Originally Posted by
ST Dog
But what gets tested in regression testing? Regression testing isn't usually a full formal qualification test, it's a subset of the most important functions. If you get that list wrong, or don't exercise it fully you can miss things.
A full qualification test on one box/software item could take months. That's why they only do regression tests along with testing the new functions.
Say the issue was an assumption that some parameter wouldn't be exceeded. The regression test suite developed isn't going to exceed that either. They are just going to run the same suite as the last time, and the time before that. Only if new functionality is considered important enough will it be added to the regression test. The code could go through a dozen updates without the regression suite being changed.
If the design documentation is good then the impact of changing items is known through forward and backward traceability. I have worked in systems where a line of code could be traced back the a specific functional requirement (or set of functional requirements). There can also be cascade impacts so many regression tests are required - although this should not happen with good loosely coupled designs. This level of design governance is hard but is essential if you are maintaining safety related systems.
So given a change(s) is(are) required at a certain point the required regression tests are known as the documentation allows that (automated) traceability.