Converted from SourceForge issue 3418691, submitted by ceball
Submit Date: 2011-10-04 22:27 GMT
Although we have a test of instantiating as many classes as possible, only those components that can be instantiated in isolation are tested (and it's only tested that they can be instantiated - nothing more).
We could do better by using a simulation, doing things like using all subclasses of a particular type in the appropriate position (e.g. every subclass of TransferFn could used as an output function). This way, components such as projections that cannot be instantiated in isolation can be tested at least a little.
Another cheap way to test would be to have a test set up for each family of component, run it on every subclass, save the output, and check that output never changes. For instance, every subclass of TransferFn could be presented with a sine grating.
Converted from SourceForge issue 3418691, submitted by ceball Submit Date: 2011-10-04 22:27 GMT
Although we have a test of instantiating as many classes as possible, only those components that can be instantiated in isolation are tested (and it's only tested that they can be instantiated - nothing more).
We could do better by using a simulation, doing things like using all subclasses of a particular type in the appropriate position (e.g. every subclass of TransferFn could used as an output function). This way, components such as projections that cannot be instantiated in isolation can be tested at least a little.
Another cheap way to test would be to have a test set up for each family of component, run it on every subclass, save the output, and check that output never changes. For instance, every subclass of TransferFn could be presented with a sine grating.