Add MachMessageServer and its test

Add MachMessageServer and its test.

MachMessageServer is much like mach_msg_server() and
mach_msg_server_once(), but with a C++ interface and with a number of
deficiencies corrected.

TEST=util_test MachMessageServer.*
R=rsesek@chromium.org

Committed: https://chromium.googlesource.com/crashpad/crashpad/+/67082c93b922eda317b325c1fc4b8bfe09103dd2

[Mark Mentovai, 2014-09-05 23:12:52]

mark@chromium.org changed reviewers:
+ rsesek@chromium.org

[Robert Sesek, 2014-09-08 16:28:47]

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
File util/mach/mach_message_server.cc (right):

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
util/mach/mach_message_server.cc:228: : MACH_SEND_TIMEOUT);
Why is SEND_ONCE a gate for SEND_TIMEOUT?

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_server.h
File util/mach/mach_message_server.h (right):

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
util/mach/mach_message_server.h:73: //! \brief Runs a Mach message server to
handle a Mach RPC request.
I'd call out explicitly that this is used for MIG servers, since in the
implementation you freely cast the response to mig_reply_error_t.

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
util/mach/mach_message_server.h:79: //! The size
Incomplete comment.

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
util/mach/mach_message_server.h:96: //! \param[in] persistent When `true`, this
function operates in a loop rather
I'm not sure how clear these two bool parameters are going to be at the
callsites.

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
File util/mach/mach_message_server_test.cc (right):

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
util/mach/mach_message_server_test.cc:36: enum TestType {
No tests for the error reply scenario? What about when mach_msg() fails because
of a dead remote port, for example?

https://codereview.chromium.org/544393002/diff/1/util/mach/mach_message_serve...
util/mach/mach_message_server_test.cc:223: virtual void MachMultiprocessChild()
override {
The test is pretty comprehensive, which is good, but I think the way the logic
is organized may be difficult to maintain. Could this be a table-type test,
where there's test options for {timeout, number to send, blocking, child expect,
parent expect}, and then rather than handling the conditions by test_type_ name,
you could examine the table test options. It would allow you to mix-and-match
the flavors more easily, too.

[Mark Mentovai, 2014-09-08 22:29:02]

All feedback addressed.

[Robert Sesek, 2014-09-09 00:21:58]

LGTM. Awesome test :).

https://codereview.chromium.org/544393002/diff/20001/util/mach/mach_message_s...
File util/mach/mach_message_server.h (right):

https://codereview.chromium.org/544393002/diff/20001/util/mach/mach_message_s...
util/mach/mach_message_server.h:70: //!     be received as the \a in parameter
of Run().
Run() or MachMessageServerFunction()? Same on line 74.

[Mark Mentovai, 2014-09-09 01:06:30]

Thanks, Robert!

[Mark Mentovai, 2014-09-09 01:06:37]

Committed patchset #3 (id:40001) manually as 67082c9 (presubmit successful).