Mojo: Change the way message pipes are passed over channels.

Mojo: Change the way message pipes are passed over channels.

Before:
* We'd send the message "containing" the message pipe (handle) first.
* Upon receiving it, the other side would create an endpoint (and
  dispatcher, etc.)
* The other side then sent a message back indicating the new ID (and
  telling the first side to run the message pipe.
* Upon receiving that "run" message, the first side would start sending
  messages for that message pipe.

This makes for rather complicated handshaking, which would only get more
complicated if, e.g., the second side wanted to forward the message pipe
handle that it received to someone else (e.g., over another channel).

Now:
* We "remotely allocate" an ID for the other side and send a message
  telling it to create an endpoint with that ID.
* We forward all the messages queued to that message pipe endpoint
  immediately (since we already know the destination ID).
* We send the message "containing" the message pipe (handle).
* The other side then creates a dispatcher and associates it with the
  (already-existing) endpoint.

Still to do:
* We need to track IDs that we've allocated for the remote side more
  carefully (in case of wrap-around, to avoid collisions).
* Convert everything over to the "attach and run" mode of operation (and
  possibly rename "attach and run" to something shorter), and delete the
  various separate "attach" and "run" methods.

R=brettw@chromium.org

Committed: https://crrev.com/df8e15ae21198cb6ac1587a07a179fa26655048c
Cr-Commit-Position: refs/heads/master@{#300157}

[brettw, 2014-10-17 20:11:13]

LGTM. How does the forwarding case work in the new design? It seems like the
concept of local and remote ID will get more complicated.

[viettrungluu, 2014-10-17 20:26:33]

The CQ bit was checked by viettrungluu@chromium.org

[commit-bot: I haz the power, 2014-10-17 20:27:34]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/664763002/1

[viettrungluu, 2014-10-17 20:30:50]

On 2014/10/17 20:11:13, brettw wrote:
> LGTM. How does the forwarding case work in the new design? It seems like the
> concept of local and remote ID will get more complicated.

The main complication (which existed even under the old scheme) will be in
making sure that the first side and the third side have a direct channel, and
then allocating a message pipe on that channel (assigning IDs).

But at least assigning IDs can be done by one of those sides (the first or the
third), and you don't have to worry about waiting for startup handshake
completion (or doing it simultaneous with the handshake).

[commit-bot: I haz the power, 2014-10-17 20:41:08]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2014-10-17 20:43:26]

Patchset 1 (id:??) landed as
https://crrev.com/df8e15ae21198cb6ac1587a07a179fa26655048c
Cr-Commit-Position: refs/heads/master@{#300157}