cases.json contains all 147 HTML fixtures from upstream test/index.html, plus the 50 independently written cases in additional.json. The upstream fixtures retain their MIT license in LICENSE. Production Rust code does not include the upstream JavaScript implementation.

What is compared

Turndown is configured to match Moli's ATX headings, dash bullets, star emphasis, fenced code, inline links, and horizontal rules. Per-case style-only options (such as referenced links or tilde fences) are normalized to this configuration. The semantic preformattedCode option is preserved and tested against Moli's preformatted_code option. This corpus tests conversion behavior, not Turndown's JavaScript plugin API or every output-style option.

Each case parses the source HTML through an independent test DOM implementing Dom, converts it, and checks that the DOM is unchanged. Both Markdown outputs are rendered with pulldown-cmark, then their HTML is compared exactly. This accepts equivalent delimiter choices and list indentation while detecting lost text, incorrect links, changed paragraph/list structure, and visible delimiters. There are also exact Markdown assertions in the original and regression suites.

expectations.json documents 16 deliberate differences with exact Moli outputs. These preserve bare pre blocks and avoid upstream losses of literal characters, emphasis, whitespace, and list boundaries, and normalize textual attributes for single-line dumps. They remain active assertions; no case is skipped. A reference change that makes a difference obsolete also fails the test. Separate converter tests cover GFM tables and strikethrough as Moli extensions. Explicit table headers follow Turndown's GFM plugin; simple headerless tables receive empty headings. Complex tables retain content-only HTML so merged cells and nested tables keep their associations. Presentational and layout-only containing tables expand into blocks; their simple inner tables can use GFM.

Findings tracked outside the reference corpus

The 16 exceptions above describe this corpus, not every design difference or every correctness issue. tracking.rs records these open review findings with active assertions of both Markdown and rendered HTML:

Tests	Status	Current behavior and comparison
`track_nested_list_spacing_through_a_transparent_wrapper`, `track_nested_list_spacing_before_a_trailing_empty_element`	Open compatibility decision	Moli keeps the outer list tight when an `ins` wraps the nested list or an empty `span` follows it. Turndown makes it loose. Both preserve the list items.

These tests are not ignored or expected to panic. A change to a recorded output requires review; when fixing a bug, replace the current-output assertion with a regression asserting the desired semantics and remove its open entry here.

The inline-code br separator and nested-emphasis closing-delimiter fixes have independent semantic regressions in inline_boundaries.rs. Inline-code breaks follow Moli's whitespace mode (one space after collapsing), without copying the extra spaces from Turndown's Markdown hard-break syntax.

Regeneration

Normal Rust tests use the checked-in data and require neither Node nor network access. To update the reference intentionally:

git clone https://github.com/mixmark-io/turndown.git /tmp/moli-turndown/upstream
git -C /tmp/moli-turndown/upstream checkout aa84dfa3e2361edea8c43acbfc2b7a9363494bfd
npm install --prefix /tmp/moli-turndown --ignore-scripts --no-audit --no-fund @mixmark-io/domino@2.2.0
node moli-html2md/tests/turndown/generate.mjs /tmp/moli-turndown/upstream
cargo test -p moli-html2md

The generator executes the pinned upstream source. It adapts import extensions for Node's ESM loader and provides CommonJS require for Domino, without changing conversion rules. The corpus records both the upstream revision and normalized options. cases.rs provides a separate named Rust test for every fixture.

The multiline image-alt, multiline link-title, paragraph-end nested emphasis, and intraword mixed-emphasis bugs are fixed. Their exact rendered-HTML regressions are in inline_boundaries.rs; tracking.rs retains only the two open list-spacing compatibility decisions.