Has anyone had a chance to review the revised algorithm yet?

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

Has anyone had a chance to review the revised algorithm yet?

Jim Barnett
We need to publish a new Last Call draft once we're satisfied that we have things right.  The sooner we do that, the better, but it's most important to get things right.  
Jim
Reply | Threaded
Open this post in threaded view
|

Re: Has anyone had a chance to review the revised algorithm yet?

Gavin Kistner-4
On Apr 12, 2014, at 6:27 AM, Jim Barnett <[hidden email]> wrote:
> We need to publish a new Last Call draft once we're satisfied that we have things right.  The sooner we do that, the better, but it's most important to get things right.  

I have not yet tried to re-implement the revised algorithm in my interpreter, but will attempt to do so this weekend.

I would like to reiterate what my coworker Chris Nuernberger suggested in February 2013: I think that it would have saved us a lot of trouble if there were a reference implementation in the spec.

Right now we have an official set of unit tests and an official algorithm for interpretation. We do not know for sure if these match up, because there is no way to run the algorithm against them. We have to instead port the pseudo-code algorithm to real languages, and hope that we did not make a mistake and interpret the pseudo-code in a manner other than intended. Only then can we see if this matches the unit tests. And if it doesn’t, which then is right? Is the implementation an invalid representation of the algorithm? Does the algorithm need to be changed to match the unit tests? Is there a bug in the unit tests?

If, instead, there was an official reference implementation, you could independently ensure that the algorithm and unit tests agreed. At that point, you have a definitive “right” answer, guaranteed.
Reply | Threaded
Open this post in threaded view
|

Re: Has anyone had a chance to review the revised algorithm yet?

Zjnue Brzavi
In reply to this post by Jim Barnett
We need to publish a new Last Call draft once we're satisfied that we have things right.  The sooner we do that, the better, but it's most important to get things right.

Hi,

After the recent changes a number of tests fail.
I've updated some instructions for running the tests in a dedicated branch here : https://github.com/zjnue/hscxml/tree/provisional_20140409/test
Happy to assist during a debugging session.

Thanks,
Zjnue
Reply | Threaded
Open this post in threaded view
|

Re: Has anyone had a chance to review the revised algorithm yet?

Jim Barnett
OK, there are clearly problems.  I'll take a look when I'm back in the office. Thanks for running the tests. 

Jim Barnett
Genesys

On Apr 13, 2014, at 11:54 AM, Zjnue Brzavi <[hidden email]> wrote:

We need to publish a new Last Call draft once we're satisfied that we have things right.  The sooner we do that, the better, but it's most important to get things right.

Hi,

After the recent changes a number of tests fail.
I've updated some instructions for running the tests in a dedicated branch here : https://github.com/zjnue/hscxml/tree/provisional_20140409/test
Happy to assist during a debugging session.

Thanks,
Zjnue
Reply | Threaded
Open this post in threaded view
|

RE: Has anyone had a chance to review the revised algorithm yet?

Jim Barnett-2
In reply to this post by Gavin Kistner-4
Gavin,
  It certainly would be nice if there were a reference implementation, but someone has to volunteer to create one.  I think that the ideal solution would  be to have all the necessary helper routines defied in a publicly available place, and then convert the algorithm into actual code in some reasonably well-known language, so that users could combine the algorithm with the helper routines and execute the IR test suite.  

- Jim

-----Original Message-----
From: Gavin Kistner [mailto:[hidden email]]
Sent: Saturday, April 12, 2014 3:59 PM
To: Jim Barnett
Cc: Voice List
Subject: Re: Has anyone had a chance to review the revised algorithm yet?

On Apr 12, 2014, at 6:27 AM, Jim Barnett <[hidden email]> wrote:
> We need to publish a new Last Call draft once we're satisfied that we have things right.  The sooner we do that, the better, but it's most important to get things right.  

I have not yet tried to re-implement the revised algorithm in my interpreter, but will attempt to do so this weekend.

I would like to reiterate what my coworker Chris Nuernberger suggested in February 2013: I think that it would have saved us a lot of trouble if there were a reference implementation in the spec.

Right now we have an official set of unit tests and an official algorithm for interpretation. We do not know for sure if these match up, because there is no way to run the algorithm against them. We have to instead port the pseudo-code algorithm to real languages, and hope that we did not make a mistake and interpret the pseudo-code in a manner other than intended. Only then can we see if this matches the unit tests. And if it doesn't, which then is right? Is the implementation an invalid representation of the algorithm? Does the algorithm need to be changed to match the unit tests? Is there a bug in the unit tests?

If, instead, there was an official reference implementation, you could independently ensure that the algorithm and unit tests agreed. At that point, you have a definitive "right" answer, guaranteed.