Thanks for taking your time to review this feature and share your thoughts.
On Thu, May 26, 2011 at 1:37 PM, Mitch Capper
<mitch.capper@gmail.com> wrote:
Hi Moy,
First of all I wanted to thank you and Comrex for the new improvements
and code you have done I think a lot of people will find these
features very useful!
You're certainly welcomed, it is nice to work with companies like Comrex that realize is better to contribute and be part of the community.
Thanks for clarifying some things on IRC. The changes look good my
main concerns with the current code are how well it reacts to dynamic
changes (usb devices plugged or unplugged) and the fact all streams
and endpoints have to be specified in the xml config itself.
Where else should be specified if not in the XML config? doing it dynamically would have created extra complexity and I did not feel added much value for the current user.
I am not sure if there is a large reason to have endpoints that are
defined separately from streams. An endpoint is just a name, a the
indev, the outdev, and the channel for each. Streams already define
an indev and outdev which makes things a bit confusing. If channel
selection was moved to the stream definition it could all be defined
there. The only downside to this that I see is when you have an
audio device with multiple channels you are going to end up
duplicating the stream section partially for each but I am not sure if
that is enough to warrant to require the double configuration of an
audio device before you can start using it. I would say that if the
device selection was expanded so that rather than just a device ID
/name to also include the channel I think it would be an improved
interface.
Well, you have also the sampling rate and IO sizes (codec-ms) which belong to the stream. The main reason for the separation is that they are meant to be used separate. You can define 10 different endpoints using the same stream. As long as they don't conflict in the channel they use and the direction. You can define an endpoint that only uses the input of a given stream and channel, and does not use the output, and another endpoint doing the opposite, both at the same time having a call using the same stream.
Short of that I think before it is merged into trunk the changes
should be merged in a bit better with the existing commands and the
existing stream code should be merged with the shared stream code as
right now some functionality is being duplicated. While the way the
changes were made looks like relatively few new code paths are
followed if you don't use the new shared streams/endpoints, it makes
for a confusing interface for someone trying to figure out how things
work.
The decision here had to do mostly with backwards compatibility and the amount of things I'd most likely break if I touched the original code paths and the amount of testing required to make sure the previous functionality wasn't affected. I'd rather slowly start moving chunks from the old code compatible with the new one as the need arises.
My suggested list of changes would be:
-Merge the existing stream code with the new stream code, this should
eliminate some of the redundancies and allow the new code to be used
everywhere, streams would be expanded to have the naming, channels,
rates, and other features that shared streams allow
I don't think I'd be able to do that at this point.
-Ability to create streams through the API (with the merge of the
above code its just expanding the API to take the additional params)
-Remove endpoints and allow bridging directly to a stream by using the
stream's name
For the reasons I explained before (the same stream being used by multiple endpoints), I can't see how this would work.
-Do not validate streams until you go to use them, right now if a
device is not ready when the xml config is read you can't use it
I like failing early, is easier to spot mistakes in the configuration. I understand this might be a problem if you're planing to plug/unplug USB devices, but that is not our case yet :-)
-pa answer, pa call, pa play are expanded to optionally take the
streamname to act on
-pa switchstream takes the call to switch the stream on
-pa switch this is a tougher one but would need to figure something else out
pa answer, pa call are replaced by uuid_outgoing_answer, originate from the command line respectively.
I think it will probably be a decent bit of work to do these things
but should allow for the great new features you have added without
causing confusion for other users.
The current state of things is exactly to avoid confusing existing users. They can go ahead and keep being happy with their current monodevice configuration and usage. If, and only if they want to move to multi devices they will have to learn something new. I found too complicated expanding the current API's to support the new architecture without changing the usage of the commands.
By all means I am not asking you
do this work any time someone contributes code back to FS its great
and something much appreciated and certainly others can work to fully
integrate it into freeswitch. I figured I would just give some other
feedback about it as you asked for comments.
Your input is very much appreciated. Although I can certainly tell you, given that my own needs and Comrex needs are completely fulfilled with the current feature set, I am looking mostly for bugs or small improvements. Most the things you described make the interface nicer and consistent, but will require a reasonable amount of work and a high chance of breaking existing users (either monodevice users or Comrex themselves).
Once the code is merged is easier to start improving one thing at the time.
-
Moy