Adding Tests to Untestable Legacy Code

My work consists of improving the code quality in legacy applications, most in the millions of lines. How to approach testing when the code is untestable? Where to begin with an application that has no tests? What to do with existing tests that are either hard to understand or have fallen into disrepair? I will answer all these questions by showing how these problems were solved in real-world projects.

Transcript

1. Adding Tests to Untestable Legacy Code HEISENBUG MOSCOW | OCT 5, 2021 @afilina
2. "We need tests to refactor, but we need to refactor the code to make it testable." Only a problem if we write the wrong tests at the wrong time of the refactoring process.
3. Anna Filina • Coding since 1997 • VB, PHP, Java, Ruby, C#, etc. • Legacy archaeology • Test automation • Public speaking • Mentorship • Twitter legacy woes • YouTube videos
4. No Tests Hard to justify a testing spree.
5. Always write tests with other work. Don't make it a standalone task.
6. • Bug report. • New feature. • Refactoring. • Build coverage over time.
7. Not Unit-Testable
8. • Bootstraps the framework. • Hits the database. • Call external APIs. • Requires extensive mocking. • Can't mock. The code is not untestable, just not unit-testable.
9. Unit tests end up being convoluted and not useful. Natural response: refactor to make code more testable.
10. Unit tests are sensitive to structure changes. Say you test a class.
11. That class might not exist after refactoring, so you throw the test away. What was the point?
12. • Break up class. • Extract method. • Change method signature. • Reorganize dependencies. Such changes will invalidate unit tests. This defies the purpose of regression tests.
13. You need to keep the same test for the "before" and "after" code, so don't write unit tests yet.
14. Start with acceptance tests. Interact with the whole system using request/response.
15. ASP Classic Tests like that don't even care in which language the application is written.
16. PHP I used this approach to migrate an application from ASP Classic to PHP.
17. Scenario: User can subscribe with a credit card Given I selected a subscription level When I enter valid credit card details Then I should see a payment receipt For acceptance tests, I use any framework that supports Gherkin.
18. • PHP: Behat • C#: SpecFlow • Java: Cucumber There are even frameworks for testing mobile and desktop applications.
19. Slower but fewer You'll still use unit tests for the fine-grained logic and implementation details.
20. Unit Tests
21. Acceptance tests will pass with both the original and refactored code.
22. Any time you refactor, write new unit tests for the new code.
23. class ProductController extends AbstractController { public function search() { //... return $this->render("products/search.html.twig", $products); } } To test this controller, you need to bring in the entire framework.
24. class ProductController { //... public function __construct(Templating $templating) { $this->templating = $templating; } public function search() { //... return $this->templating->render("products/search.html.twig", $products); } } Don't extend. Inject dependencies so you can mock them.
25. • Write high-level tests that survive refactoring. • Write unit tests for new or refactored code. This is the plan for an application that has no tests.
26. Broken Tests Some applications already have tests, but they might be broken or of poor quality.
27. • Didn't compile. • Most failed due to DB changes. • After fixes, many kept failing. • After review, most were wrong or redundant. This was an application when one small feature had 300 acceptance tests, but we couldn't use them.
28. • Checking that something did not happen. • Comparing large DB dump after execution. • Many unnecessary permutations. • Many scenarios still untested. Example: e-mail not sent, but this can happen due to a crash.
29. Afraid to throw code away The alternative can be even more costly.
30. • Fixing estimated at 6 months. • The tests would still be unmaintainable. • Many scenarios would still require new tests.
31. • 46 new tests. • All scenarios covered. • No redundancies. • Easy to read and to maintain.
32. Scenario: TLA1217-A01 Given File is present with records for DE-BER / 2019-01-01 Given Table import_log is empty Given Table country contains: | id | name | area | population | timezone_offset | timezone_offset_dst | created_date | updated_date | | 1 | Germany | 357386 | 83000000 | 1 | 2 | 2019-01-01 | 2019-01-01 | | 2 | Lithuania | 357386 | 83000000 | 1 | 2 | 2019-01-01 | 2019-01-01 | | 3 | Argentina | 357386 | 83000000 | 1 | 2 | 2019-01-01 | 2019-01-01 | Given Table city contains: | id | country_id | area | population | elevation | latitude | longitude | geocode | created_date | updated_date | name | code | | 1 | 1 | 891.7 | 3748148 | 34 | 52.516667 | 13.388889 | DE3 | 2019-01-01 | 2019-01-01 | Berlin | DE-BER | | 2 | 1 | 891.7 | 3748148 | 34 | 52.516667 | 13.388889 | DE3 | 2019-01-01 | 2019-01-01 | Vilnius | LT-VLN | | 3 | 1 | 891.7 | 3748148 | 34 | 52.516667 | 13.388889 | DE3 | 2019-01-01 | 2019-01-01 | Buenos Aires | AR-BAI | Given Table weather_line contains: | id | min_temp | max_temp | rainfall | evaporation | sunshine | wind_gust_dir | wind_gust_speed | wind_dir_9am | wind_dir3_pm | wind_speed9_am | wind_speed3_pm | humidity9_am | humidity3_pm | pressure9_am | pressure3_pm | cloud9_am | cloud3_pm | temp9_am | temp3_pm | rain_today | risk_mm | 1 | 8 | 24.3 | 0 | 3.4 | 6.3 | NW | 30 | SW | NW | 6 | 20 | 68 | 29 | 1019.7 | 1015 | 7 | 7 | 14.4 | 23.6 | No | 3.6 | 2 | 14 | 26.9 | 3.6 | 4.4 | 9.7 | ENE | 39 | E | W | 4 | 17 | 80 | 36 | 1012.4 | 1008.4 | 5 | 3 | 17.5 | 25.7 | Yes | 3.6 | 3 | 13.7 | 23.4 | 3.6 | 5.8 | 3.3 | NW | 85 | N | NNE | 6 | 6 | 82 | 69 | 1009.5 | 1007.2 | 8 | 7 | 15.4 | 20.2 | Yes | 39.8 | 4 | 13.3 | 15.5 | 39.8 | 7.2 | 9.1 | NW | 54 | WNW | W | 30 | 24 | 62 | 56 | 1005.5 | 1007 | 2 | 7 | 13.5 | 14.1 | Yes | 2.8 | 5 | 7.6 | 16.1 | 2.8 | 5.6 | 10.6 | SSE | 50 | SSE | ESE | 20 | 28 | 68 | 49 | 1018.3 | 1018.5 | 7 | 7 | 11.1 | 15.4 | Yes | 0 | 6 | 6.2 | 16.9 | 0 | 5.8 | 8.2 | SE | 44 | SE | E | 20 | 24 | 70 | 57 | 1023.8 | 1021.7 | 7 | 5 | 10.9 | 14.8 | No | 0.2 | 7 | 6.1 | 18.2 | 0.2 | 4.2 | 8.4 | SE | 43 | SE | ESE | 19 | 26 | 63 | 47 | 1024.6 | 1022.2 | 4 | 6 | 12.4 | 17.3 | No | 0 | 8 | 8.3 | 17 | 0 | 5.6 | 4.6 | E | 41 | SE | E | 11 | 24 | 65 | 57 | 1026.2 | 1024.2 | 6 | 7 | 12.1 | 15.5 | No | 0 | 9 | 8.8 | 19.5 | 0 | 4 | 4.1 | S | 48 | E | ENE | 19 | 17 | 70 | 48 | 1026.1 | 1022.7 | 7 | 7 | 14.1 | 18.9 | No | 16.2 | 10 | 8.4 | 22.8 | 16.2 | 5.4 | 7.7 | E | 31 | S | ESE | 7 | 6 | 82 | 32 | 1024.1 | 1020.7 | 7 | 1 | 13.3 | 21.7 | Yes | 0 | 11 | 9.1 | 25.2 | 0 | 4.2 | 11.9 | N | 30 | SE | NW | 6 | 9 | 74 | 34 | 1024.4 | 1021.1 | 1 | 2 | 14.6 | 24 | No | 0.2 | 12 | 8.5 | 27.3 | 0.2 | 7.2 | 12.5 | E | 41 | E | NW | 2 | 15 | 54 | 35 | 1023.8 | 1019.9 | 0 | 3 | 16.8 | 26 | No | 0 | 13 | 10.1 | 27.9 | 0 | 7.2 | 13 | WNW | 30 | S | NW | 6 | 7 | 62 | 29 | 1022 | 1017.1 | 0 | 1 | 17 | 27.1 | No | 0 | 14 | 12.1 | 30.9 | 0 | 6.2 | 12.4 | NW | 44 | WNW | W | 7 | 20 | 67 | 20 | 1017.3 | 1013.1 | 1 | 4 | 19.7 | 30.7 | No | 0 | 15 | 10.1 | 31.2 | 0 | 8.8 | 13.1 | NW | 41 | S | W | 6 | 20 | 45 | 16 | 1018.2 | 1013.7 | 0 | 1 | 18.7 | 30.4 | No | 0 When I execute TLA1217 Then Table import_summary contains: | id | created_date | status | rows_imported | rows_skipped | | 1 | 2019-01-01 | success | 7 | 5 | Then Table weather_line contains: | id | min_temp | max_temp | rainfall | evaporation | sunshine | wind_gust_dir | wind_gust_speed | wind_dir_9am | wind_dir3_pm | wind_speed9_am | wind_speed3_pm | humidity9_am | humidity3_pm | pressure9_am | pressure3_pm | cloud9_am | cloud3_pm | temp9_am | temp3_pm | rain_today | risk_mm | 1 | 8 | 24.3 | 0 | 3.4 | 6.3 | NW | 30 | SW | NW | 6 | 20 | 68 | 29 | 1019.7 | 1015 | 7 | 7 | 14.4 | 23.6 | No | 3.6 | 2 | 14 | 26.9 | 3.6 | 4.4 | 9.7 | ENE | 39 | E | W | 4 | 17 | 80 | 36 | 1012.4 | 1008.4 | 5 | 3 | 17.5 | 25.7 | Yes | 3.6 | 3 | 13.7 | 23.4 | 3.6 | 5.8 | 3.3 | NW | 85 | N | NNE | 6 | 6 | 82 | 69 | 1009.5 | 1007.2 | 8 | 7 | 15.4 | 20.2 | Yes | 39.8 | 4 | 13.3 | 15.5 | 39.8 | 7.2 | 9.1 | NW | 54 | WNW | W | 30 | 24 | 62 | 56 | 1005.5 | 1007 | 2 | 7 | 13.5 | 14.1 | Yes | 2.8 | 5 | 7.6 | 16.1 | 2.8 | 5.6 | 10.6 | SSE | 50 | SSE | ESE | 20 | 28 | 68 | 49 | 1018.3 | 1018.5 | 7 | 7 | 11.1 | 15.4 | Yes | 0 | 6 | 6.2 | 16.9 | 0 | 5.8 | 8.2 | SE | 44 | SE | E | 20 | 24 | 70 | 57 | 1023.8 | 1021.7 | 7 | 5 | 10.9 | 14.8 | No | 0.2 | 7 | 6.1 | 18.2 | 0.2 | 4.2 | 8.4 | SE | 43 | SE | ESE | 19 | 26 | 63 | 47 | 1024.6 | 1022.2 | 4 | 6 | 12.4 | 17.3 | No | 0 | 8 | 8.3 | 17 | 0 | 5.6 | 4.6 | E | 41 | SE | E | 11 | 24 | 65 | 57 | 1026.2 | 1024.2 | 6 | 7 | 12.1 | 15.5 | No | 0 | 9 | 8.8 | 19.5 | 0 | 4 | 4.1 | S | 48 | E | ENE | 19 | 17 | 70 | 48 | 1026.1 | 1022.7 | 7 | 7 | 14.1 | 18.9 | No | 16.2 | 10 | 8.4 | 22.8 | 16.2 | 5.4 | 7.7 | E | 31 | S | ESE | 7 | 6 | 82 | 32 | 1024.1 | 1020.7 | 7 | 1 | 13.3 | 21.7 | Yes | 0 | 11 | 9.1 | 25.2 | 0 | 4.2 | 11.9 | N | 30 | SE | NW | 6 | 9 | 74 | 34 | 1024.4 | 1021.1 | 1 | 2 | 14.6 | 24 | No | 0.2 | 12 | 8.5 | 27.3 | 0.2 | 7.2 | 12.5 | E | 41 | E | NW | 2 | 15 | 54 | 35 | 1023.8 | 1019.9 | 0 | 3 | 16.8 | 26 | No | 0 | 13 | 10.1 | 27.9 | 0 | 7.2 | 13 | WNW | 30 | S | NW | 6 | 7 | 62 | 29 | 1022 | 1017.1 | 0 | 1 | 17 | 27.1 | No | 0 | 14 | 12.1 | 30.9 | 0 | 6.2 | 12.4 | NW | 44 | WNW | W | 7 | 20 | 67 | 20 | 1017.3 | 1013.1 | 1 | 4 | 19.7 | 30.7 | No | 0 | 15 | 10.1 | 31.2 | 0 | 8.8 | 13.1 | NW | 41 | S | W | 6 | 20 | 45 | 16 | 1018.2 | 1013.7 | 0 | 1 | 18.7 | 30.4 | No | 0 | 16 | 12.4 | 32.1 | 0 | 8.4 | 11.1 | E | 46 | SE | WSW | 7 | 9 | 70 | 22 | 1017.9 | 1012.8 | 0 | 3 | 19.1 | 30.7 | No | 0 | 17 | 13.8 | 31.2 | 0 | 7.2 | 8.4 | ESE | 44 | WSW | W | 6 | 19 | 72 | 23 | 1014.4 | 1009.8 | 7 | 6 | 20.2 | 29.8 | No | 1.2 | 18 | 11.7 | 30 | 1.2 | 7.2 | 10.1 | S | 52 | SW | NE | 6 | 11 | 59 | 26 | 1016.4 | 1013 | 1 | 5 | 20.1 | 28.6 | Yes | 0.6 | 19 | 12.4 | 32.3 | 0.6 | 7.4 | 13 | E | 39 | NNE | W | 4 | 17 | 60 | 25 | 1017.1 | 1013.3 | 1 | 3 | 20.2 | 31.2 | No | 0 | 20 | 15.6 | 33.4 | 0 | 8 | 10.4 | NE | 33 | NNW | NNW | 2 | 13 | 61 | 27 | 1018.5 | 1013.7 | 0 | 1 | 22.8 | 32 | No | 0 | 21 | 15.3 | 33.4 | 0 | 8.8 | 9.5 | WNW | 59 | N | NW | 2 | 31 | 60 | 26 | 1012.4 | 1006.5 | 1 | 5 | 22.2 | 32.8 | No | 0.4 | 22 | 16.4 | 19.4 | 0.4 | 9.2 | 0 | E | 26 | ENE | E | 6 | 11 | 88 | 72 | 1010.7 | 1008.9 | 8 | 8 | 16.5 | 18.3 | No | 25.8 Then Table import_log contains: | id | created_date | type | message | | 1 | 2019-01-01 | warning | Invalid city code "FR-PAR" | | 2 | 2019-01-01 | warning | Invalid city code "FR-PAR" | | 3 | 2019-01-01 | warning | Invalid city code "FR-PAR" | | 4 | 2019-01-01 | warning | Invalid city code "FR-PAR" | | 5 | 2019-01-01 | warning | Invalid city code "FR-PAR" | Example of existing tests: basically DB dump in and DB dump out.
33. Scenario: Add warning for invalid city Given city exists for RU-MSK And file contains entry for RU-MSK / 2021-01-01 And file contains entry for CA-MTL / 2021-01-01 When I execute Import Daily Weather Then 1 weather entry should have been imported And warning Invalid city code "CA-MTL" should be logged New tests were readable and maintainable. Only 46 of them, and they covered all use cases.
34. • Will the tests survive refactoring? • Effort of fixing vs rewriting tests. • Do the tests give you confidence? • Can you maintain these tests? Based on those answers, you'll know whether to salvage or rewrite the tests.
35. Thank you! @afilina