[Insight-developers] Array Initialization

[Bill Lorensen]

Samson,
These are certainly bugs. The dimension of the scales should be checked 
before they are used. I'm modifying some optimizers and metrics to check 
for the proper size. If the scales are incorrect, then an exception will be 
raised with a descriptive message.

I want to do some further testing, so won't check these in until tomorrow. 
The classes that I have changed so far are:

Algorithms/itkHistogramImageToImageMetric
Numerics/itkGradientDecentOptimizer
Numerics/itkOnePlusOneEvolutionaryOptimizer
Numerics/itkQuaternionRigidTransformGradientDescentOptimier
Numerics/itkRegularStepGradientDescentBaseOptimizer

If you have others that have caused similar problems, let us know,

Bill

At 07:37 PM 12/17/2003, Samson Timoner wrote:

>I've run into an annoyance with array initialization in several classes.
>
>As an example, consider Code/Algorithms/itkHistogramImageToImageMetric.h.
>There is an array, DerivativeStepLengthScales, which should have the number
>of parameters of a transformation as its length.
>
>However, when the class is instatiated, the transform is unknown so that
>the default length of DerivativeStepLengthScales is unknown. Thus, to use
>this class, one MUST set DerivativeStepLengthScales. And, if one does not,
>there eventually is a seg-fault.
>
>Now, if one looks at the class, one might well suspect that this array does
>not need to be initialized. I find this to be true for two reasons:
>
>  1) In the constructor, the array calls the function Fill(1). Thus it
>     appears that the array has been set and need not be initialized. In
>     fact, since the array has the default length 0, this command does 
> nothing.
>
>  2) There is no note in the class that says: "To use this class, you must
>     set DerivativeStepLengthScales".
>
>Several times now, I have found myself chasing down seg-faults which turn
>out to be errors on my part assuming that arrays did not need to be
>initialized, when in fact they did.
>
>Is there something we can do about this?
>
>   Does it makes sense to get rid of initializations of arrays of length 0?
>
>   Adding a simple note to the header of a class that says "You must set the
>   following variables or this class will crash and burn" would be very
>   helpful.
>
>   A different way would be to explicitly check for array lengths to make
>   sure they are of the correct size. Maybe we only do this while debugging
>   is turned on.
>
>In short, I'd like to get rid of the most common way that I cause
>seg-faults when using a new class.
>
>  Any thoughts?
>
>-- Samson
>_______________________________________________
>Insight-developers mailing list
>Insight-developers at itk . org
>http://www . itk . org/mailman/listinfo/insight-developers