Awesome
stream-http
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
andWritableStream
)
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)
All other supported browsers support pseudo-streaming, where the data is available before the request finishes, but the entire response must be held in memory. This works for both text and binary data.
IE note:
As of version 3.0.0, IE10 and below are no longer supported. IE11 support will remain for now.
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 intohttp.request
orhttp.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-streaming': 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.
- 'default' (or any falsy value, including
-
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 nodesetTimeout
functions, which apply to pauses in data transfer over the underlying socket, or the nodetimeout
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, other than
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
andmessage.rawTrailers
will remain empty.- Redirects are followed silently by the browser, so it isn't possible to access the 301/302 redirect pages.
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 .airtaprc
file.
You will also need to run a Sauce Connect Proxy
with the same credentials.
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
the link shown in your terminal.
License
MIT. Copyright (C) John Hiesey and other contributors.