Land Recent QUIC Changes.

QUIC client StartConnect and CryptoConnect return void instead of bool.
No behavior change.

CryptoConnect always returns true in QuicCryptoClientStream, and
StartConnect just passes this up the stack. No point checking an always
true value.

Merge internal change: 83014290
https://codereview.chromium.org/811973006/

After internal change: cl/82509991, setting of a sensible client flow
control window is done in QuicClient::Initialize() and so no longer
needs to be set explicitly in each client.

Only touching QUIC clients: example client

Merge internal change: 82959151
https://codereview.chromium.org/830573003/

Minor change to implementation of QUIC Pacing Sender.
No behavior change, not flag protected.

Move the setting of burst_tokens_ from TimeUntilSend to OnPacketSent, so
it no longer needs to be mutable.

Merge internal change: 82655556
https://codereview.chromium.org/833563002/

Delete inaccurate comment from QuicSpdyClientStream

Merge internal change: 82545048
https://codereview.chromium.org/826273002/

QUIC DCHECK becomes LOG_IF(DFATAL) for better error signals.

NextPendingRetransmissions will dereference an invalid pointer if it is
called with an empty list.  Right now, this is enforced via a DCHECK,
which amounts to just documentation in opt mode.

Rather than just dying with an invalid pointer reference, I've changed
this to a LOG_IF(DFATAL,..), so that we have a slightly better signal
about where the problem occurred when it occurred, not later in a bad
dereference.

Merge internal change: 82541587
https://codereview.chromium.org/825733003/

Adds varz for tracking QUIC connections on time-wait list.

Third try on this seemingly simple CL.

Unfortunately, the plan to use callbacks from varz into the time-wait
list manager is a no-go.  The reasons for NOT using callbacks are:

- Messy memory semantics.  We needed a permanent callback, registered
  with the global stats object, for stats collection.  At the same
  time, we had to guarantee that the global stats object never called
  the callback after the time-wait list manager was deleted.  The
  memory-management aspect of the callbacks was very brittle.

- Need for new locks.  We'd need to add a locks to prevent
  thread-contention between the stats-collection object and the
  time-wait list manager.  It's much cleaner to leave all the locking
  inside the ExportedVariableMap, than introduce a new lock inside the
  time-wait list manager or the stats collection object

Reverted back my original design.

There is now a varz that is mapped per-silo, per-port, as Alyssa
suggested.  It tracks the length of the time-wait list per dispatcher.

  Added QUIC varz tracking per-dispatcher time-wait list length.

Merge internal change: 82527019
https://codereview.chromium.org/803723003/

R=rch@chromium.org

Review URL: https://codereview.chromium.org/832713003

Cr-Commit-Position: refs/heads/master@{#310000}

1 files changed, 1 insertions, 1 deletions

diff --git a/net/tools/quic/quic_client.h b/net/tools/quic/quic_client.h
index c095a4d..47d4cec 100644
--- a/net/tools/quic/quic_client.h
+++ b/net/tools/quic/quic_client.h
@@ -75,7 +75,7 @@ class QuicClient : public EpollCallbackInterface,
   // Start the crypto handshake.  This can be done in place of the synchronous
   // Connect(), but callers are responsible for making sure the crypto handshake
   // completes.
-  bool StartConnect();
+  void StartConnect();
 
   // Returns true if the crypto handshake has yet to establish encryption.
   // Returns false if encryption is active (even if the server hasn't confirmed