Recent commits to derivepassphrase.git (59d73c8320650a6bcb77eed759c694949d350cdb)

Update the CLI to use the translatable strings

2024-12-30T23:45:01+01:00

We implement the translations in a deferred manner, via objects that
stringify to their correct translation.  Since `click` 8.1, help texts
are (mostly) stored first and processed only when the help page is
emitted.  Accordingly, this system can be adapted to support our
translatable strings: we need only ensure the help text is stringified
just before it is formatted and emitted.  This in turn can be achieved
by reimplementing the help page assembly and formatting in a custom
`click.Command` class, as has already been done to implement help/option
group support.  Once this is in place, it remains to exchange all inline
help texts with the corresponding translatable string, and adapt the
tests to the new messages.

While most of the changes are straightforward, some subtleties remain.

First, the help texts are now supplied as (separately translatable)
paragraphs to the `click.command` decorator, instead of being extracted
from the respective function docstring.  The docstrings in return have
been shortened down.

Second, since the new capabilities are implemented via custom
`click.Command`, `click.MultiCommand` and `click.Option` subclasses,
every option and command must derive from one of these classes.  This
necessitates the creation of an empty `StandardOption` class for options
without an option group, and making `_DefaultToVaultGroup` a subclass of
`CommandWithHelpGroups`.

Third, since the `click.Option` class actually processes its help text
eagerly, the subclass must hide the (non-string) help text from the
`click.Option` constructor and later restore it.

Finally, while the tests currently pass as before, they now implicitly
rely on *not* loading any translations.  Furthermore, there is no
mechanism in place with which to isolate the translation system from the
user's environment for testing purposes, nor is there a (controlled)
mechanism to test `derivepassphrase` under a different language.  Once
there are actual translations, future commits will need to establish
such a missing mechanism for translation loading/switching/forcing.
</pre>

Add more translatable strings and serialization machinery

2024-12-30T21:34:42+01:00

Add all help texts and metavars as translatable strings.  Furthermore,
add extra machinery for introspection to the serialized translatable
strings class, which will make debugging values in pytest much easier.

Add another couple of error messages (and update the manpages as well)
for low-level errors while querying the SSH agent for the key list or
while requesting a signature.  (Previously, we forwarded the internal
English exception message directly.)  Update the manpage to include
these as well.
</pre>

Extract translatable log messages into separate module

2024-12-25T21:41:22+01:00

Extract all log messages into a separate module, so that they may be
translated in a future version.  We anticipate the use of such a message
in a `logging` call, so we provide a wrapper object that defers the
string interpolation until its serialization is called, and caches the
result.  (This is all as per the recommendation from the standard
`logging` module.)  Implemented this way, it is easy to run easy
transformations on the specific log message template, e.g. substituting
a translated message template, or trimming the filename placeholder if
the filename is `None`.

As an added benefit, this organization also makes it somewhat easy to
ensure message consistency across different use sites and to generate
diagnostics lists for inclusion in the manpage.

While we intend to use the gettext toolset to implement the translation
later, we do *not* currently use the standard `_` gettext alias or
otherwise mark up the strings in an `xgettext`-compatible manner.  Both
the GNU version of `xgettext` and the Python work-alike `xgettext.py`
suck: the GNU version does not completely understand Python syntax and
cannot sensibly parse `black`-/`ruff`-formatted code using implicit
string concatenation or trailing commas in function argument lists, and
`xgettext.py` is so outdated (even in current Pythons) that it does not
support `xgettext`'s `--add-comment` option for extracting extra
comments for the translators.  So instead of attempting to cram the
strings and comments into some formatter-resistant and
`xgettext`-compatible shape (a very futile endeavor so far), store
enough information that we could generate the `.po` files ourselves with
very little additional effort.
</pre>

Add an actual derivepassphrase-vault(1) manpage, plus better manpage contents

2024-12-25T20:55:45+01:00

The barebones HTML "manpage" for `derivepassphrase` was essentially an
HTML-ized version of the help text.  In particular, the option
descriptions only contained the one-line help text, and many typical
sections (FILES, ENVIRONMENT, EXAMPLES, DIAGNOSTICS, AUTHOR, BUGS) were
missing.

This commit contributes a proper manpage, in `*roff`/`mdoc` format.  It
was authored using guidance from the `groff_mdoc` documentation, the
only sensible-appearing authoritative source on manpage style issues.
(Perl's `perlpodstyle`, `groff_man` and `groff_man_style` were also
consulted, but ultimately disregarded.)  The content changes were also
folded back into the HTML version, save for some styling choices that
aren't possible in the `*roff` version.  In particular, both versions
are hand-written, and manually synchronized with each other.

The manpages use the option group headings "Passphrase generation" and
"Compatibility and extension options" instead of "Password generation"
and "Options concerning compatibility with other tools", for better text
flow.  Accordingly, this phrasing has been adapted in
`derivepassphrase_vault` as well.
</pre>

Make the mutually exclusive options helper function print the correct parameter name

2024-12-25T18:59:32+01:00

The `click.Parameter.human_readable_name` attribute yields the metavar
for arguments, and the function parameter name for options.  The
documentation calls the latter the "name" for the option, perhaps under
the assumption that you would never pass an explicit parameter name to
the option decorator.  I disagree with this, it is horrible design:
sometimes the value is usable as a function parameter (options),
sometimes it isn't (arguments), and vice versa for constructing
a command-line, so it is neither usable as a command-line parameter
display name nor as a function parameter name.

So what *is* it even good for...?

Include our own logic to select an appropriate display name for our
options.  This requires querying `click.Parameter` attributes that
should be public, but are internal.  Why...?!
</pre>

Add small fixes to changelog, docstrings and variable names

2024-12-25T17:15:23+01:00

</pre>

Instruct `coverage` to record the test function for each covered line

2024-12-21T01:02:34+01:00

Enable the standard dynamic context setting `test_function` for
`coverage`.  This is immensely helpful for debugging whether and which
tests trigger or don't trigger a certain branch of code in
a multi-branch block (`if`/`elif`/`else`, `match`/`case` blocks,
dispatch tables, etc.).
</pre>

Fix formatting, some coverage pragmas and some linting rules in tests

2024-12-21T00:57:49+01:00

Aside from formatting fixes, disable the E501/line-too-long and
C419/unnecessary-comprehension-in-call linting rules in tests, where
they hinder debuggability of the tests.  (See embedded comments for full
rationale.)  Furthermore, make `coverage` recognize `@overload` lines
automatically as lines that should not be included in coverage, like
`assert False` and friends.
</pre>

Support exporting the `vault` configuration as a POSIX shell script

2024-12-21T00:23:33+01:00

When exporting `vault` configurations via `--export`, a new option
`--export-as=FORMAT` allows selecting the export format (defaulting to
`json`).  Setting `FORMAT` as `sh` selects the new POSIX shell script
export format, which yields a sh(1)-compatible script to reimport the
existing configuration as a series of calls to the `derivepassphrase`
command.

Because the output is very regular, the test suite also includes an
interpreter specifically for the sh(1) subset emitted during an `sh`
export, on top of the usual adaptations to existing tests for the new
functionality.
</pre>

Allow unsetting settings when configuring `vault`

2024-12-20T17:00:48+01:00

When configuring a `vault` service, or the global settings, a new
configuration option `--unset FIELD` will unset the specified `FIELD`
from the service or global settings prior to and addition to applying
the requested settings changes.  This way, the user can update all
settings on the command-line (except for the "notes") without manually
editing and reimporting a configuration export.  (It is permissible to
only unset settings, without applying any other configuration changes.)
</pre>