When using the fetch api, exceptions thrown inside
ClientRequest._onFinish, which also calls the user callback,
were being caught and turned into 'error' events. This is
not the correct behavior; these exceptions should remain
uncaught to let the browser print them to the console.
From now on, IE8 is only supported if polyfills for `Object.keys`,
`Array.prototype.forEach`, and `Array.prototype.indexOf` are
provided. The suggested polyfills from MDN are provided in
ie8-polyfill.js, or you can use https://github.com/es-shims/es5-shim
for a more complete set of polyfills.
If multiple chunks of data got queued up, the abort on data test
would erroneously fail. Wait till the end of the tick before
considering 'data' events as errors.
Make options handling logic match node's `http` module more
closely and let the browser handle all of the relative url
cases itself.
This has a few important advantages:
* Behavior is closer to native `http` modula
* Relative paths in urls are handled properly (fixes#16)
* Code is simpler and more obviously correct
Previously, testing for 'ms-stream' and 'moz-chunked-arraybuffer'
support on the xhr object caused warnings to appear in Chrome.
Since fetch will always be used instead of these when available,
short-circuit and return false when fetch is available.
The default behavior now preserves the 'content-type' header, at
the expense of not providing pseudo-streming in Safari, generic
WebKit, and older Chrome. Applications that require streaming
and already use 'prefer-stream' mode will not be affected.
The 'prefer-fast' mode is now deprecated, as that behavior now
matches the default. The 'allow-wrong-content-type' mode is new;
it gives the same behavior as the old default in case binary
streaming is essential.
Fixes#8
In order to have parity with XMLHttpRequest fetch credentials should use same-origin instead of omit
Setting XMLHttpRequest.withCredentials to false has no effect on same-site request
John Hiesey
John Hiesey
Matt Crute
Jan Bölsche
Tejesh Mehta
Cesar Andreu