We are developing a WPF GUI. We use TDD, but not everything is coverable, so manual testing plays an important role.
We have a set of about 400 manual test cases. Before a release our tester runs the full refression. If something is found and developers fix it the tester has to repeat his work. Now with agile it is even worse, since we have very short release cycles.
So I think that this can be done better. To reduce the regression I want to be able to detect which test scenarios get touched by the particular code change.
GUI is a set of screens. Testers work with them. So we can link test cases to the screens.
Each screen has an object model. It is possible to build up a graph of dependencies of the particular screen. These dependencies are classes and interfaces, that actually live in files (source or binaries). Using that graph and SVN diff it should be possible to detect which screens get affected. From screens we get to test-cases.
Has anyone used such approach? Are there any tools for that?
In theory, it may be possible in your case, but that depends heavily on the details. There may be run time dependencies or hidden dependencies which are not reflected by the compile time dependencies detectable by a tool like NDepend. For example, a specific code change may allow to enter data with Form1 into your database which could not be entered before, and another Form2 using this data will get affected though Form2 is not changed (perhaps it should have been changed and the issue is that one forgot this).
