If a 100 Continue interim response was received after the continue
timeout was reached, dispatch_continue() would force a double submission
of DATA frames. This patch prevents dispatch_continue() from doing
anything if the timer callback has already been invoked. This makes
ContinueTimer a single-shot mechanism, as originally intended.
To deal with servers that don't conform to RFC 7231 (or, potentially,
connections with a large round-trip time), don't wait forever for a 100
Continue status to come back. Currently, the timeout is hard-coded to
one second.
A ContinueTimer encapsulates the handshake timeout logic for a single
request. Somewhat counterintuitively, ContinueTimers are owned by the
HttpClient instead of the Request object, because their lifetime must be
bound to the life of the connection (which is owned by the HttpClient
and not the Requests). A Request is associated with its corresponding
ContinueTimer through a std::weak_ptr.
In this commit, we made --dep-idle behaviour by default. This is
because the previous default behaviour is not reflect current usage of
dependency priority and never will be because of fragility of tree due
to stream closure.
Tatsuhiro Tsujikawa
Jacob Champion