Requests

-------------------

**Improvements**

- It is now possible to pass `no_proxy` as a key to the `proxies`
dictionary to provide handling similar to the `NO_PROXY` environment
variable.
- When users provide invalid paths to certificate bundle files or
directories Requests now raises `IOError`, rather than failing at
the time of the HTTPS request with a fairly inscrutable certificate
validation error.
- The behavior of `SessionRedirectMixin` was slightly altered.
`resolve_redirects` will now detect a redirect by calling
`get_redirect_target(response)` instead of directly querying
`Response.is_redirect` and `Response.headers['location']`. Advanced
users will be able to process malformed redirects more easily.
- Changed the internal calculation of elapsed request time to have
higher resolution on Windows.
- Added `win_inet_pton` as conditional dependency for the `[socks]`
extra on Windows with Python 2.7.
- Changed the proxy bypass implementation on Windows: the proxy bypass
check doesn't use forward and reverse DNS requests anymore
- URLs with schemes that begin with `http` but are not `http` or
`https` no longer have their host parts forced to lowercase.

**Bugfixes**

- Much improved handling of non-ASCII `Location` header values in
redirects. Fewer `UnicodeDecodeErrors` are encountered on Python 2,
and Python 3 now correctly understands that Latin-1 is unlikely to
be the correct encoding.
- If an attempt to `seek` file to find out its length fails, we now
appropriately handle that by aborting our content-length
calculations.
- Restricted `HTTPDigestAuth` to only respond to auth challenges made
on 4XX responses, rather than to all auth challenges.
- Fixed some code that was firing `DeprecationWarning` on Python 3.6.
- The dismayed person emoticon (`/o\\`) no longer has a big head. I'm
sure this is what you were all worrying about most.

**Miscellaneous**

- Updated bundled urllib3 to v1.21.1.
- Updated bundled chardet to v3.0.2.
- Updated bundled idna to v2.5.
- Updated bundled certifi to 2017.4.17.

-------------------

**Features**

- Only load the `idna` library when we've determined we need it. This
will save some memory for users.

**Miscellaneous**

- Updated bundled urllib3 to 1.20.
- Updated bundled idna to 2.2.

-------------------

**Bugfixes**

- Fixed an issue with JSON encoding detection, specifically detecting
big-endian UTF-32 with BOM.

-------------------

**Bugfixes**

- Fixed regression from 2.12.2 where non-string types were rejected in
the basic auth parameters. While support for this behaviour has been
re-added, the behaviour is deprecated and will be removed in the
future.

-------------------

**Bugfixes**

- Fixed regression from v2.12.1 for URLs with schemes that begin with
"http". These URLs have historically been processed as though they
were HTTP-schemed URLs, and so have had parameters added. This was
removed in v2.12.2 in an overzealous attempt to resolve problems
with IDNA-encoding those URLs. This change was reverted: the other
fixes for IDNA-encoding have been judged to be sufficient to return
to the behaviour Requests had before v2.12.0.

-------------------

**Bugfixes**

- Fixed several issues with IDNA-encoding URLs that are technically
invalid but which are widely accepted. Requests will now attempt to
IDNA-encode a URL if it can but, if it fails, and the host contains
only ASCII characters, it will be passed through optimistically.
This will allow users to opt-in to using IDNA2003 themselves if they
want to, and will also allow technically invalid but still common
hostnames.
- Fixed an issue where URLs with leading whitespace would raise
`InvalidSchema` errors.
- Fixed an issue where some URLs without the HTTP or HTTPS schemes
would still have HTTP URL preparation applied to them.
- Fixed an issue where Unicode strings could not be used in basic
auth.
- Fixed an issue encountered by some Requests plugins where
constructing a Response object would cause `Response.content` to
raise an `AttributeError`.