Improved Testability Leads to Better Design

From WikiContent

(Difference between revisions)
Jump to: navigation, search
Current revision (14:40, 6 August 2009) (edit) (undo)
 
(6 intermediate revisions not shown.)
Line 1: Line 1:
-
''Economic Testability'' is a simple concept but not one often seen in practice. Essentially it boils down to recognizing that since tested code should be a requirement, and that we have in place some nice, economical, standard tools for testing (such as xUnit, Fit, etc.), then the products that we build should always satisfy the requirement of ease-of-test in addition to any other requirements. Happily this requirement reinforces rather than contradicts best practice. Perhaps a few examples will help?
+
''Economic Testability'' is a simple concept yet one seen infrequently in practice. Essentially, it boils down to recognizing that since code-testing should be a requirement and that we have in place some nice, economical, standard tools for testing (such as xUnit, Fit, etc.), then the products that we build should always satisfy the new requirement of ease-of-test, in addition to any other requirements. Happily the ease-of-test requirement reinforces rather than contradicts best practice.
-
It is often the case that tests are impeded because certain major functions are not yet implemented. For example,an object needs to be tested but it makes use of a yet-to-be-built object is a pretty common occurrence. A way around this is to allow the provision of the yet-to-be-built object via a parameterized constructor. Then, in the test-case situation we can provide a dummy object without changing the internals at all; in the production model we can provide the real (tested) object to deliver the required functionality. When implemented via the constructor, as recommended, in the set-up region of the tests, the actual method signatures are not changed at all.
+
If you build your systems so that testing is made economic — while simultaneously of course preserving simplicity in the production model — the interfaces that you finally end up with are likely to be greatly improved over the one-environment system. Code which has to operate successfully and unchanged in two or more environments (the test and production environments) must pay more than lip service to clean interfaces and maximum encapsulation if the task of embedding within the multiple environments is not to become overwhelming. The discipline needed leads to better design and more modular construction. It really is a win–win situation.
 +
 
 +
Progressive testing is often stymied because certain necessary functions have not yet been implemented. For example, a common occurrence involves an object that needs to be tested, but makes use of a yet-to-be-built object. An incompleteness that means the test cannot be run. A way around this is to allow the provision of the yet-to-be-built object via a parametrized constructor: When testing we can provide a test double object without changing the internals at all; in the production code we can provide the real (tested) object to deliver the required functionality. The API of the test double and the production object are identical and the object under test is unaware of whether it is running in a test or a production environment.
 +
 
 +
We can go further of course — a lot further. This very soft style of building systems brings advantages all down the line when compared to the hard-wired logic commonly encountered in code. For example, if we need to introduce some kind of logging trail for complex bug diagnosis during development, we can provide the logging logic via the constructor parameters, using the plug-in technique already outlined. In production, using the ''null object'' pattern (which will provide a "do nothing" object with the same API as the logger), we can replace the logger with a harmless null alternative. This means of course that the code being monitored is unchanged, an important point for some types of bug. Should it be necessary that you dynamically enable logging in production mode, this technique makes it very simple to enable or disable logging dynamically — or indeed anything else that you need. Then we have a production system that gracefully slips into logging mode when needed. It may be unfortunate that you should need to do this, but if you do, then sensible application of these techniques can be seriously reputation enhancing!
-
We can go further of course .. a lot further. If we wanted to introduce some kind of logging-trail for complex bug diagnosis, then we can provide the logging logic via the technique outlined above, by providing the logging object in the constructor parameters. At production time, using the null-object pattern, we can replace the logger with a harmless alternative. On the other hand, why not make this dynamically selectable? Then we could have a production system which can gracefully slip into logging mode. It may be unfortunate that you should need to do this, but if you do, then you can look like a hero, i.e., a professional!
 
-
Finally, if you build your systems so that testing is made simple, either by following the suggestion here, or by just making the interfaces very simple to use in the xUnit world, and simultaneously of course preserving simplicity in the production model, then our experience has been that the interfaces that you finally end up with are superior in terms of encapsulation and portability than if you had built you system with just one target environment in mind. In other words, the discipline in building an interface which is equally at home in two environments causes you to do a better job in the design-essentials than otherwise. It really is a win-win situation.
 
By [[George Brooke]]
By [[George Brooke]]

Current revision

Economic Testability is a simple concept yet one seen infrequently in practice. Essentially, it boils down to recognizing that since code-testing should be a requirement and that we have in place some nice, economical, standard tools for testing (such as xUnit, Fit, etc.), then the products that we build should always satisfy the new requirement of ease-of-test, in addition to any other requirements. Happily the ease-of-test requirement reinforces rather than contradicts best practice.

If you build your systems so that testing is made economic — while simultaneously of course preserving simplicity in the production model — the interfaces that you finally end up with are likely to be greatly improved over the one-environment system. Code which has to operate successfully and unchanged in two or more environments (the test and production environments) must pay more than lip service to clean interfaces and maximum encapsulation if the task of embedding within the multiple environments is not to become overwhelming. The discipline needed leads to better design and more modular construction. It really is a win–win situation.

Progressive testing is often stymied because certain necessary functions have not yet been implemented. For example, a common occurrence involves an object that needs to be tested, but makes use of a yet-to-be-built object. An incompleteness that means the test cannot be run. A way around this is to allow the provision of the yet-to-be-built object via a parametrized constructor: When testing we can provide a test double object without changing the internals at all; in the production code we can provide the real (tested) object to deliver the required functionality. The API of the test double and the production object are identical and the object under test is unaware of whether it is running in a test or a production environment.

We can go further of course — a lot further. This very soft style of building systems brings advantages all down the line when compared to the hard-wired logic commonly encountered in code. For example, if we need to introduce some kind of logging trail for complex bug diagnosis during development, we can provide the logging logic via the constructor parameters, using the plug-in technique already outlined. In production, using the null object pattern (which will provide a "do nothing" object with the same API as the logger), we can replace the logger with a harmless null alternative. This means of course that the code being monitored is unchanged, an important point for some types of bug. Should it be necessary that you dynamically enable logging in production mode, this technique makes it very simple to enable or disable logging dynamically — or indeed anything else that you need. Then we have a production system that gracefully slips into logging mode when needed. It may be unfortunate that you should need to do this, but if you do, then sensible application of these techniques can be seriously reputation enhancing!


By George Brooke

This work is licensed under a Creative Commons Attribution 3

Back to 97 Things Every Programmer Should Know home page

Personal tools