---
title: "Introduction to V8 for R"
date: "`r Sys.Date()`"
output:
  html_document:
    fig_caption: false
    toc: true
    toc_float:
      collapsed: false
      smooth_scroll: false
    toc_depth: 3
vignette: >
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteIndexEntry{Introduction to V8 for R}
  \usepackage[utf8]{inputenc} 
---

```{r, echo = FALSE, message = FALSE}
knitr::opts_chunk$set(comment = "")
library(V8)
```


V8 is Google’s open source, high performance JavaScript engine. It is written in C++ and implements ECMAScript as specified in ECMA-262, 5th edition. The V8 R package builds on the C++ library to provide a completely standalone JavaScript engine within R:

```{r}
# Create a new context
ct <- v8()

# Evaluate some code
ct$eval("var foo = 123")
ct$eval("var bar = 456")
ct$eval("foo + bar")
```

A major advantage over the other foreign language interfaces is that V8 requires no compilers, external executables or other run-time dependencies. The entire engine is contained within a 6MB package (2MB zipped) and works on all major platforms.

```{r}
# Create some JSON
cat(ct$eval("JSON.stringify({x:Math.random()})"))

# Simple closure
ct$eval("(function(x){return x+1;})(123)")
```

However note that V8 by itself is just the naked JavaScript engine. Currently, there is no DOM (i.e. no *window* object), no network or disk IO, not even an event loop. Which is fine because we already have all of those in R. In this sense V8 resembles other foreign language interfaces such as Rcpp or rJava, but then for JavaScript.

## Loading JavaScript Libraries

The `ct$source` method is a convenience function for loading JavaScript libraries from a file or url.

```{r}
ct$source('https://cdnjs.cloudflare.com/ajax/libs/underscore.js/1.13.6/underscore-min.js')
ct$source("https://cdnjs.cloudflare.com/ajax/libs/crossfilter/1.3.11/crossfilter.min.js")
```

## Data Interchange

By default all data interchange between R and JavaScript happens via JSON using the bidirectional mapping implemented in the [jsonlite](https://arxiv.org/abs/1403.2805) package.

```{r}
ct$assign("mydata", mtcars)
ct$get("mydata")
```

Alternatively use `JS()` to assign the value of a JavaScript expression (without converting to JSON):

```{r}
ct$assign("foo", JS("function(x){return x*x}"))
ct$assign("bar", JS("foo(9)"))
ct$get("bar")
```

## Function Calls

The `ct$call` method calls a JavaScript function, automatically converting objects (arguments and return value) between R and JavaScript:

```{r}
ct$call("_.filter", mtcars, JS("function(x){return x.mpg < 15}"))
```

It looks a bit like `.Call` but then for JavaScript instead of C.

## Promises

If a call to `ct$eval()`, `ct$get()`, or `ct$call()` returns a JavaScript promise, you can set `await = TRUE` to wait for the promise to be resolved. It will then return the result of the promise, or an error in case the promise is rejected.

```{r error=TRUE}
js = 'function test_number(x){
  var promise = new Promise(function(resolve, reject) {
    if(x == 42)
      resolve(true)
    else
      reject("This is wrong")
  })
  return promise;
}'

# Call will just show a promise
ctx <- V8::v8()
ctx$eval(js)

# A promise does not return anything in itself:
ctx$call("test_number", 42)

# Resolve the promise to the result
ctx$call("test_number", 42, await = TRUE)

# A rejected promise will throw an error
ctx$call("test_number", 41, await = TRUE)
```

## Interactive JavaScript Console

A fun way to learn JavaScript or debug a session is by entering the interactive console:

```{r, eval=FALSE}
# Load some data
data(diamonds, package = "ggplot2")
ct$assign("diamonds", diamonds)
ct$console()
```

From here you can interactively work in JavaScript without typing `ct$eval` every time:

```javascript
var cf = crossfilter(diamonds)
var price = cf.dimension(function(x){return x.price})
var depth = cf.dimension(function(x){return x.depth})
price.filter([2000, 3000])
output = depth.top(10)
```

To exit the console, either press `ESC` or type `exit`. Afterwards you can retrieve the objects back into R:

```{r, eval=FALSE}
output <- ct$get("output")
print(output)
```


## warnings, errors and console.log

Evaluating invalid JavaScript code results in a SyntaxError:

```{r, error=TRUE, purl = FALSE}
# A common typo
ct$eval('var foo <- 123;')
```

JavaScript runtime exceptions are automatically propagated into R errors:

```{r, error=TRUE, purl = FALSE}
# Runtime errors
ct$eval("123 + doesnotexit")
```

Within JavaScript we can also call back to the R console manually using `console.log`, `console.warn` and `console.error`. This allows for explicitly generating output, warnings or errors from within a JavaScript application.

```{r, error = TRUE, purl = FALSE}
ct$eval('console.log("this is a message")')
ct$eval('console.warn("Heads up!")')
ct$eval('console.error("Oh no! An error!")')
```

A example of using `console.error` is to verify that external resources were loaded:

```{r, eval=FALSE}
ct <- v8()
ct$source("https://cdnjs.cloudflare.com/ajax/libs/crossfilter/1.3.11/crossfilter.min.js")
ct$eval('var cf = crossfilter || console.error("failed to load crossfilter!")')
```

## The Global Namespace

Unlike what you might be used to from Node or your browser, the global namespace for a new context is very minimal. By default it contains only a few objects: `global` (a reference to itself), `console` (for `console.log` and friends) and `print` (an alias of console.log needed by some JavaScript libraries)

```{r}
ct <- v8();
ct$get(JS("Object.keys(global)"))
```

A context always has a global scope, even when no name is set. When a context is initiated with `global = NULL`, it can still be reached by evaluating the `this` keyword within the global scope:

```{r}
ct2 <- v8(global = NULL, console = FALSE)
ct2$get(JS("Object.keys(this).length"))
ct2$assign("cars", cars)
ct2$eval("var foo = 123")
ct2$eval("function test(x){x+1}")
ct2$get(JS("Object.keys(this).length"))
ct2$get(JS("Object.keys(this)"))
```

To create your own global you could use something like:

```{r}
ct2$eval("var __global__ = this")
ct2$eval("(function(){var bar = [1,2,3,4]; __global__.bar = bar; })()")
ct2$get("bar")
```

## Syntax Validation

V8 also allows for validating JavaScript syntax, without actually evaluating it.

```{r}
ct$validate("function foo(x){2*x}")
ct$validate("foo = function(x){2*x}")
```

This might be useful for all those R libraries that generate browser graphics via templated JavaScript. Note that JavaScript does not allow for defining anonymous functions in the global scope:

```{r}
ct$validate("function(x){2*x}")
```

To check if an anonymous function is syntactically valid, prefix it with `!` or wrap in `()`. These are OK:
```{r}
ct$validate("(function(x){2*x})")
ct$validate("!function(x){2*x}")
```


## Callback To R

A recently added feature is to interact with R from within JavaScript using the `console.r` API`. This is most easily demonstrated via the interactive console. 

```{r, eval=FALSE}
ctx <- v8()
ctx$console()
```

From JavaScript we can read/write R objects via `console.r.get` and `console.r.assign`. The final argument is an optional list specifying arguments passed to `toJSON` or `fromJSON`.

```javascript
// read the iris object into JS
var iris = console.r.get("iris")
var iris_col = console.r.get("iris", {dataframe : "col"})

//write an object back to the R session
console.r.assign("iris2", iris)
console.r.assign("iris3", iris, {simplifyVector : false})
```

To call R functions use `console.r.call`. The first argument should be a string which evaluates to a function. The second argument contains a list of arguments passed to the function, similar to `do.call` in R. Both named and unnamed lists are supported. The return object is returned to JavaScript via JSON.

```javascript
//calls rnorm(n=2, mean=10, sd=5)
var out = console.r.call('rnorm', {n: 2,mean:10, sd:5})
var out = console.r.call('rnorm', [2, 20, 5])

//anonymous function
var out = console.r.call('function(x){x^2}', {x:12})
```

There is also an `console.r.eval` function, which evaluates some code. It takes only a single argument (the string to evaluate) and does not return anything. Output is printed to the console.

```javascript
console.r.eval('sessionInfo()')
```

Besides automatically converting objects, V8 also propagates exceptions between R, C++ and JavaScript up and down the stack. Hence you can catch R errors as JavaScript exceptions when calling an R function from JavaScript or vice versa. If nothing gets caught, exceptions bubble all the way up as R errors in your top-level R session.

```javascript
//raise an error in R
console.r.call('stop("ouch!")')

//catch error from JavaScript
try {
  console.r.call('stop("ouch!")')
} catch (e) {
  console.log("Uhoh R had an error: " + e)
}
//# Uhoh R had an error: ouch!
```