[kepler-dev] [Bug 5546] SDF director iterations parameter default of 0 is unfriendly

Tue Aug 14 11:43:17 PDT 2012

http://bugzilla.ecoinformatics.org/show_bug.cgi?id=5546

Christopher Brooks <cxh at eecs.berkeley.edu> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
       Hours Worked|                            |1.0
    Remaining Hours|2.0                         |1.0

--- Comment #8 from Christopher Brooks <cxh at eecs.berkeley.edu> 2012-08-14 11:43:17 PDT ---
I'm  appending the thread in pt-dev about this.
I made the fix (call isEmbedded()).

Derik asked about creating a filter that would set iterations to 0.
We don't have such a filter, but I believe that the implementation below
will do the trick.  For preexisting models that do not have iterations set,
then the default will be "AUTO", which will provide the correct functionality.

Also, I created ptolemy/domains/sdf/kernel/fixSDFIterations, which is a script
that forces iterations to be set to 0.  I'm not sure if it is needed.

I'll open a documentation bug to update the docs for this change.

This bug is almost ready to be closed.

On 8/10/12 7:00 PM, Edward A. Lee wrote:
>
> One minor point: The director should call isEmbedded()
> instead of seeing whether the container's container is null.
> The reason for this is RunCompositeActor, where the container's
> container is not null, but you still want the model to behave
> as if it were at the top level...
>
> I admit that this design is not ideal. Any ideas for a better
> way?
>
> Edward
>
>
> On 8/10/12 4:33 PM, Christopher Brooks wrote:
>> Ok, I implemented your suggested changes.
>>
>> The SDF Constructor eventually calls:
>>> // AUTO and UNBOUNDED are used to set the value of iterations,
>>> // see the getIterations() method.
>>>
>>> Parameter AUTO = new Parameter(this, "AUTO");
>>> AUTO.setToken(_auto);
>>> AUTO.setVisibility(Settable.EXPERT);
>>>
>>> Parameter UNBOUNDED = new Parameter(this, "UNBOUNDED");
>>> UNBOUNDED.setToken(IntToken.ZERO);
>>> UNBOUNDED.setVisibility(Settable.EXPERT);
>>>
>>> iterations = new Parameter(this, "iterations");
>>> iterations.setTypeEquals(BaseType.INT);
>>> iterations.addChoice("AUTO");
>>> iterations.addChoice("UNBOUNDED");
>>> iterations.setExpression("AUTO"); 
>>
>>
>>
>> I added the following method that determines how iterations are handled:
>>> /** Return the number of iterations.
>>> *
>>> * <p>The number of iterations returned depends on the value of
>>> * the <i>iterations</i> parameter and whether the container
>>> * of the director is at the top level.
>>> * See the {@link #interations} documentation for details.</p>
>>> *
>>> * <p>Code that uses SDFDirector should call getIterations()
>>> * instead of directly referring to the value of the
>>> * <i>iterations</i> parameter.
>>> *
>>> * @return the number of iterations
>>> * @exception If thrown while getting the value of the
>>> * iterations parameter.
>>> */
>>> public int getIterations() throws IllegalActionException {
>>> // See "SDF director iterations parameter default of 0 is unfriendly"
>>> // http://bugzilla.ecoinformatics.org/show_bug.cgi?id=5546
>>> int iterationsValue = ((IntToken) (iterations.getToken())).intValue();
>>> if (iterationsValue > 0) {
>>> return iterationsValue;
>>> }
>>> NamedObj container = getContainer();
>>> if (container != null
>>> && container.getContainer() == null) {
>>> // The container of this director is at the toplevel
>>> if (iterations.equals(_auto)) {
>>> return 1;
>>> }
>>> }
>>> return 0;
>>> } 
>>
>> It will take a few days for any bugs to settle out of the nightly build.
>>
>> _Christopher
>>
>> On 8/9/12 2:12 PM, Edward A. Lee wrote:
>>>
>>> How about if we define a local variable called AUTO
>>> (that has value -1), and document that AUTO means what
>>> you say below. This seems like a good solution...
>>> We should also define a variable called UNBOUNDED
>>> that has value 0, with the current meaning of 0.
>>> Then we should enhance the parameter with
>>> addChoice("AUTO") and addChoise("UNBOUNDED").
>>>
>>> Edward
>>>
>>>
>>> On 8/9/12 9:06 AM, Christopher Brooks wrote:
>>>> In the perfect world, if the SDFDirector was at the top level,
>>>> the initial number of iterations would be 1.
>>>>
>>>> If the SDFDirector was not at the top level,
>>>> the initial number of iterations would be 0.
>>>>
>>>> Perhaps we should set the initial value to -1, which would have
>>>> the above behavior?
>>>>
>>>> I dunno.
>>>>
>>>> Right now, the initial default for iterations is 1.
>>>>
>>>> The test failures are not that bad, so let's leave it be for awhile.
>>>>
>>>> The bug is at
>>>> http://bugzilla.ecoinformatics.org/show_bug.cgi?id=5546
>>>>
>>>> _Christopher
>>>>
>>>>
>>>> On 8/9/12 8:39 AM, Edward A. Lee wrote:
>>>>>
>>>>> Oops... I forgot about the non-top-level problem.
>>>>> Setting it to 1 by default might create major problems...
>>>>> People will be very puzzled ...
>>>>> Maybe we need to be smarter about this...
>>>>>
>>>>> Edward
>>>>>
>>>>>
>>>>>
>>>>> On 8/9/12 8:34 AM, Christopher Brooks wrote:
>>>>>> Jennifer,
>>>>>> Also, the SDFDirector iterations parameter now defaults to 1.
>>>>>> Previously, it defaulted to 0, which meant "run forever".
>>>>>>
>>>>>> Edward, my understanding is that iterations should usually be set to 0
>>>>>> when SDFDirector is not at the top level. Is this the case?
>>>>>> Should we have some text about this?

-- 
Configure bugmail: http://bugzilla.ecoinformatics.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA Contact for the bug.