JSON *must* be encoded using UTF-8, UTF-16 or UTF-32 (see the [RFC][1]; detect the encoding based on the fact that JSON always starts with 2 ASCII characters.
[1]: http://tools.ietf.org/html/rfc4627#section-3
Prior to this, you could sneak a list of anything to register_hook and it
would accept it. This will check if the items in the list are callable before
registering them. Also added a regression test to make sure if this gets
changed it will be noticed.
If encoding is None, decoding will throw the following TypeError:
TypeError: unicode() argument 2 must be string, not None
If this is the case, attempt to run without any set encoding
Prior to this, you could sneak a list of anything to register_hook and it
would accept it. This will check if the items in the list are callable before
registering them. Also added a regression test to make sure if this gets
changed it will be noticed.
If encoding is None, decoding will throw the following TypeError:
TypeError: unicode() argument 2 must be string, not None
If this is the case, attempt to run without any set encoding
This allows iter_content and iter_lines to succeed without
crashing even after the response content has been fetched
(iter_content gives you an iterator over the prefetched
content)
* Properly handle repeated data fields for multipart/form-data requests (#737)
* Allow a list of 2-tuples as the `files` agument.
* Consistently serialize lists a of parameters (#729).
* prefetch now defaults to True, ensuring that by default, sockets
are returned to the urllib3 connection pool on request end
* sessions now have a close() method, notifying urllib3 to close pooled
connections
* the module-level API, e.g., `requests.get('http://www.google.com')`,
explicitly closes its session when finished
When prefetch is False, the open socket becomes part of the state of the
Response object, and it's the client's responsibility to read the whole
body, at which point the socket will be returned to the pool.
It uses the same logic whenever we're trying to use a list of key values and
avoid what I had started doing -- using the same pattern everywhere.
proxies, and headers should both be covered by this commit.
Ian Cordasco