Recent commits to derivepassphrase.git (986509a3106caaec52888c75820b2406eee266c4)

Restructure testing data and testing machinery tests for better coverage

2026-08-29T19:17:34+02:00

We restructure `tests.data.SUPPORTED_KEYS` and
`tests.data.UNSUPPORTED_KEYS`, introducing a third category
`tests.data.CONDITIONALLY_SUPPORTED_KEYS` that contains the DSA-class
entries from the `UNSUPPORTED_KEYS` table, which are suitable for use
with vault if the agent supports deterministic DSA signatures.  We then
test the stub SSH agent's sign operation against these conditionally
supported keys as well, for the "with deterministic DSA" variant of the
stub agent.

This improves code coverage, because although such code paths were
previously already exercised in other tests of the test suite, with this
change, the `test_000_testing_machinery` module alone is sufficient to
achieve full code coverage on the stub SSH agent, in all three
variations.

Since we plan to extract the stub SSH agent from derivepassphrase's test
suite, the `test_000_testing_machinery` module alone contains all the
necessary code to test the extracted stub SSH agent (and its test keys).
</pre>

Fix typing and linting errors after tool upgrade

2026-08-29T18:45:38+02:00

</pre>

Update hatch use to "hatch check", and ruff to >= 0.16.0

2026-08-29T18:35:09+02:00

As part of the ruff upgrade, the hatch configuration uses rule names now
instead of rule codes.  The code still uses the rule codes, because
these are currently the only non-tool-specific notation available.
</pre>

Fix minor mistakes in old changelog entries

2026-08-29T18:03:13+02:00

  - The `"correctness"/"regression" split` phrase does not break
    properly in the browser when used with quotes.  The browser appears
    to have trouble finding the word boundaries in this case.

  - The entry for the introduction of SSH agent socket providers
    mentions that an Annoying OS-specific provider is yet to be written,
    even though two entries further down, that same socket provider is
    being introduced.
</pre>

Document recent test suite changes in the changelog

2026-08-29T18:01:11+02:00

  - Mention the "import unavailable" simulation, and the resulting
    reduction in test configurations.

  - Mention that Python 3.15 is now supported (and tested).
</pre>

Update compatibility info for Python 3.15

2026-08-29T17:32:01+02:00

</pre>

Repair the --version output test for derivepassphrase export vault

2026-08-23T17:23:09+02:00

The test was no longer triggering the fallback behavior for when the
"export" extra/the "cryptography" library was not available.  Partially,
this was because we now need to parametrize this test for both cases,
but forgot to do so.  Partially, this was *also* because querying the
installed version of "cryptography" needs to be separately mocked,
beyond just blocking the import.
</pre>

Officially support and test on Python 3.15

2026-08-23T14:58:43+02:00

Currently at release candidate 1.
</pre>

Fix hypothesis dependency on PyPy 3.9 and PyPy 3.10

2026-08-23T12:56:29+02:00

Because `hypothesis` requires a Rust toolchain for building 6.156 and
higher, and because this is not installable on PyPy 3.10 and lower, pin
`hypothesis` to 6.155.* and lower on PyPy 3.9 and 3.10.

This was already broken for quite some time, but cropped up again after
redoing the virtualenvs in commit
4fbfed4a4de887bd35f295863ea7548989024bf7.
</pre>

Always test the "no cryptography support" scenario, shrinking the test matrix

2026-08-23T12:44:07+02:00

Even when the `cryptography` library is available, we test the fallback
behavior of `derivepassphrase export vault` by inhibiting the loading of
`cryptography`.  As a result, we no longer need a separate testing
environment for the "no cryptography support" case, and can therefore
halve the number of configurations in our testing matrix.  We also no
longer need some associated configuration and helper objects.

In hindsight, I cannot believe I never tried to do a web search on how
to inhibit loading of a module in Python, and rather assumed that just
because pytest does not prominently offer a dedicated top-level function
or a dedicated `monkeypatch` method, that this is not supported.  But
seeing the underlying documentation for `importlib.import_module` and
the documentation of how `sys.modules[...] = None` affects the module
resolution, it becomes immediately obvious that this can be handled with
monkeypatching, and does not need a separate virtualenv.
</pre>