Recent commits to derivepassphrase.git (d1977d9a206d1bc787a00d1d71abb1374d492ba9)

Convert Parametrizations enums to Parametrize namespace

2025-02-01T23:25:35+01:00

Switch out `enum.Enum` for `types.SimpleNamespace` to collect the
various `pytest.mark.parametrize` decorators: it works equally well in
practice when coupled with a type checker, and it avoids having to
unpack the enum value when actually calling the decorator.  Opt for the
shorter name `Parametrize` instead of `Parametrizations`.

Also consolidate the two versions of the `VAULT_CONFIG_FORMATS_DATA`
parametrization set by moving it into a dedicated namespace in the
top-level `tests` package.
</pre>

Consolidate test parametrizations per test module

2025-02-01T15:55:49+01:00

Collect all calls to `pytest.mark.parametrize` in a per-module enum.
This deduplicates the definitions and highlights inconsistencies.

(For the CLI test module, the shell formatter functions were moved above
the enum to avoid NameErrors.)
</pre>

Resolve dependency versions explicitly in pyproject.toml

2025-01-31T18:03:02+01:00

Remove the `~=` version specifiers in favor of `>=` specifiers, and only
use explicitly tested versions as minimum version requirements.

Also explicitly list pytest plugins and coverage plugins as
dependencies.
</pre>

Fix inconsistent indentation in pyproject.toml

2025-01-31T15:37:32+01:00

</pre>

Add static analysis and testing dependencies to "dev" extra

2025-01-31T15:20:21+01:00

This also enables the static analysis environment to directly use the
"dev" extra instead of re-declaring all dependencies.  (Sadly, the
testing environment needs manual re-declaration, because `cryptography`
is a transitive dependency of `hatch`, but the test suite needs to test
the "no cryptography support" scenario too.)
</pre>

Remove type annotations from enum value definitions

2025-01-31T15:18:20+01:00

Apparently type checkers have always internally assigned new Literal
types to those enum names, and are now also instructed to flag explicit
type annotations as errors.

For Griffe/mkdocstrings-python, this means we'll need to fill the
attribute table manually.  *sigh*
</pre>

Fix formatting nitpicks from ruff

2025-01-31T15:03:58+01:00

</pre>

Document general changes to the test suite since 0.4.0

2025-01-31T15:00:24+01:00

</pre>

Document change in testing setup

2025-01-31T14:54:53+01:00

</pre>

Adjust hypothesis deadlines globally and redo the profiles and slowdown factors

2025-01-31T14:43:03+01:00

Previously, we would define a "worst-case" hypothesis profile for use
under instrumentation (coverage measurement), and decorate every
hypothesis test which we experienced to have deadline problems with this
decorator.  This step was manual, and very error-prone when generating
new hypothesis tests.  Conceptually, it also indicated that adjusted
deadlines applied for some tests, but not for others, even though the
deadline adjustment is actually a side-effect of global instrumentation.

We therefore now adopt a global solution, where the pytest configuration
file detects which coverage instrumentation (if any) is running, and
adjusts the default hypothesis settings appropriately.  The decorator is
removed.

In the course of this, we re-evaluated the profiles and the slowdown
factors, instead of assuming the worst-case instrumentation.  hypothesis
also defines a "ci" profile for use in continuous integration, together
with some commentary on how this profile is expected to be used.  We
cede this profile name to hypothesis (and remove it from our testing
matrix in hatch), and define our own "intense" profile for high-volume
hypothesis testing based on the "ci" profile, to be invoked manually as
a "pre-release" (or perhaps "nightly") run.

Finally, because the test suite now is rather slow in sequential use, we
add `--dist=worksteal` to the test options to better distribute load
during parallel testing.  We choose this over other more "balanced"
strategies because our single-test runtimes wildly fluctuate thanks to
hypothesis, so work stealing appears to be the most reliable
distribution heuristic for us.
</pre>