Add device port unmapping support to forwarder2.

Add device port unmapping support to forwarder2.

There used not to be any way to unmap a previously forwarded port. Since
the host and device forwarder processes are kept alive during a whole test
suite, both the host and device ended up with an unnecessarily high amount of
unused open sockets. That is source of flakiness (e.g. too many opened file
descriptors).

This adds an UnmapDevicePort() method to the Python Forwarder wrapper class and
modifies the native forwarder2 to accept a negative port on the command line.
When a negative port is provided, the host forwarder tries to unmap that port
(that was mapped previously).

BUG=229685, 239014
R=bulach@chromium.org, digit@chromium.org

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=200257

[Philippe, 2013-05-14 16:45:38]

FYI, another CL will follow to make the Python Chrome test server spawner use a
single Python Forwarder instance for each shard. This means that we will do a
single adb forward for each shard rather than one for each unit test using the
test server spawner as we do currently.

[bulach, 2013-05-14 18:07:38]

lgtm, but I'm not too familiar here.. go ahead once david is happy :)

[digit1, 2013-05-15 10:13:46]

lgtm, thanks :)

https://codereview.chromium.org/15008004/diff/6001/tools/android/forwarder2/h...
File tools/android/forwarder2/host_forwarder_main.cc (right):

https://codereview.chromium.org/15008004/diff/6001/tools/android/forwarder2/h...
tools/android/forwarder2/host_forwarder_main.cc:152: adb_port,
std::abs(device_port));
I believe -device_port might work as well since you know the value is negative
here.

https://codereview.chromium.org/15008004/diff/14001/tools/android/forwarder2/...
File tools/android/forwarder2/socket.cc (right):

https://codereview.chromium.org/15008004/diff/14001/tools/android/forwarder2/...
tools/android/forwarder2/socket.cc:339: for (int i = 0; i < events_.size(); ++i)
better use size_t.
Also please check with clang-format of git cl format that you can actually omit
the braces here.

https://codereview.chromium.org/15008004/diff/14001/tools/android/forwarder2/...
tools/android/forwarder2/socket.cc:367: int max_event_fd = -1;
suggestion: initialize this to socket_ to avoid the std::max() call below.

[Philippe, 2013-05-15 11:40:50]

Thanks David!

https://codereview.chromium.org/15008004/diff/6001/tools/android/forwarder2/h...
File tools/android/forwarder2/host_forwarder_main.cc (right):

https://codereview.chromium.org/15008004/diff/6001/tools/android/forwarder2/h...
tools/android/forwarder2/host_forwarder_main.cc:152: adb_port,
std::abs(device_port));
On 2013/05/15 10:13:46, digit1 wrote:
> I believe -device_port might work as well since you know the value is negative
> here.

Done.

https://codereview.chromium.org/15008004/diff/14001/tools/android/forwarder2/...
File tools/android/forwarder2/socket.cc (right):

https://codereview.chromium.org/15008004/diff/14001/tools/android/forwarder2/...
tools/android/forwarder2/socket.cc:339: for (int i = 0; i < events_.size(); ++i)
On 2013/05/15 10:13:46, digit1 wrote:
> better use size_t.
> Also please check with clang-format of git cl format that you can actually
omit
> the braces here.

Done.

https://codereview.chromium.org/15008004/diff/14001/tools/android/forwarder2/...
tools/android/forwarder2/socket.cc:367: int max_event_fd = -1;
On 2013/05/15 10:13:46, digit1 wrote:
> suggestion: initialize this to socket_ to avoid the std::max() call below.

Good point.

[Philippe, 2013-05-15 11:42:39]

+Peter in case he would like to use this new feature in Blink.

[Philippe, 2013-05-15 13:55:14]

Committed patchset #4 manually as r200257 (presubmit successful).