RE: Proposal for device "enumeration"

> From: Martin Thomson [mailto:martin.thomson@gmail.com]
> 
> On 7 December 2012 10:29, Travis Leithead <travis.leithead@microsoft.com>
> wrote:
> > I say less because my exposure doesn't tie-in to any device-specific
> data to create the identifier. If the one-way hash you propose is ever
> compromised, then it might be possible to start to correlate specific
> devices. But other than that, yes, they are essentially the same.
> 
> That was an example method of what an implementation could do, not the
> canonical method.  You could equally persist mappings on a per-origin
> basis and create far less exposure.
> 
> > I understand this. I'm just not convinced that devices are going to
> change that often.
> 
> You obviously use communication devices very differently to me.  My
> devices change constantly.  I have devices at the office and home.  My
> Skype installation knows of at least 5 microphones and 3 cameras and
> the subset of those that are connected change often.  The builtin mic
> and camera are awful, but they serve when I'm travelling; my apple
> monitor at my desk has both, and these are marginally better; the
> headset is good for one-to-one calls; and I frequently borrow a
> colleague's speaker device for group calls.  I no longer worry about
> things because the application has remembered my priorities and it
> chooses the right behavior based on what is currently present.  Having
> to fight with settings every time I launch the app would be a
> deal-breaker.

Great context, thank you!

It sounds like at least Harld was bought into this approach, so I'll see about committing this into my draft.

One question: since you have to presumably first perform some setup (involving requesting, gathering source capabilities, etc.) before remembering a given deviceId, it may be OK to _only_ disclose deviceIds once you have a source device, and not before (say using the static method). If this is the case, then the current v5 proposal of returning a number of devices still makes sense from a discoverability standpoint. Does that make sense?

Received on Friday, 7 December 2012 18:54:12 UTC