Motivation:
- state clearly what is covered with tests without reading non-trivial code
- avoid cleverness - tests should be no-brainer
- get closer to one test case targets one feature/beaviour with one check
- replace obscure `throw std::exception` with diagnostics that are actually
useful to pin-point failure cause and location - makes CI logs useful.
- allow to select and run specific tests
- make tests maintenance easy, quick and fun
Propose new structure of tests that reflects the previous tests hierarchy,
but organizes channel tests in test/channel directory with test programs
each covering specific library feature (or set of closely related features).
The refactored tests cover 100% of checks from the old `channel.cpp`,
plus it refines or adds a bunch more.
NOTE: old test/channel.cpp has not been removed yet.
Common definitions from the single test/channel.cpp moved to
channel_test_fixtures.hpp and namespace boost::gil::test::fixture:
Classes and typedefs:
- `do_test` as `fixture::channel`
- `value_core` as `fixture::channel_value`
- `reference_core` as `fixture::channel_reference`
- `packed_reference_core` as `fixture::packed_channel_reference`
- `packed_dynamic_reference_core` as `fixture::packed_dynamic_channel_reference`
- `channel_archetype` and relatives to `channel_concepts.cpp` which is compile
test in Jamfile
- `test_packed_channel_reference()` parts as `fixture::packed_channels565`
- `test_packed_dynamic_channel_reference()` parts as `fixture::packed_dynamic_channels565`
Test case functions called from `do_test<T>::test_all`:
- `test_channel_invert()` to `algorithm_channel_invert.cpp` suite
- `test_channel_convert()` to `algorithm_channel_convert.cpp` suite
- `test_channel_multiply()` to `algorithm_channel_multiply.cpp` suite
- `test_channel_math()` split to `algorithm_channel_relation.cpp`
and `algorithm_channel_arithmetic.cpp`
Add test cases for each channel value type T as used to run from
`test_channel_value_impl<T>`, `test_packed_channel_reference<T>` and
`test_packed_dynamic_channel_reference<T>`.
Add list of possible T-s defined as type-lists `fixture::channel_byte_types`,
`fixture::channel_integer_types`, `channel_float_types` and
`channel_bitfield_types` which used with `BOOST_AUTO_TEST_CASE_TEMPLATE`
generate all possible combination of inputs.
Add new `channel_test_fixture.cpp` is a self-test suite verifying the fixtures.
Import the selection of integer types into boost::gil namespace,
and move from channel.hpp to typedefs.hpp for easier access.
Replace bits32f with float32_t and bits64f with float64_t
- kept as alias of scoped_channel_value.
Move float64_t (bits64f) to typedefs.hpp.
Replace the four {float|double}_{zero|one} min/max channel values
with float_point_zero and float_point_one templates.
Replace <boost/cstdint,hpp> with C++11 <cstdint>.
Introduce preference of using declaration instead of typedef.
Reformat typedefs.hpp to take advantage of the using declaration
- works much better for left-to-right reading, alias name as
most important detail comes first.
Add some of missing #include typedefs.hpp, sort some headers.
Add cast to explicitly indicate val promotion to at least
32-bit integer. The arithmetic operations are performed on
int of higher rank type. Finally, since the result value is
guaranteed to fit range of 8/16-bit result integer type,
it is safe to cast.
This should make it clearer for a reader that subtleties
of implicit promotions and conversions have been considered.
- unused parameter
- shadowed typedef
- implicit conversion changes signedness:
int to signed <int>
int to unsigned <int>
Note, arithmetic op is performed on int as higher rank type
with result is in range of return type, safe to cast.
This enables test runners to copy the file to the target. It also makes
the code that looks for the file in a different path unnecessary since
Boost.Build will pass the proper relative path.
