Provide a way to disable Nagle's algorithm on Pepper TCP sockets.

ppapi/proxy/ppapi_messages.h

Index: ppapi/proxy/ppapi_messages.h
diff --git a/ppapi/proxy/ppapi_messages.h b/ppapi/proxy/ppapi_messages.h
index 6175d02a72b47ebb698c51ebd40ae1209a85b177..9fab30485e1490a4d3e64693b24d76923eeea89b 100644
--- a/ppapi/proxy/ppapi_messages.h
+++ b/ppapi/proxy/ppapi_messages.h
@@ -671,6 +671,10 @@ IPC_MESSAGE_ROUTED4(PpapiMsg_PPBTCPSocket_WriteACK,
                     uint32 /* socket_id */,
                     bool /* succeeded */,
                     int32_t /* bytes_written */)
+IPC_MESSAGE_ROUTED3(PpapiMsg_PPBTCPSocket_SetBoolOptionACK,
+                    uint32 /* plugin_dispatcher_id */,
+                    uint32 /* socket_id */,
+                    bool /* succeeded */)
 
 // PPB_URLLoader_Trusted
 IPC_MESSAGE_ROUTED1(
@@ -1213,6 +1217,10 @@ IPC_MESSAGE_CONTROL2(PpapiHostMsg_PPBTCPSocket_Write,
                      std::string /* data */)
 IPC_MESSAGE_CONTROL1(PpapiHostMsg_PPBTCPSocket_Disconnect,
                      uint32 /* socket_id */)
+IPC_MESSAGE_CONTROL3(PpapiHostMsg_PPBTCPSocket_SetBoolOption,
+                     uint32_t /* socket_id */,

[yzshen1, 2013/02/12 17:39:58]

Please be consistent with other messages about the type of |socket_id|.

[Wez, 2013/02/12 18:45:21]

Done.

+                     uint32_t /* name */,

[yzshen1, 2013/02/12 17:39:58]

Please use the enum type.

[Wez, 2013/02/12 18:45:21]

This is an IPC message definition. Using the enum type here would imply that
recipients can rely on the field having one of the defined enum values, which is
not the case.

[yzshen1, 2013/02/13 08:07:14]

I don't understand, why the recipients shouldn't rely on the field having one of
the defined enum values?

[Wez, 2013/02/13 22:06:22]

Because over IPC an enum is transferred as an int. If the peer is buggy or
compromised then the int's value may fall outside the enum's range. Since the
IPC layer simply static_cast<>s the int back to the enum, the resulting value is
undefined for values outside the enum's range.

[yzshen1, 2013/02/14 08:35:38]

Your concern makes sense. I am okay to do it your way (for now).

However, we need to find a better way to fix the general problem. Using integer
as a replacement of enum doesn't seem to be a real fix:
- It hurts readability.
- It is still cast back to the enum type (in pepper_tcp_socket.cc). And I think
you rely on the 'default' branch to catch out-of-range values. There are two
problems:
    (1) the validation is not in the IPC layer, instead it happens long after we
receive the IPC message. Imagine that we have a lot of messages passing enum
types as integers. The validation will be scattered in different parts of our
codebase. And it relies on every user of those messages to do the validation,
which is error-prone.
    (2) Since the result of the cast is undefined, theoretically it can be any
value. Therefore, it may run any branch of the switch statement. That
uncertainty bothers me as well. :)

 
On 2013/02/13 22:06:22, Wez wrote:

[Wez, 2013/02/14 08:57:53]

Agreed; I've filed crbug.com/176110 for fixing the IPC issue.

[Wez, 2013/02/14 08:57:53]

D'oh!  Yes, I should remove the cast.

 There are two
Yup.  See above.

I'll update the CL to remove the cast - if you're otherwise happy with it then
please wield your awesome OWNER powers. ;)

+                     bool /* value */)
 
 // UDPSocketPrivate.
 IPC_MESSAGE_CONTROL0(PpapiHostMsg_UDPSocketPrivate_Create)