A tiny wrapper around Node.js streams.Transform (Streams2/3) to avoid explicit subclassing noise

Inspired by Dominic Tarr's through in that it's so much easier to make a stream out of a function than it is to set up the prototype chain properly: through(function (chunk) { ... }).

fs.createReadStream('ex.txt')
  .pipe(through2(function (chunk, enc, callback) {
    for (let i = 0; i < chunk.length; i  )
      if (chunk[i] == 97)
        chunk[i] = 122 // swap 'a' for 'z'

    this.push(chunk)

    callback()
   }))
  .pipe(fs.createWriteStream('out.txt'))
  .on('finish', () => doSomethingSpecial())

Or object streams:

const all = []

fs.createReadStream('data.csv')
  .pipe(csv2())
  .pipe(through2.obj(function (chunk, enc, callback) {
    const data = {
        name    : chunk[0]
      , address : chunk[3]
      , phone   : chunk[10]
    }
    this.push(data)

    callback()
  }))
  .on('data', (data) => {
    all.push(data)
  })
  .on('end', () => {
    doSomethingSpecial(all)
  })

Note that through2.obj(fn) is a convenience wrapper around through2({ objectMode: true }, fn).

Since Node.js introduced Simplified Stream Construction, many uses of through2 have become redundant. Consider whether you really need to use through2 or just want to use the 'readable-stream' package, or the core 'stream' package (which is derived from 'readable-stream'):

const { Transform } = require('readable-stream')

const transformer = new Transform({
  transform(chunk, enc, callback) {
    // ...
  }
})

through2([ options, ] [ transformFunction ] [, flushFunction ])

Consult the stream.Transform documentation for the exact rules of the transformFunction (i.e. this._transform) and the optional flushFunction (i.e. this._flush).

The options argument is optional and is passed straight through to stream.Transform. So you can use objectMode:true if you are processing non-binary streams (or just use through2.obj()).

The options argument is first, unlike standard convention, because if I'm passing in an anonymous function then I'd prefer for the options argument to not get lost at the end of the call:

fs.createReadStream('/tmp/important.dat')
  .pipe(through2({ objectMode: true, allowHalfOpen: false },
    (chunk, enc, cb) => {
      cb(null, 'wut?') // note we can use the second argument on the callback
                       // to provide data as an alternative to this.push('wut?')
    }
  ))
  .pipe(fs.createWriteStream('/tmp/wut.txt'))

The transformFunction must have the following signature: function (chunk, encoding, callback) {}. A minimal implementation should call the callback function to indicate that the transformation is done, even if that transformation means discarding the chunk.

To queue a new chunk, call this.push(chunk)—this can be called as many times as required before the callback() if you have multiple pieces to send on.

Alternatively, you may use callback(err, chunk) as shorthand for emitting a single chunk or an error.

If you do not provide a transformFunction then you will get a simple pass-through stream.

The optional flushFunction is provided as the last argument (2nd or 3rd, depending on whether you've supplied options) is called just prior to the stream ending. Can be used to finish up any processing that may be in progress.

fs.createReadStream('/tmp/important.dat')
  .pipe(through2(
    (chunk, enc, cb) => cb(null, chunk), // transform is a noop
    function (cb) { // flush function
      this.push('tacking on an extra buffer to the end');
      cb();
    }
  ))
  .pipe(fs.createWriteStream('/tmp/wut.txt'));

through2.ctor([ options, ] transformFunction[, flushFunction ])

Instead of returning a stream.Transform instance, through2.ctor() returns a constructor for a custom Transform. This is useful when you want to use the same transform logic in multiple instances.

const FToC = through2.ctor({objectMode: true}, function (record, encoding, callback) {
  if (record.temp != null && record.unit == "F") {
    record.temp = ( ( record.temp - 32 ) * 5 ) / 9
    record.unit = "C"
  }
  this.push(record)
  callback()
})

// Create instances of FToC like so:
const converter = new FToC()
// Or:
const converter = FToC()
// Or specify/override options when you instantiate, if you prefer:
const converter = FToC({objectMode: true})

through2 is Copyright © Rod Vagg and additional contributors and licensed under the MIT license. All rights not explicitly granted in the MIT license are reserved. See the included LICENSE file for more details.