)

[jrounds]

So,

the "always plotting something" didn't quite work with mixing plot() and rbokeh, but using empty figure instead of returning NULL works.

For example this makes badPanelFn good:

badPanelFn = function(v) {
    if(v$i == 1)
        return(figure())
    p = figure() %>% ly_points(Sepal.Length, Sepal.Width, data=v$data, hover=list(Sepal.Length))
    return(p)
}

If you are wondering how this comes up. This is an error detection use-case. v is being inspected for appropriateness in the panel that is being applied. Occasionally v is inappropriate and the panel function needs to leave without plotting. In using rbokeh panels need to leave with return(figure()) or trelliscope fails to function properly. Which is fine when you know that =)