Yes, there is a better way to do it. I found this particular problem at the Programming Praxis web site, which took it from http://butunclebob.com/ArticleS.UncleBob.TheBowlingGameKata
This is a superb example of how an informal specification and a set of test cases can diverge. The specification says to score a bowling game according to a particular set of rules. Here is exactly the algorithm called for, written in the functional programming language Haskell: score (10:b:c:xs) = b+10+c+(if null xs then 0 else score (b:c:xs)) score (a:b:c:xs) | a+b = 10 = 10+c+(if null xs then 0 else score (c:xs)) score (a:b:xs) = a+b+(if null xs then 0 else score xs) That's *it*. Explanation, (10:b:c:xs) means "a sequence starting with 10, followed by something (call it b), followed by something (call it c), followed by the rest of the sequence (call that xs). This is actually strictly statically typed as well. And it shows the while the data are *provided* from first to last, the score is most easily *computed* from last to first. Let's remind ourselves of Kent Beck's four rules of good code: 1. Runs all the tests 2. Contains no duplication 3. Expresses all the important design intentions 4. Minimizes entities (e.g. classes and methods). Hence the Extreme Programming slogan "You Ain't Gonna Need It." Now we come to the divergence between the specification (which calls for #roll: and #score methods) and the tests (which call for a #scoreAfterRolling: method and then in test 15 calls for #scoreRolling:after:). There are two kind of test case for this exercise. + check that the result is correct in a good case - check that the right error is raised given bad data. Now the second kind go far beyond the informal specification and require not only that certain errors be detected but *when* they are detected. And this is something you have to bear in mind when writing down your test cases: you have to prevent ideas you may have about *how* the code might work creeping into your test cases and forcing the code to work that way. Most of my code, and most of the trouble I had writing the code, was about these "does it report the right error at the right time and oh yeah I made up these 'right' definitions with no warrant in the customer's requirements" test cases. I mean, seriously, NO methods in common between the specifications and tests? Returning to Kent Beck's maxims. How many classes do we need? The test code calls for ONE class, "Bowling". There are no tests for any Frame pr Roll classes, and no need for any. How many methods does it need? The test code checks #scoreAfterRolling and #scoreRolling:after: There are not tests for any others. What state should the instances of the class hold? There is hint You will need to store some game state. but that applies to the #roll:/#score interface, NOT to the interface that's actually tested. Use as many classes and methods as you NEED, but remember that you need a lot fewer than you think, On Mon, 21 Dec 2020 at 23:05, r.wobben--- via Pharo-users < pharo-users@lists.pharo.org> wrote: > Hello, > > > To come back at this one. > > I made a class Bowling and a class Frame which contains two instance > variables named firstThrow and secondThrow. > > > but I still wonder how do I know which one to fill in. > > I can do ` (firstThrow nil?) ` but that feels not right. > > > Is there a more idiomatic solution for this ? > > > Roelof > > >
