Fixed not to append the model id when fake video device is enabled.

media/capture/video/video_capture_device.cc

Index: media/capture/video/video_capture_device.cc
diff --git a/media/capture/video/video_capture_device.cc b/media/capture/video/video_capture_device.cc
index 4b771e564ee9c05c0641bd14b5174cd4d005f6b6..49ca78a05165a43e2998cf4d8eab7cc368f25df4 100644
--- a/media/capture/video/video_capture_device.cc
+++ b/media/capture/video/video_capture_device.cc
@@ -4,8 +4,10 @@
 
 #include "media/capture/video/video_capture_device.h"
 
+#include "base/command_line.h"
 #include "base/i18n/timezone.h"
 #include "base/strings/string_util.h"
+#include "media/base/media_switches.h"
 
 namespace media {
 
@@ -14,7 +16,10 @@ const std::string VideoCaptureDevice::Name::GetNameAndModel() const {
   if (model_id.empty())
     return device_name_;
   const std::string suffix = " (" + model_id + ")";
-  if (base::EndsWith(device_name_, suffix, base::CompareCase::SENSITIVE))
+  if (base::EndsWith(device_name_, suffix, base::CompareCase::SENSITIVE) ||
+      base::CommandLine::ForCurrentProcess()->HasSwitch(
+          switches::kUseFakeDeviceForMediaStream))
+    // Ignore |model_id| if |kUseFakeDeviceForMediaStream| flag is present.

[mcasas, 2015/09/04 02:24:50]

Hmm it's not that simple :)

Command line flags, specially long standing ones like this,
should be contained in its spread, otherwise we create two
what essentially amounts to two code paths.

The problem here is that GetModel() is linked on compile
time to whatever platform we're compiling, whereas we
should use more a run time call. And that will prove hard
since we'd like to collect the Model, to paste to the Name,
/before/ actually creating the VideoCaptureDevice proper.

[msu.koo, 2015/09/05 07:58:11]

Thank you for your comment :)
A little bit hard to understand what "should be contained in its spread" means.
Can you explain a little bit more?

On 2015/09/04 02:24:50, mcasas wrote:

[mcasas, 2015/09/08 17:50:15]

Yeah, so command line flags are supposed to be 
limited in its scope and to be used at
minimal points and for limited functionality. 
If a command line flag is used in several 
points, it'd break this implicit limitation, 
and it could potentially be used for several 
things. A flag should be used only once, in
one place, and, if possible, with a view on
removing it altogether long term. 
Hmmm... makes sense?

[msu.koo, 2015/09/11 07:08:50]

I fully agree with you that should minimize using command-line branch
statements.
So not to use the command-line checking in here, I think it's required to change
the root approach(select platform at build time) because fake device is selected
at runtime.
I'm not sure that it's a better way to change the root approach for fixing this
bug instead of using command-line branching.
If you are not hardly opposite on this way, I would take the current way(using
commandline switch). 
How do you think?

On 2015/09/08 17:50:15, mcasas wrote:

[mcasas, 2015/09/15 17:11:49]

Hmm is true that changing the whole infra might prove daunting
at this point. Proposal: leave and land this CL as is after adding 
and creating another bug, adding a line here:

// TODO(esnada): http://crbug.com/ABCDEF, remove checking the switch in favour
// of deferring GetModel() call to the actual VideoCaptureDevice object.

Or any other wording of you choice :)
SGTY?

     return device_name_;
   return device_name_ + suffix;
 }