[makecheck]

I do agree with the point that it is possible to spend a lot of time fixing "broken" tests. The lesson that should often be learned is that certain tests are not adding enough value to warrant the constant maintenance. (Code alone requires plenty of maintenance as it is.)

One thing that works extremely well is to have a test environment. In other words, spend the time to replicate your entire production flow 100% (but in isolation), so that you can "deploy" changes and see if anything blows up. This not only catches stupid deployment bugs, but it also allows you to avoid some unit testing by simply relying on customers themselves: have them log into the isolated environment and do exactly what they would have done in production, and you'll know if it works.