Recent commits to derivepassphrase.git (b725e5f10e6031cf0781c8593fe854b2046a3bae)

Add missing tests for `SSHAgentClient.query_extensions` runtime errors

2024-11-27T20:23:06+01:00

Currently, this is still a white-box, maliciously compliant design that
relies heavily on implementation details of the function under test.
That said, the failure cases being triggered here are all SSH agent
protocol violations, which would otherwise be difficult to trigger in
"real" SSH agents.
</pre>

Retire the use of symlinks for the README and the changelog

2024-11-27T18:21:17+01:00

Symlinks interact very poorly with online git repository viewers, and
while they are offline viewable on Unixish systems, readers only see the
source view, not the rendered view.

For the documentation frontpage `docs/index.md`, use the Snippets
extension (Material for MkDocs/Python-Markdownx) to include the contents
of `README.md`.  This will allow future customization of the frontpage
by either appending further content or only embedding parts of the
README.

For the changelog, because we use `scriv` to maintain it
semi-automatically and thus want the changelog snippets to stay out of
the top-level directory, we do not want `docs/changelog.md` to symlink
to or to snippet-include the top-level `CHANGELOG.md`.  Instead, we let
`CHANGELOG.md` point to the `docs` directory (in prose), and purge the
symlinks to the SVG badges.  The original intent was to make the
changelog renderable at the top-level, without the MkDocs machinery, but
this is only really feasible if the changelog is handwritten, which
hasn't been the case anymore for quite some time already.
</pre>

Use canonical test keys in all SSH agent key listings

2024-11-27T15:10:52+01:00

For all documentation, whenever a formatted key listing of suitable keys
is presented, use the test keys from the test suite if possible.

Also, correct a typo.
</pre>

Unmark icon- and badge-type external links as external

2024-11-27T14:16:59+01:00

Icon-/badge-only external links lose their visual succintness if they
include the external link marker afterwards.

Also use this opportunity to rewrite the external links CSS selectors
using the `:is` and `:has` pseudo-classes, which leads to much less code
repetition.  Supposedly, every major browser has had a release with
`:has` support since December 2023, so browser compatibility shouldn't
be a major issue.
</pre>

Document that `is_suitable_ssh_key` now takes an optional agent client in the changelog

2024-11-27T13:29:15+01:00

</pre>

Add an ecdsa-sha2-nistp521 SSH test key

2024-11-27T00:09:55+01:00

This case was apparently forgotten when adding the 256- and 384-bit
keys.
</pre>

Publish polished `is_suitable_ssh_key` and `has_deterministic_dsa_signatures` interfaces

2024-11-26T23:28:43+01:00

The `has_deterministic_signatures` function internally only ever checked
whether DSA signatures were known deterministic, because currently,
signature schemes are either deterministic by design or they are
DSA-like and can be derandomized via RFC 6979 or a similar procedure.
There's no guarantee this dichotomy will stay this way in the future.
Thus it is better to rename the function to match what it actually
tests: Does this agent use deterministic DSA and ECDSA signatures?  We
do just that.

In a similar vein, the `Vault._is_suitable_ssh_key` only really checks
if the key type is known deterministic, not whether the key is suitable;
the latter depends on the SSH agent, and requires a call to the old
`has_deterministic_signatures` function.  We could of course analogously
rename `_is_suitable_ssh_key` into `is_known_deterministic_key_type` or
similar, but this feels too much like exposing implementation details to
the API user.  It seems better to expose a `Vault.is_suitable_ssh_key`
method that actually does what it advertises: check whether a key type
is known deterministic under a given SSH agent, or under all SSH agents
in general.  So we do just that.

Finally, we clean up some inconsistencies in the `query_extensions`
docstring, and some missing SSH agent clients not passed on to the calls
to the `Vault.phrase_from_key` function in the tests.
</pre>

Let the `running_ssh_agent` test fixture report the agent type

2024-11-26T14:26:21+01:00

In the current test scenario, where multiple SSH agents are spawned if
possible, it is highly unhelpful to know *that* a running SSH agent
failed, but not *which* agent did.  For debugging purposes, it is better
if the `running_ssh_agent` test fixture reports not only the agent's
socket, but also its type.

It is sufficient to have the type passed as a fixture output/test
function input, because `pytest` will then pretty-print it when a test
function fails.
</pre>

Decouple deterministic signatures from general SSH agent detection

2024-11-26T14:12:53+01:00

Instead of tying deterministic signatures directly to the detection of
Pageant specifically, add a general mechanism for attempting to infer
the connected SSH agent from its reported list of extensions.  This
moves the question of *how* we detect certain SSH agents out of the
deterministic signature checking function.

Alas, OpenSSH does not support the extension query message we issue,
despite them supporting the extension system in general *and* stewarding
the SSH agent protocol specification which defines this message
normatively.  So our implementation must tolerate a moderate level of
spec violation.
</pre>

Fix test suite to actually test deterministic signature support

2024-11-26T14:03:34+01:00

So far, the test suite was silently passing for me, because it requires
either a patched version or a not-yet-released version of PuTTY to
actually run the tests against Pageant (which is the main beneficiary of
deterministic signature detection).  Actually plugging in a suitable
patched Pageant version revealed a couple of key places where we
silently assume that the key type alone determines its suitability for
`derivepassphrase`.  This commit rectifies that.
</pre>