Pain of testing legacy batch software

Pain of batch testing legacy software

Problem description

• You have a legacy software that takes a lot of time to test:
  • 2.5 months manually
  • hours automatically
• Testing through UI
• Preparing data for tests takes hours
• Tests are grouped so preparing the environment takes less time. Downside: the tests in a group are depending in each other, hence a failure in the beginning of the group compromises the rest of the test results

Brainstorming

• When writing the tests, focus on the core functionality; then go with tests from the inside to the outside
• the need of tests: "do agile right"
• small steps: refactoring / making tests
• manual smoke tests are not good enough
• Solution to a specific problem: test SQL stored procedures in SQL. A.k.a. keep your tests simple
• Find independent test batches
• Test application changes too: test the code without the migration script; then run the migration script and run the after-migration asserts. This way you can verify the correctness of:
  • the original systems (you might need to roll back to it)
  • the migration script (you'll need to run it on the productin database)
  • the state of the system after the migration

Maintaining automated tests

• Maintaining automated tests is important - as maintaining code is important too. Do it by:
  • keep them maintainable (:
  • make them abstract: use acceptance test frameworks as FitNesse. So you can define the tests in a human-readable format in 1 place and add the glue in another
    • Keyword-driven acceptantce test frameworks:
      • FitNesse http://fitnesse.org/
      • Robot Framework http://code.google.com/p/robotframework/
      • They define high-level keywords
      • You can add deeper description to them

Too many broken legacy tests

• You cannot have too much tests
• In case of broken test hard to tell apart:
  • broken functionality
  • changed functionality
• Canary tests: an old, 100% good branch with its tests. If it fails ==> the environment is failing
• QA tends to write shallow tests: they cover every page but not every case
• Dev tends to write whitebox tests