- so far, `gf2()` runs in a `strict` mode by default
=> `gf2(42)` results in a `ValueError`
- we adapt `gf2()` (a.k.a. `lalib.elements.galois.GF2Element`)
such that it behaves more like the built-in `bool()`
=> `gf2(42)` returns `one`, like `bool(42)` returns `True`
- further, the `GF2Element` class is adjusted such that
`one` and `zero` assume their counterparts to be
`1`-like or `0`-like objects during binary operations;
other values result in an error
=> for example, `one + 1` works but `one + 42` does not
=> so, when used in binary operations, `one` and `zero`
continue to cast their counterparts in `strict` mode
- simplify the casting logic
- make `value` a positional-only argument in `gf2()`
(like most of the built-in constructors)
- provide more usage examples in the docstrings
clarifying when `strict` mode is used and when not
- propagate the above changes to the test suite:
+ adapt test cases with regard to the `strict` mode logic
+ add new test cases to `assert` how `gf2()`
can process `str`ings directly
+ rename two test cases involving `complex` numbers
to mimic the naming from the newly added test cases
- extend `pytest` with an option to run only the minimum number
of (unit) test cases to just keep the coverage at 100%
- rationale:
+ many of the unit test cases partly overlap with
respect to the lines of source code executed
+ also, integration tests, by definition, do not
contribute to a higher test coverage
- implementation: mark "redundant" test cases as one of:
+ `pytest.mark.integration_test`
=> code usage from the perspective of the end user
+ `pytest.mark.overlapping_test`
=> tests not contributing to the 100% coverage
+ `pytest.mark.sanity_test`
=> tests providing confidence in the test data
- add `tests.conftest` module
=> programatically convert the above markers into
`@pytest.mark.no_cover` and collect the non-"redundant" tests
- add nox session "test-fast" to run only the minimum
number of (unit) test while holding coverage at 100%
- refactor some test modules
+ wrap some test cases in a class
+ move sanity tests to the end of the files
- the `gf2` sub-classes `GF2One` and `GF2Zero` have no purpose
other than to provide unique `help(one)` and `help(zero)` messages
- delete them at the bottom of `lalib.elements.galois`
to prevent them from being imported at all
(`type(one)` and `type(zero)` still appear to be the `gf2` class,
which is a little white lie, provided by some meta class)
- differentiate between official and inofficial API in the test suite
- `to_gf2()` is only to be used within `GF2Element.__new__()`
=> rename it into `_to_gf2()`
- the test cases in `TestGF2ConstructorWithCastedValue`
cover everything in `TestGF2Casting`
=> retire the redundant test cases
and make the test suite run faster
- the future (concrete) Galois `Field` implementation
shall receive the name `GF2` (as per common math notation)
=> name conflict with the current `GF2` class
implementing the elements of the future Galois `Field`
=> rename the current `GF2` class into `GF2Element`
- because `GF2Element` is a bit tedius to type for the end user,
we introduce a `gf2` alias in line with the naming convention
for the built-in data types (e.g., `int` or `float`)
that are also used as elements of (other) `Field`s
=> name conflict with the current `lalib.elements.gf2` module
=> rename the module into `lalib.elements.galois`
- adjust the docstrings to refer to "the `gf2` type"
- adjust the top-level imports and tests
2024-09-19 12:14:20 +02:00
Renamed from tests/elements/test_gf2.py (Browse further)
