CMake: Add separate unit test targets

[ferdnyc]

Each unit test file ClassName.cpp, which results in the creation of a test target openshot-ClassName-test, will now also be labeled with the CTest label ClassName, and a target ClassName_coverage will be generated that runs only the tests in that file.

The names

Just like the main coverage target, the target name ClassName_coverage is just a generic name. Running the target will collect coverage data if ENABLE_COVERAGE is set, but will run the tests without coverage if it is left disabled. It does not imply collection of coverage data.

(That all started because I couldn't redefine the test target to also collect coverage, so I had to use a different name. And since it's easier to always use the same target name, I made coverage work to run tests with or without ENABLE_COVERAGE. I could have called the individual targets ClassName_test, but I decided to be consistent with coverage.)

Y THO?

The separate targets are especially useful when developing tests for a class, as the tests in the class being worked on can be re-run without having to wait a minute or more while the other tests run.

Example workflow

# Configure project
cmake -B build -S .
# Work on tests for MyClass
vim tests/MyClass.cpp
# Build and run only the tests for MyClass
cmake --build build --target MyClass_coverage

(The first time through, libopenshot itself will still have to be built before the tests can run. But in subsequent executions of cmake --build build --target MyClass_coverage, the build will be quick because the only file that will be recompiled is tests/MyClass.cpp. When running the tests, the very lengthy runtimes of certain tests in FrameMapper, FFmpegWriter, and CVTracker are avoided.)

[ferdnyc]

When running the tests, the very lengthy runtimes of certain tests in FrameMapper, FFmpegWriter, and CVTracker are avoided.

In fact, the labeling has an unexpected side benefit: This summary, which CTest displays at the end of the run if the full suite of tests are enabled:

100% tests passed, 0 tests failed out of 135

Label Time Summary:
CVStabilizer     =   3.56 sec*proc (2 tests)
CVTracker        =  62.05 sec*proc (2 tests)
CacheDisk        =   6.82 sec*proc (6 tests)
CacheMemory      =  10.67 sec*proc (11 tests)
Clip             =   7.18 sec*proc (6 tests)
Color            =   8.81 sec*proc (9 tests)
Coordinate       =   5.10 sec*proc (5 tests)
DummyReader      =   7.08 sec*proc (7 tests)
FFmpegReader     =  14.32 sec*proc (8 tests)
FFmpegWriter     =  10.08 sec*proc (2 tests)
Fraction         =   4.33 sec*proc (5 tests)
Frame            =   5.55 sec*proc (5 tests)
FrameMapper      =  38.04 sec*proc (13 tests)
ImageWriter      =   5.00 sec*proc (1 test)
KeyFrame         =  23.89 sec*proc (24 tests)
Point            =  13.41 sec*proc (11 tests)
QtImageReader    =   3.54 sec*proc (3 tests)
ReaderBase       =   1.14 sec*proc (1 test)
Settings         =   1.96 sec*proc (2 tests)
Timeline         =  14.52 sec*proc (12 tests)

Total Test time (real) =  63.19 sec

So we can now see at a glance that CVTracker is by far the most time-consuming test (as one would expect, really, given the complexity of its function), but FrameMapper is shockingly the second-place occupant on that list, which feels worth investigating.

The long runtime may simply be a result of test thoroughness and/or complexity. (That's certainly why KeyFrame takes so long. It runs 24 separate test cases, almost twice as many as FrameMapper.) But long runtimes may also may be a sign of inefficiencies — either in the tests, or the class itself.

[codecov]

Codecov Report

Merging #660 (317c0be) into develop (4e4a95c) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff            @@
##           develop     #660   +/-   ##
========================================
  Coverage    52.42%   52.42%           
========================================
  Files          151      151           
  Lines        12346    12346           
========================================
  Hits          6473     6473           
  Misses        5873     5873           

Impacted Files	Coverage Δ
tests/KeyFrame.cpp	100.00% <100.00%> (ø)
tests/Point.cpp	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4e4a95c...317c0be. Read the comment docs.

[ferdnyc]

Merging this, as it's relatively inconsequential unless used.

(Except for a switch I snuck in, to use the standard CTest variable BUILD_TESTING as the preferred way to enable/disable unit tests. It's on by default, but can be switched off to disable all unit test / coverage targets. The old DISABLE_TESTS and ENABLE_TESTS variables will be forwarded to BUILD_TESTING if they're set.)