fix(presentations): separate slides with a thematic break

[Ready22Race]

Fixes #31.

What

Emits Block::Rule between slides in the three presentation parsers (pptx, ppt, odp), which previously concatenated every slide's blocks with nothing in between.

A slide's title becomes a Heading, so decks that title every slide read correctly by accident. A slide with no title placeholder contributes no structural block, so its content is indistinguishable from a continuation of the previous slide — two untitled slides in a row produce two adjacent paragraphs, and bullet lists from different slides merge into one list.

Why a thematic break

No new concept and no page number: Block::Rule already exists, is already emitted for <hr> (src/shared/html.rs:432), and already renders as --- (src/render/markdown/mod.rs:201).

This is deliberately not the pagination #26 declined. That issue settled that a w:type="page" break is layout and dropping it is correct. A slide is a container in the source model rather than a layout artifact, so its boundary is document structure.

Shape of the change

Each parser now accumulates one slide into a local Vec<Block> and appends it only if non-empty, pushing a Rule first when output already exists. That gives two guarantees worth stating:

• never leading — no --- before the first slide
• never doubled — a slide that produces no blocks (empty layout, chrome-only placeholders) cannot leave a dangling or repeated break

ppt already accumulated per slide, so it only needed the guard.

Scope: core only, no binding changes

Block::Rule already exists and every binding already handles it, so nothing outside src/formats/ needs to move:

binding	maps model::Block::Rule	declares the rule kind
node	node/src/document.rs:88	node/index.d.ts:60
python	python/src/document.rs:83	python/anydoc/_anydoc.pyi:84
wasm	wasm/src/document.rs:96	wasm/src/typescript.rs:47

No new Block variant, no public API change, no type-surface change. Confirmed end to end through the Python binding on the deck from the issue:

>>> anydoc.to_document(data).blocks
['heading', 'rule', 'paragraph', 'rule', 'paragraph']   # kinds

Tests

Three unit tests in src/formats/pptx/mod.rs, building minimal packages in memory in the style of the sheet module's tests — they pin the behaviour rather than just the output:

• untitled_slides_are_separated — the defect itself: two title-less slides must not read as one
• separator_never_leads — a deck must not open with ---, and a single slide yields only its own content
• empty_slides_leave_no_dangling_separator — a slide contributing no blocks must not add a break, neither doubled between two real slides nor trailing

Verification

cargo fmt --all --check                                          clean
cargo clippy --workspace --all-targets --all-features -D warnings clean
cargo clippy -p anydoc-wasm --target wasm32-unknown-unknown       clean
cargo test --locked                                              194 passed, 0 failed
python -m unittest discover -s tests                             9 passed

Snapshot delta is 6 added --- lines and 6 added blank lines across 6 snapshots, with zero removals — no content changed, only separators appeared:

tests/snapshots/snapshots__odp__pres.odp.snap                  | 2 ++
tests/snapshots/snapshots__ppt__handmade-multimaster.ppt.snap  | 2 ++
tests/snapshots/snapshots__ppt__handmade-sparsenotes.ppt.snap  | 2 ++
tests/snapshots/snapshots__ppt__pres.ppt.snap                  | 2 ++
tests/snapshots/snapshots__pptx__handmade-links.pptx.snap      | 2 ++
tests/snapshots/snapshots__pptx__pres.pptx.snap                | 2 ++

Also checked by hand against a deck whose slides 2 and 3 have no title placeholder (the case in the issue) and against tests/fixtures/ppt/pres.ppt.

Note

Happy to take this in a different direction if you would rather the boundary be expressed some other way, or scope it to a subset of the three parsers. The behaviour change is visible in output, so it is your call whether it belongs in a minor bump.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[H0rowitz]

Independent verification, for whatever it's worth — we hit #31 on a Pandoc-authored deck while evaluating anydoc as the conversion engine for our ingestion pipeline, before finding the issue already filed.

Checked out 4b0edae and ran it against that deck (3 slides; slide 2 has no title placeholder — exactly the pathological case from the issue):

• cargo test --locked: 194 passed, 0 failed, matching the PR's own verification block.
• The two separators appear at exactly the two slide boundaries, none leading, none doubled.
• A diff of the full converted output against 8eecca5 is 4 added lines (two ---, two blanks), zero removals — and this is a producer (Pandoc) the fixture corpus doesn't currently cover, so it complements the PR's LibreOffice/handmade snapshot evidence rather than repeating it.

One note for anyone reading along, orthogonal to this PR: on that same deck the remaining structure loss is bullet formatting inherited from the layout (<p:ph idx="1"/> content placeholders lose their list markers) — the separate, bigger inheritance topic. The boundary fix makes such decks navigable regardless, and untitled slides stop merging into their predecessor.

[blueagledev]

Independent verification. We're evaluating anydoc as the extraction layer for a documentation-ingestion pipeline that consumes the document model rather than the Markdown serializer, so slide boundaries are load-bearing for us. In fairness: we tested this PR and #95 against the same deck, and we've posted parallel notes there. Both work; the differences are below.

Setup: checked out from refs/pull/32/head, cargo/rustc 1.95.0. Test deck is Pandoc-authored: 2 slides, slide 1 titled with an image and speaker notes, slide 2 has no title placeholder — the pathological case from #31.

cargo test --locked: 194 passed, 0 failed, 1 ignored. (The lower count vs. main's 217 is staleness, not skipped tests — this branch is based on 8eecca5, roughly 23 commits behind e754e1d/v0.1.9. It still applies cleanly to current main.)

Document model. Baseline v0.1.9 gives 4 top-level blocks with no marker of any kind — Heading -> Paragraph(Image) -> BlockQuote -> Table. With this PR, 5 blocks:

Heading -> Paragraph(Image) -> BlockQuote -> Rule -> Table

The untitled slide 2 separates correctly, exactly as the PR describes, and the never-leading / never-doubled guarantees hold on our fixture.

Where it left us short, and why we'd bet on #95. Our output format numbers slides (## Slide N), so we need an ordinal, and it can't be recovered here by counting rules. This PR's own passing test empty_slides_leave_no_dangling_separator asserts that a 3-slide deck with an empty middle slide yields exactly one rule — correct behaviour for a separator, but it means a consumer counting rules labels the third slide "2". #95 carries the ordinal in the marker itself (Inline::Anchor("slide-N"), derived from slide_paths.enumerate()), so it survives empty and unparseable slides.

The secondary point is that Block::Rule is a content construct — indistinguishable from a genuine <hr> in source content — where an anchor is a dedicated identity. Your write-up argues a slide boundary is document structure rather than a layout artifact, and we agree; that argument seems to us to point at a structural marker rather than a content one.

One thing this PR has that #95 does not: ODF coverage. On tests/fixtures/odp/pres.odp this separates slides (11 blocks, one rule) while #95's output is byte-identical to base (10 blocks, no marker). If #95 is the one that lands, .odp needs the same treatment as a follow-up — that gap is real and it's yours.

The two conflict: this applies cleanly to main, after which #95 conflicts in 6 files (both edit the same statement), so whichever lands first forces the other to rebase.

None of this is a knock on the change — it does what it says and the reasoning in the description is sound. Posting the comparison because a maintainer choosing between two competing PRs is better served by it than by a thumbs-up on each.