Catch invalid path in build method

Ship 0.5.3

Fix attaching with UNIX sockets

Signed-off-by: Ben Firshman <ben@firshman.co.uk>

Signed-off-by: Ben Firshman <ben@firshman.co.uk>

Ship 0.5.2

Stop TLS sockets being closed when attaching

Signed-off-by: Ben Firshman <ben@firshman.co.uk>

When calling attach_socket, the response is garbage collected,
which triggers a close method on the file object, closing the socket.
This stop the response from being garbage collected until we've
finished with the underlying socket.

Fixes #350
Signed-off-by: Ben Firshman <ben@firshman.co.uk>

Ship 0.5.1

Enable the overriding of `assert_hostname` on TLSConfig

Signed-off-by: Ben Firshman <ben@firshman.co.uk>

Signed-off-by: Aanand Prasad <aanand.prasad@gmail.com>

- Check that `urllib_ver` is not None, instead of `urllib3`
- Do version comparison only if `urllib_ver` is *not* "dev"

Fixes #325. Original work by @sidprak.

The server responds with an unuseful message (server error EOF), which
happens because an empty tar file is sent over. So check in the client,
who knows why that's happening and tell the user.

Wrap response socket for platform compatibility.

Don't fallback to HTTP unless insecure_registry is specified when pushing/pulling

add support to add or drop capabilities

Restart Policy for Containers Implemented fixes #302

Closes #282

Volumes parameter on create_container should be validated as string

Stop timeout should be added to the request timeout

Using the max of the stop timeout and request timeout did not entirely
make sure that a stop timeout greater than a request timeout wouldn't
fail prematurely with a HTTPTimeout exception. The correct behavior is
to add the timeouts together, as the stop timeout is understood to be
part of the "request processing time". Any transport-level timeout thus
comes in addition to that.
Signed-off-by: Maxime Petazzoni <max@signalfuse.com>