[rahul05ranjan]

Fixes #155

What's wrong

parse() only used the frontmatter block when it parsed to a dict:

fm_data = handler.load(fm)
if isinstance(fm_data, dict):
    metadata.update(fm_data)

Anything else — a plain string, a list, a number — was thrown away, along with the source text it came from. Nothing warned, so a loads() → dumps() round-trip silently deleted part of the file:

>>> frontmatter.loads("---\nsome markdown text\n---\nbody\n").content
'body'

The most realistic case is a Markdown file using --- thematic breaks near the top: everything between the first two rules parsed as a YAML string, wasn't a dict, and vanished. A list-valued block (---\n- a\n- b\n---) is valid YAML with the same fate.

The fix

Treat the whole file as content when the block parses to a non-mapping value — the same thing the no-frontmatter path already does. Empty frontmatter (which parses to None) keeps its existing behavior, so check() and the empty-frontmatter tests are unchanged.

>>> post = frontmatter.loads("---\nsome markdown text\n---\nbody\n")
>>> post.metadata
{}
>>> post.content
'---\nsome markdown text\n---\nbody'

parse()'s docstring is updated to document this; tests are in tests/unit_test.py as suggested in CONTRIBUTING.

Verification

• New tests (test_non_mapping_frontmatter_is_content, test_list_frontmatter_is_content, test_thematic_break_file_roundtrip_loses_no_text) fail on main, pass with this change.
• pytest . --doctest-modules --doctest-glob "README.md": 61 passed, 9 failed with the fix vs 58 passed, 9 failed on main — the failures are all the pre-existing TOML-suite ones (this host has no toml package installed) and are identical on both runs; the +3 delta is exactly the new tests.
• black --check . clean; mypy couldn't run locally (host lacks the toolchain), happy to fix anything CI flags.

AI tools

Yes — this issue (#155) and this PR were prepared with the help of an AI coding assistant (GLM, via OpenRouter): finding the repro, drafting the fix, tests and this write-up. The verification commands above were run directly against this branch and against main, and their output is quoted as it was produced.

[rahul05ranjan]

Saw this was closed — thanks for looking. If the behavior change itself was the concern (I read the empty-None case as intentional), I'm happy to drop this. If a different approach to --- thematic breaks would be welcome instead, point me at it and I'll rework.