Description
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 

7.0 KiB

stream-http Build Status

Sauce Test Status

This module is an implementation of Node's native http module for the browser. It tries to match Node's API and behavior as closely as possible, but some features aren't available, since browsers don't give nearly as much control over requests.

This is heavily inspired by, and intended to replace, http-browserify.

What does it do?

In accordance with its name, stream-http tries to provide data to its caller before the request has completed whenever possible.

Backpressure, allowing the browser to only pull data from the server as fast as it is consumed, is supported in:

  • Chrome >= 58 (using fetch and WritableStream)

The following browsers support true streaming, where only a small amount of the request has to be held in memory at once:

  • Chrome >= 43 (using the fetch API)
  • Firefox >= 9 (using moz-chunked-arraybuffer responseType with xhr)

The following browsers support pseudo-streaming, where the data is available before the request finishes, but the entire response must be held in memory:

  • Chrome
  • Safari >= 5, and maybe older
  • IE >= 10
  • Most other Webkit-based browsers, including the default Android browser

All browsers newer than IE8 support binary responses. All of the above browsers that support true streaming or pseudo-streaming support that for binary data as well except for IE10. Old (presto-based) Opera also does not support binary streaming either.

IE8 note:

As of version 2.0.0, IE8 support requires the user to supply polyfills for Object.keys, Array.prototype.forEach, and Array.prototype.indexOf. Example implementations are provided in ie8-polyfill.js; alternately, you may want to consider using es5-shim. All browsers with full ES5 support shouldn't require any polyfills.

How do you use it?

The intent is to have the same API as the client part of the Node HTTP module. The interfaces are the same wherever practical, although limitations in browsers make an exact clone of the Node API impossible.

This module implements http.request, http.get, and most of http.ClientRequest and http.IncomingMessage in addition to http.METHODS and http.STATUS_CODES. See the Node docs for how these work.

Extra features compared to Node

  • The message.url property provides access to the final URL after all redirects. This is useful since the browser follows all redirects silently, unlike Node. It is available in Chrome 37 and newer, Firefox 32 and newer, and Safari 9 and newer.

  • The options.withCredentials boolean flag, used to indicate if the browser should send cookies or authentication information with a CORS request. Default false.

This module has to make some tradeoffs to support binary data and/or streaming. Generally, the module can make a fairly good decision about which underlying browser features to use, but sometimes it helps to get a little input from the developer.

  • The options.mode field passed into http.request or http.get can take on one of the following values:

    • 'default' (or any falsy value, including undefined): Try to provide partial data before the request completes, but not at the cost of correctness for binary data or correctness of the 'content-type' response header. This mode will also avoid slower code paths whenever possible, which is particularly useful when making large requests in a browser like Safari that has a weaker JavaScript engine.
    • 'allow-wrong-content-type': Provides partial data in more cases than 'default', but at the expense of causing the 'content-type' response header to be incorrectly reported (as 'text/plain; charset=x-user-defined') in some browsers, notably Safari and Chrome 42 and older. Preserves binary data whenever possible. In some cases the implementation may also be a bit slow. This was the default in versions of this module before 1.5.
    • 'prefer-stream': Provide data before the request completes even if binary data (anything that isn't a single-byte ASCII or UTF8 character) will be corrupted. Of course, this option is only safe for text data. May also cause the 'content-type' response header to be incorrectly reported (as 'text/plain; charset=x-user-defined').
    • 'disable-fetch': Force the use of plain XHR regardless of the browser declaring a fetch capability. Preserves the correctness of binary data and the 'content-type' response header.
    • 'prefer-fast': Deprecated; now a synonym for 'default', which has the same performance characteristics as this mode did in versions before 1.5.
  • options.requestTimeout allows setting a timeout in millisecionds for XHR and fetch (if supported by the browser). This is a limit on how long the entire process takes from beginning to end. Note that this is not the same as the node setTimeout functions, which apply to pauses in data transfer over the underlying socket, or the node timeout option, which applies to opening the connection.

Features missing compared to Node

  • http.Agent is only a stub
  • The 'socket', 'connect', 'upgrade', and 'continue' events on http.ClientRequest.
  • Any operations, including request.setTimeout, that operate directly on the underlying socket.
  • Any options that are disallowed for security reasons. This includes setting or getting certain headers.
  • message.httpVersion
  • message.rawHeaders is modified by the browser, and may not quite match what is sent by the server.
  • message.trailers and message.rawTrailers will remain empty.
  • Redirects are followed silently by the browser, so it isn't possible to access the 301/302 redirect pages.
  • The timeout event/option and setTimeout functions, which operate on the underlying socket, are not available. However, see options.requestTimeout above.

Example

http.get('/bundle.js', function (res) {
	var div = document.getElementById('result');
	div.innerHTML += 'GET /beep<br>';

	res.on('data', function (buf) {
		div.innerHTML += buf;
	});

	res.on('end', function () {
		div.innerHTML += '<br>__END__';
	});
})

Running tests

There are two sets of tests: the tests that run in Node (found in test/node) and the tests that run in the browser (found in test/browser). Normally the browser tests run on Sauce Labs.

Running npm test will run both sets of tests, but in order for the Sauce Labs tests to run you will need to sign up for an account (free for open source projects) and put the credentials in a .zuulrc file.

To run just the Node tests, run npm run test-node.

To run the browser tests locally, run npm run test-browser-local and point your browser to http://localhost:8080/__zuul

License

MIT. Copyright (C) John Hiesey and other contributors.