# Class 7: Test Driven Development ## Today Today's [assignment](https://classroom.github.com/a/4tXUyq_N) ### Debugging We'll have more opportunity to talk about it more, but in a nutshell,t there is a very large range of possibilities, from the very basic (compiler warnings, adding print statements to the code) to the rather fancy (debuggers, debugger integration in the IDE, parallel debuggers, sanitizers, etc.) For today, though, we'll focus on avoiding bugs in the first place, and get more into testing, which is a very powerful tool to help with that. In particular, we'll talk about test-driven development, which is a software development process that relies on writing tests first, and then writing the code to satisfy those tests. ### Test-driven development People's approach to testing varies wildly. It is not uncommon to not test a resarch code in detail at all, other than running entire simulations and checking for results to look reasonable. This obviously save lots of time writing tests, finding bugs, and fixing them. On the other hand, lots of bugs still happen, and finding them when running the whole application is often very hard and time-intensive, and some bugs may never be found a all. The other extreme is to test everything, ie., achieve "100% coverage" of your test suite, which means that there is a test for every code path. While commercial software aims to go this way, this is often not feasible for science/research project, and probably not worth the effort. Nevertheless, I highly recommend doing lot of testing as you write code, and if those tests can be formalized into a test suite, that helps future development / maintenance a lot. A side effect is that in order to be able to test small units at a time, one is essentially forced to design smaller, independent structures / components / modules, which greatly helps keeping a code manageable. Traditionally, one writes code first, and then tests to make sure that it works afterwards. An interesting alternative approach is test-driven development (TDD), where one writes the tests first, and then the code to satisfy the tests next. In a nutshell, test-driven developments follows these steps: * Add a test * Run all tests and make sure the new test fails * Write the code * Run all tests and make sure they now all pass * Refactor / clean up * Repeat until your project is done ### Bowling game I will demonstrate this process (or at least the beginning of it) on a classical example, ie., let's implement a function that scores a game of bowling. Here are the scoring rules for 10-pin bowling (straight from Wikipedia): > In traditional scoring, one point is scored for each pin that is knocked over, and when less than all ten pins are knocked down in two rolls in a frame (an open frame), the frame is scored with the total number of pins knocked down. However, when all ten pins are knocked down with either the first or second rolls of a frame (a mark), bonus pins are awarded as follows. > * Strike: When all ten pins are knocked down on the first roll (marked "X" on the scoresheet), the frame receives ten pins plus a bonus of pinfall on the next two rolls (not necessarily the next two frames). A strike in the tenth (final) frame receives two extra rolls for bonus pins. > * Spare: When a second roll of a frame is needed to knock down all ten pins (marked "/" on the scoresheet), the frame receives ten pins plus a bonus of pinfall in the next roll (not necessarily the next frame). A spare in the first two rolls in the tenth (final) frame receives a third roll for bonus pins. ### Set up the project I already set things up for you in today's assignment. In particular, I added a `CMakeLists.txt` that builds the `test_bowling` executable, but `test_bowling.cxx` and `bowling.cxx` are still basically empty. `bowling.cxx` will have the `bowlingScore()` function, and, as you might guess, `test_bowling.cxx` will contain the tests for that function. I'm using googletest as a testing framework, and the cmake bits I put in there should make sure that it automatically gets downloaded if it's not already found on your system. ### Write the first test One of the simplest cases is doing all gutter rolls, in which case the final score should be 0. ```cpp TEST(Bowling, AllZeros) { std::vector rolls = { 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, }; EXPECT_EQ(bowlingScore(rolls), 0); } ``` I roll 20 zeros in a row, and I expect to get a zero. In order to store my 20 zeros, I'm using `std::vector`, which is essentially the basic C++ data type for an array of ints. ### Make sure test fails Well, when run this test, one does not get very far, since it already fails at the compilation stage. It fails with an unpleasant error, but the reason is simple: We're calling the `bowlingScore()` function, which we haven't written yet at all. So let's add a simple, and intentionally wrong, first attempt at providing the function: ```cpp int bowlingScore(const std::vector& rolls) { return -1; } ``` Now we can compile the code, and when running it, the test fails, as it is supposed to: ```sh vscode ➜ /workspaces/class-7/build (main) $ ./test_bowling Running main() from /workspaces/class-7/build/_deps/gtest-src/googletest/src/gtest_main.cc [==========] Running 1 test from 1 test suite. [----------] Global test environment set-up. [----------] 1 test from Bowling [ RUN ] Bowling.AllZeros /workspaces/class-7/test_bowling.cxx:10: Failure Expected equality of these values: bowlingScore(rolls) Which is: -1 0 [ FAILED ] Bowling.AllZeros (0 ms) [----------] 1 test from Bowling (0 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test suite ran. (0 ms total) [ PASSED ] 0 tests. [ FAILED ] 1 test, listed below: [ FAILED ] Bowling.AllZeros 1 FAILED TEST ``` This is where googletest shines, providing some helpful information on what's going on. It's running our single test, and it fails. And it also shows the problem: `bowlingScore(rolls)` returns -1, but we expect the final score to be 0. ### Write the code / fix the test There isn't a whole lot needed to fix the test -- just returning 0 will do it: ```diff diff --git a/bowling.cxx b/bowling.cxx index 4660674..bc39996 100644 --- a/bowling.cxx +++ b/bowling.cxx @@ -3,5 +3,5 @@ int bowlingScore(const std::vector& rolls) { - return -1; + return 0; } ``` And in fact the test now passes: ```sh vscode ➜ /workspaces/class-7/build (main) $ ./test_bowling Running main() from /workspaces/class-7/build/_deps/gtest-src/googletest/src/gtest_main.cc [==========] Running 1 test from 1 test suite. [----------] Global test environment set-up. [----------] 1 test from Bowling [ RUN ] Bowling.AllZeros [ OK ] Bowling.AllZeros (0 ms) [----------] 1 test from Bowling (0 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test suite ran. (0 ms total) [ PASSED ] 1 test. ``` ### Refactor / clean up There isn't really anything to clean up yet, so we're going to repeat the process. ### Write another test (that fails) So let's assume I improved a little bit and occasionally manage to get the ball to knock over some pins: ```cpp TEST(Bowling, RegularGame) { std::vector rolls = { 0,0, 2,3, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 3,6, }; EXPECT_EQ(bowlingScore(rolls), 14); } ``` Note that instead of just picking, say rolling 20 times "3", I picked something with a bit more variation, in order to make the test a bit more meaningful. Once I add this test, it fails, as it should: ```sh vscode ➜ /workspaces/class-7/build (main) $ ./test_bowling Running main() from /workspaces/class-7/build/_deps/gtest-src/googletest/src/gtest_main.cc [==========] Running 2 tests from 1 test suite. [----------] Global test environment set-up. [----------] 2 tests from Bowling [ RUN ] Bowling.AllZeros [ OK ] Bowling.AllZeros (0 ms) [ RUN ] Bowling.RegularGame /workspaces/class-7/test_bowling.cxx:17: Failure Expected equality of these values: bowlingScore(rolls) Which is: 0 14 [ FAILED ] Bowling.RegularGame (0 ms) [----------] 2 tests from Bowling (0 ms total) [----------] Global test environment tear-down [==========] 2 tests from 1 test suite ran. (0 ms total) [ PASSED ] 1 test. [ FAILED ] 1 test, listed below: [ FAILED ] Bowling.RegularGame 1 FAILED TEST ``` ### Fix the second test The test expects that score should be 14 (which is correct), but our `bowlingScore()` function apparently returns 0, and that's not too surprising, considering it doesn't look at the array of rolls that it gets passed at all. A good start to actually scoring a bowling game would be to sum up all the pins that were knocked over in each roll: ```diff diff --git a/bowling.cxx b/bowling.cxx index c5cbdbb..bf0f841 100644 --- a/bowling.cxx +++ b/bowling.cxx @@ -12,7 +12,11 @@ int bowlingScore(const std::vector& rolls) { - return 0; + int score = 0; + for (int roll : rolls) { + score += roll; + } + return score; } // ====================================================================== ``` This does in fact work: ```sh vscode ➜ /workspaces/class-7/build (main) $ ./test_bowling Running main() from /workspaces/class-7/build/_deps/gtest-src/googletest/src/gtest_main.cc [==========] Running 2 tests from 1 test suite. [----------] Global test environment set-up. [----------] 2 tests from Bowling [ RUN ] Bowling.AllZeros [ OK ] Bowling.AllZeros (0 ms) [ RUN ] Bowling.RegularGame [ OK ] Bowling.RegularGame (0 ms) [----------] 2 tests from Bowling (0 ms total) [----------] Global test environment tear-down [==========] 2 tests from 1 test suite ran. (0 ms total) [ PASSED ] 2 tests. ``` ### Refactor / clean up There still isn't really any need to refactor at this point, it's about as concise as it can be. So I'm doing this instead: The above function uses a somewhat fancy "range-based for loop", which is a newer C++ feature. It's nice and compact, and I would think pretty intuitive to understand, but it'll be hard using that approach to look at results frame-by-frame, as we will probably have to do once it comes to handling spares and strikes correctly. So let's refactor the code into a more traditional for loop for now: ```diff diff --git a/bowling.cxx b/bowling.cxx index bf0f841..0916195 100644 --- a/bowling.cxx +++ b/bowling.cxx @@ -13,8 +13,9 @@ int bowlingScore(const std::vector& rolls) { int score = 0; - for (int roll : rolls) { - score += roll; + int n_rolls = rolls.size(); + for (int i = 0; i < n_rolls; i++) { + score += rolls[i]; } return score; } ``` ### Repeat And the story continues, but now it's your turn: #### Your turn Get started on doing your own test-driven development. Starting from the current state of your assignment repo, perform the steps above to get started, and then keep following the process of test-driven development to finish the `bowlingScore` function to handle all the possible scenarios that can happen in a real game. In order for me to follow your steps, please commit at every step, including commit where you add the test and make sure it fails, then commit how you fix it, then commit again after refactoring (if there is anything to refactor). **Note**: In a real bowling game, not every frame does have two rolls and there may be extra rolls at the end, too. Rolls that don't happen **are not** part of the input. E.g., the first two frames of a game may look like `{ 10, 3, 4 }`, ie., a strike, and then 3 and 4 pins falling in the next frame. ### Homework * Finish the bowling project. It may not hurt to throw some arbitrary tests from the internet on it. E.g., here's one: . Of course, you'll need to translate it into input for your code, and remember that only actually rolls should be in the input. So the score card from this link should start like this: `{ 10, 9,1, 5,5, 7,2, 10, 10, ...}`.