Provide support for multiple outputs

[rpkyle]

This PR provides support for multiple outputs, as originally described in plotly/dash#436. Several changes are proposed:

• a createCallbackId function is introduced to simplify generating output IDs for either single outputs (container.style) or multiple outputs (..text-box.children...container.style..)
• an insertIntoCallbackMap function is also introduced to add additional callback handlers into the map
• callback handlers may now return values which are targeted at multiple callbacks, and these targets are inferred solely from the sequence of the output declaration (i.e. no names are required for the return object elements)
• partial updates are properly supported
• modifications to CircleCI test config to install the correct package versions for a given plotly/dashR version, and to avoid repeatedly launching R for package installs

The following checks are now performed within Dash for R:

• validateOutput now checks the ID formatting for single and multiple outputs
• assert_valid_callbacks now checks for circular inputs and outputs for multiple outputs also
• insertIntoCallbackMap checks for duplicated outputs across callbacks before allowing a new callback to be added to the callback map

An additional test is also proposed to ensure that multiple outputs are functioning properly, and that partial outputs are allowed.

Finally, the following checks are implemented as in Dash for Python:

• Ensuring that no output is specified twice #f1f4724
• Ensuring that no output is used in two overlapping multi-output sets (e.g. you cannot have ..o1.value...o2.value.. with ..o1.value...o3.value..) #508eeb1
• Ensuring that both methods of supplying an output parameter are working (e.g. using output=list(id='graph', property='figure') or output=output(id='graph', property='figure')) #5521a93

Closes#114.

[rpkyle]

A sample app is provided here:

library(dash)
library(dashHtmlComponents)
library(dashCoreComponents)
library(plotly)
library(dashTable)
sample_data <- list(
series = list(
data = list(
list(title = 'Game of Thrones', score = 9.5),
list(title = 'Stranger Things', score = 8.9),
list(title = 'Vikings', score = 8.6)
),
style = list(
backgroundColor = '#ff998a'
)
),
movies = list(
data = list(
list(title = 'Rambo', score = 7.7),
list(title = 'The Terminator', score = 8.0),
list(title = 'Alien', score = 8.5)
),
style = list(
backgroundColor = '#fff289'
)
)
)
app <- Dash$new() app$layout(
htmlDiv(list(
htmlH1('Multi output example'),
dccDropdown(id='data-dropdown',
options = list(
list(label = 'Movies',
value = 'movies'),
list(label = 'Series',
value = 'series')
),
value = 'movies'),
htmlDiv(list(
dccGraph(id = 'graph'),
dashDataTable(id = 'data-table',
columns = list(
list(name = 'Title',
id = 'title'),
list(name = 'Score',
id = 'score')
)
)
)
)
),
id = 'container'
)
)
app$callback(output=list(
output(id='graph', property='figure'),
output(id='data-table', property='data'),
output(id='data-table', property='columns'),
output(id='container', property='style')
),
params=list(
input(id='data-dropdown', property='value')
),
function(value) {
if (is.null(value)) {
return(dashNoUpdate())
}
selected <- sample_data[[value]]
subdata <- selected[['data']]
columns <- list(
list(
name = 'Title',
id = 'title'
),
list(
name = 'Score',
id = 'score'
)
)
figure <- plot_ly(
x = vapply(subdata, function(x) x$score, numeric(1)),
text = vapply(subdata, function(x) x$title, character(1)),
name = vapply(subdata, function(x) x$title, character(1)),
type = "bar"
)
browser()
return(list(figure,
subdata,
columns,
selected[['style']]))
})
app$run_server(debug=TRUE)

[rpkyle]

Finishing up tests for the second and third checked items above, but I believe the current version of this code satisfies all requirements noted here.

[rpkyle]

@byronz I made some minor changes to the CircleCI config in 30ba139:

• R package dependencies are now installed on one line, so we don't repeatedly launch R to complete all package installs before the tests start
• we install plotly/dashR instead of the core packages, since the latter approach causes us to grab versions that are not locked to the current Dash for R release

[rpkyle]

@alexcjohnson I've implemented the unit and integration tests for all features/scenarios requested, except for the most recent one you asked about, which I'll do today.

[rpkyle]

@alexcjohnson OK, test has been added with two DashNoUpdate() returns in the same callback.

[alexcjohnson]

Very nice! Just a small note about commenting the test, then we're ready to go! 💃