2

In following the principle of testing only the exported functions on a package (using Go - or for others languages, the public functions on a class), I'm running into a scenario where related packages are causing the code base to be tested multiple times.

For example, I have a Preprocessor, which calls a Parser package, which calls a Scanner package which calls a Lexer package.

Preprocessor -> Parser -> Scanner -> Lexer

The results bubble up to the previous package and the final results are compiled in the Preprocessor. No package below the Preprocessor is called by any other part of the app.

Each package has tests, which means the lexer is tested 4x, the scanner 3x, the parser 2x and the preprocessor 1x.

I believe each feature should be its own separate package because if I combined them all into a single Preprocessor package, the parser, scanner and lexer don't need a public API, yet the functionality is too complex not to test individually as it would be difficult to track down where bugs were occurring.

Is this type of duplicate testing recommended or is it a sign that the packages are too related to be separated? Or, is it a sign that the lower level packages are doing too much and I should extract more business logic into the Preprocessor package such that it could call each of the lower level packages, instead of having their requests chained together?

  • Are you having problems with test execution time? – Robert Harvey Sep 10 '15 at 0:21
  • Not at the moment, but more curious as the app grows and what others are doing in these scenarios. – Andy Brewer Sep 10 '15 at 2:54
  • 1
    if test coverage is adequate, and testing time is not a problem, then I don't see what the issue is. – Robert Harvey Sep 10 '15 at 5:49
3

A technique to avoid this - if it is in fact a problem for your project - is to use mocks. The Scanner, for instance, would be tested with a mock Lexer; the Parser with a mock Scanner, etc. In this way your actual Lexer is only exercised by the Lexer tests.

  • True, but the lexer is called from within the scanner, rather than the scanner passed as an argument to the lexer. – Andy Brewer Sep 10 '15 at 4:00
  • 2
    @AndyBrewer: The point that Carl is making is that you can mock the lexer to test the scanner (or vice versa), or use test doubles, to isolate the individual components. If such isolation is not possible, then it's likely that your overall design could be improved to make it more testable. – Robert Harvey Sep 10 '15 at 5:49
  • 1
    @AndyBrewer: the scanner should not call the lexer directly, it should call the lexer through an interface, and before running the scanner, you pass the lexer object (or a mock lexer in test) as an argument to the scanner. This is called dependency injection. – Doc Brown Sep 10 '15 at 6:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.