16.5 Advanced topics

This section covers several aspects of creating widgets that are not required by all widgets, but are an essential part of getting bindings to certain types of JavaScript libraries to work properly. Topics covered include:

  • Transforming JSON representations of R objects into representations required by JavaScript libraries (e.g., an R data frame to a D3 dataset).

  • Passing JavaScript functions from R to JavaScript (e.g., a user-provided formatting or drawing function)

  • Generating custom HTML to enclose a widget (the default is a <div>, but some libraries require a different element, e.g., a <span>).

  • Creating a widget without creating an R package in the first place.

16.5.1 Data transformation

R objects passed as part of the x parameter to the createWidget() function are transformed to JSON using the internal function htmlwidgets:::toJSON()14, which is basically a wrapper function of jsonlite::toJSON() by default. However, sometimes this representation is not what is required by the JavaScript library you are interfacing with. There are two JavaScript functions that you can use to transform the JSON data.

16.5.1.1 HTMLWidgets.dataframeToD3()

R data frames are represented in “long” form (an array of named vectors) whereas D3 typically requires “wide” form (an array of objects each of which includes all names and values). Since the R representation is smaller in size and much faster to transmit over the network, we create the long-form representation of R data, and then transform the data in JavaScript using the dataframeToD3() helper function.

Here is an example of the long-form representation of an R data frame:

{
  "Sepal.Length": [5.1, 4.9, 4.7],
  "Sepal.Width": [3.5, 3, 3.2],
  "Petal.Length": [1.4, 1.4, 1.3],
  "Petal.Width": [0.2, 0.2, 0.2],
  "Species": ["setosa", "setosa", "setosa"]
} 

After we apply HTMLWidgets.dataframeToD3(), it will become:

[
  {
    "Sepal.Length": 5.1,
    "Sepal.Width": 3.5,
    "Petal.Length": 1.4,
    "Petal.Width": 0.2,
    "Species": "setosa"
  },
  {
    "Sepal.Length": 4.9,
    "Sepal.Width": 3,
    "Petal.Length": 1.4,
    "Petal.Width": 0.2,
    "Species": "setosa"
  },
  {
    "Sepal.Length": 4.7,
    "Sepal.Width": 3.2,
    "Petal.Length": 1.3,
    "Petal.Width": 0.2,
    "Species": "setosa"
  }
] 

As a real example, the simpleNetwork (https://christophergandrud.github.io/networkD3/#simple) widget accepts a data frame containing network links on the R side, and transforms it to a D3 representation within the JavaScript renderValue function:

renderValue: function(x) {

  // convert links data frame to d3 friendly format
  var links = HTMLWidgets.dataframeToD3(x.links);

  // ... use the links, etc ...

}

16.5.1.2 HTMLWidgets.transposeArray2D()

Sometimes a 2-dimensional array requires a similar transposition. For this the transposeArray2D() function is provided. Here is an example array:

[
  [5.1, 4.9, 4.7, 4.6, 5, 5.4],
  [3.5, 3, 3.2, 3.1, 3.6, 3.9],
  [1.4, 1.4, 1.3, 1.5, 1.4, 1.7],
  [0.2, 0.2, 0.2, 0.2, 0.2, 0.4],
  ["setosa", "setosa", "setosa", "setosa", "setosa", "setosa"]
] 

HTMLWidgets.transposeArray2D() can transpose it to:

[
  [5.1, 3.5, 1.4, 0.2, "setosa"],
  [4.9, 3, 1.4, 0.2, "setosa"],
  [4.7, 3.2, 1.3, 0.2, "setosa"],
  [4.6, 3.1, 1.5, 0.2, "setosa"],
  [5, 3.6, 1.4, 0.2, "setosa"],
  [5.4, 3.9, 1.7, 0.4, "setosa"]
] 

As a real example, the dygraphs widget uses this function to transpose the “file” (data) argument it gets from the R side before passing it on to the dygraphs library:

renderValue: function(x) {

    // ... code excluded ...

    // transpose array
    x.attrs.file = HTMLWidgets.transposeArray2D(x.attrs.file);

    // ... more code excluded ...
}

16.5.1.3 Custom JSON serializer

You may find it necessary to customize the JSON serialization of widget data when the default serializer in htmlwidgets does not work in the way you have expected. For widget package authors, there are two levels of customization for the JSON serialization: you can either customize the default values of arguments for jsonlite::toJSON(), or just customize the whole function.

  1. jsonlite::toJSON() has a lot of arguments, and we have already changed some of its default values. Below is the JSON serializer we use in htmlwidgets at the moment:

    function(x, ..., dataframe = "columns", 
      null = "null", na = "null", auto_unbox = TRUE, 
      digits = getOption("shiny.json.digits", 
        16), use_signif = TRUE, force = TRUE, 
      POSIXt = "ISO8601", UTC = TRUE, rownames = FALSE, 
      keep_vec_names = TRUE, strict_atomic = TRUE) {
      if (strict_atomic) 
        x <- I(x)
      jsonlite::toJSON(x, dataframe = dataframe, 
        null = null, na = na, auto_unbox = auto_unbox, 
        digits = digits, use_signif = use_signif, 
        force = force, POSIXt = POSIXt, 
        UTC = UTC, rownames = rownames, 
        keep_vec_names = keep_vec_names, 
        json_verbatim = TRUE, ...)
    }

    For example, we convert data frames to JSON by columns instead of rows (the latter is jsonlite::toJSON’s default). If you want to change the default values of any arguments, you can attach an attribute TOJSON_ARGS to the widget data to be passed to createWidget(), e.g.,

    fooWidget = function(data, name, ...) {
      # ... process the data ...
      params = list(foo = data, bar = TRUE)
      # customize toJSON() argument values
      attr(params, 'TOJSON_ARGS') = list(
        digits = 7, na = 'string'
      )
      htmlwidgets::createWidget(name, x = params, ...)
    }

    We changed the default value of digits from 16 to 7, and na from null to string in the above example. It is up to you, the package author, whether you want to expose such customization to users. For example, you can leave an extra argument in your widget function so that users can customize the behavior of the JSON serializer:

    fooWidget = function(
      data, name, ..., JSONArgs = list(digits = 7)
    ) {
      # ... process the data ...
      params = list(foo = data, bar = TRUE)
      # customize toJSON() argument values
      attr(params, 'TOJSON_ARGS') = JSONArgs
      htmlwidgets::createWidget(name, x = params, ...)
    }

    You can also use a global option htmlwidgets.TOJSON_ARGS to customize the JSON serializer arguments for all widgets in the current R session, e.g.

    options(htmlwidgets.TOJSON_ARGS = list(
      digits = 7, pretty = TRUE
    ))
  2. If you do not want to use jsonlite, you can completely override the serializer function by attaching an attribute TOJSON_FUNC to the widget data, e.g.,

    fooWidget = function(data, name, ...) {
      # ... process the data ...
      params = list(foo = data, bar = TRUE)
      # customize the JSON serializer
      attr(params, 'TOJSON_FUNC') = MY_OWN_JSON_FUNCTION
      htmlwidgets::createWidget(name, x = params, ...)
    }

    Here MY_OWN_JSON_FUNCTION can be an arbitrary R function that converts R objects to JSON. If you have also specified the TOJSON_ARGS attribute, it will be passed to your custom JSON function, too.

16.5.2 Passing JavaScript functions

As you would expect, character vectors passed from R to JavaScript are converted to JavaScript strings. However, what if you want to allow users to provide custom JavaScript functions for formatting, drawing, or event handling? For this case, the htmlwidgets package includes a JS() function that allows you to request that a character value is evaluated as JavaScript when it is received on the client.

For example, the dygraphs widget (https://rstudio.github.io/dygraphs) includes a dyCallbacks function that allows the user to provide callback functions for a variety of contexts. These callbacks are “marked” as containing JavaScript so that they can be converted to actual JavaScript functions on the client:

library(dygraphs)
dyCallbacks(
  clickCallback = JS(...)
  drawCallback = JS(...)
  highlightCallback = JS(...)
  pointClickCallback = JS(...)
  underlayCallback = JS(...)
)

Another example is in the DT (DataTables) widget (https://rstudio.github.io/DT), where users can specify an initCallback with JavaScript to execute after the table is loaded and initialized:

datatable(head(iris, 20), options = list(
  initComplete = JS(
    "function(settings, json) {",
    "$(this.api().table().header()).css({
      'background-color': '#000',
      'color': '#fff'
     });",
    "}")
))

If multiple arguments are passed to JS() (as in the above example), they will be concatenated into a single string separated by \n.

16.5.3 Custom widget HTML

Typically the HTML “housing” for a widget is just a <div> element, and this is correspondingly the default behavior for new widgets that do not specify otherwise. However, sometimes you need a different element type. For example, the sparkline widget (https://github.com/htmlwidgets/sparkline) requires a <span> element, so it implements the following custom HTML generation function:

sparkline_html = function(id, style, class, ...){
  tags$span(id = id, class = class)
}

Note that this function is looked up within the package implementing the widget by the convention widgetname_html, so it need not be formally exported from your package or otherwise registered with htmlwidgets.

Most widgets will not need a custom HTML function, but if you need to generate custom HTML for your widget (e.g., you need an <input> or a <span> rather than a <div>), you should use the htmltools package (as demonstrated by the code above).

16.5.4 Create a widget without an R package

As we mentioned in Section 16.3, it is possible to create a widget without creating an R package in the first place. Below is an example:

#' @param text A character string.
#' @param interval A time interval (in seconds).
blink = function(text, interval = 1) {
  htmlwidgets::createWidget(
    'blink', list(text = text, interval = interval),
    dependencies = htmltools::htmlDependency(
      'blink', '0.1', src = c(href = ''), head = '
<script>
HTMLWidgets.widget({
  name: "blink",
  type: "output",
  factory: function(el, width, height) {
    return {
      renderValue: function(x) {
        setInterval(function() {
          el.innerText = el.innerText == "" ? x.text : "";
        }, x.interval * 1000);
      },
      resize: function(width, height) {}
    };
  }
});
</script>'
    )
  )
}

blink('Hello htmlwidgets!', .5)

The widget simply shows a blinking character string, and you can specify the time interval. The key of the implementation is the HTML dependency, in which we used the head argument to embed the JavaScript binding. The value of the src argument is a little hackish due to the current restrictions in htmltools (which might be removed in the future). In the renderValue method, we show or hide the text periodically using the JavaScript function setInterval().


  1. Note that it is not exported from htmlwidgets, so you are not supposed to call this function directly.