Respect an explicit excerpt of 0 in page front-matter

[Muhtasim-Munif-Fahim]

Fixes #317.

The bug

CheckFrontMatter guards the excerpt fallback with a truthiness check:

if not metadata.get("excerpt"):
    blog = Blog(self.app)
    node["excerpt"] = blog.post_auto_excerpt

0 is falsy, so an explicit :excerpt: 0 is indistinguishable from no excerpt at all and gets overwritten with post_auto_excerpt. The first paragraph of the post then leaks into every postlist rendered with :excerpts:, which is what #317 describes.

The same option given to the post directive already behaves correctly, because process_posts guards the identical fallback with is None:

if node["excerpt"] is None:
    node["excerpt"] = auto_excerpt

So the two ways of declaring a post disagree. This aligns the front-matter path with the directive path.

Reproduction

Two posts, identical except for how the excerpt is declared:

.. post:: 2020-12-01
   :excerpt: 0

:blogpost: true
:date: December 01, 2020
:excerpt: 0

Before this change the directive post correctly renders no excerpt, while the front-matter post renders its first paragraph. After, both render nothing.

Tests

Adds roots/test-excerpt and src/ablog/tests/test_excerpt.py covering four cases:

• an explicit 0 suppresses the excerpt,
• a non-zero count still yields that many paragraphs,
• an absent value still falls back to post_auto_excerpt,
• an explicit 0 wins over a non-default post_auto_excerpt.

The first and last fail without the one-line change; the middle two are controls that pass either way. Full suite is 13 passed on Sphinx 9.1.0.

One note

The new tests carry a filterwarnings mark:

pytestmark = pytest.mark.filterwarnings(
    "ignore:'ablog.post.CheckFrontMatter.app' is deprecated"
)

CheckFrontMatter reaches the app through SphinxTransform.app, which Sphinx 9 deprecates (the replacement is the private env._app). Since the suite turns warnings into errors, any test exercising the front-matter path trips it. That is a pre-existing compatibility issue orthogonal to this bug, and migrating to a private attribute felt like your call rather than something to smuggle into a bugfix — happy to fold it in here or open a separate issue, whichever you prefer.

🤖 Generated with Claude Code