I want to know your opinions about this. Thanks!

I'd prefer option 3, merge device::TestSerialIoHandler into FakeSerialIoHandler. If more than one client of //services/device wants a fake SerialIoHandler implementation we can move it into //services/device/public/cpp/serial but otherwise it's okay to leave it in serial_apitest.cc.

1,
device::TestSerialIoHandler is a subclass of device::SerialIoHandler, means it has hard dependency on //device/serial, so we can't easily merge it into the FakeSerialIoHandler under //extensions/browser/api/serial/.
2,
device::TestSerialIoHandler is also used by //tools/battor_agent codes for testing.

And I noticed that test_serial_io_handler.* are within the //device/serial:test_support gn target, so I'd like to keep all things as is and just make this target visible to both
  //extensions:chrome_extensions_browsertests (for usage in serial_apitest.cc)
and
  //tools/battor_agent:battor_agent_unittests (for usage in tools/battor_agent/)

Does this make sense?

An eventual goal is to merge device::SerialIoHandler with device::SerialIoHandlerImpl so it makes sense to me to make FakeSerialIoHandler a direct subclass of mojom::SerialIoHandler and remove the dependency on //device/serial in the extensions API test code.

Since //tools/battor_agent still needs a non-Mojo mock serial connection we can move device::TestSerialIoHandler into battor_connection_impl_unittest.cc. Note that I don't believe the battor tests make full use of the functionality of device::TestSerialIoHandler so it may be possible to simplify it.

Sorry for late response.

I'm not very sure about my understanding, so I want to confirm:
Currently FakeSerialIoHandler in serial_apitest.cc is already a direct subclass of mojom::SerialIoHandler, your suggestion is to merge TestSerialIoHandler impl into FakeSerialIoHandler?

Understood that we'd better merge necessary parts of TestSerialIoHandler into battor_connection_impl_unittest.cc.

Yes, merge TestSerialIoHandler impl into FakeSerialIoHandler so that FakeSerialIoHandler doesn't depend on //device/serial.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/659e65f84b56393a147ba98c6e54fa30cafe9c7e

commit 659e65f84b56393a147ba98c6e54fa30cafe9c7e
Author: Han Leon <leon.han@intel.com>
Date: Sat Sep 02 04:59:54 2017

[DeviceService] Eliminate dependency on //device/serial from //extensions

This CL merges necessary parts of device::TestSerialIoHandler
implementation into serial_apitest.cc so that we can eliminate
completely the dependency on //device/serial from //extensions.

BUG= 757169 
TEST=browser_tests SerialApiTest.*
TBR=charliea@chromium.org for trivial changes in tools/battor_agent/

Change-Id: Ie981dd14c9eba40c4a162d0f51e486e8461e1997
Reviewed-on: https://chromium-review.googlesource.com/646978
Commit-Queue: Han Leon <leon.han@intel.com>
Reviewed-by: Reilly Grant <reillyg@chromium.org>
Cr-Commit-Position: refs/heads/master@{#499386}
[modify] https://crrev.com/659e65f84b56393a147ba98c6e54fa30cafe9c7e/chrome/test/BUILD.gn
[modify] https://crrev.com/659e65f84b56393a147ba98c6e54fa30cafe9c7e/device/BUILD.gn
[modify] https://crrev.com/659e65f84b56393a147ba98c6e54fa30cafe9c7e/device/serial/BUILD.gn
[modify] https://crrev.com/659e65f84b56393a147ba98c6e54fa30cafe9c7e/extensions/BUILD.gn
[modify] https://crrev.com/659e65f84b56393a147ba98c6e54fa30cafe9c7e/extensions/browser/api/BUILD.gn
[delete] https://crrev.com/910ab17ccc84f26b89417cd5cf4690a9af327d2b/extensions/browser/api/serial/DEPS
[modify] https://crrev.com/659e65f84b56393a147ba98c6e54fa30cafe9c7e/extensions/browser/api/serial/serial_apitest.cc
[modify] https://crrev.com/659e65f84b56393a147ba98c6e54fa30cafe9c7e/tools/battor_agent/BUILD.gn