Write one tableParaStyle entry per paragraph (#81)

[leogdion]

Fixes the bug behind both the #64 monospace failure and the dark-on-dark code panel in #78's render pass: item-level formatting reached only the first paragraph of a multi-paragraph text box.

Branches off v0.1.x directly — independent of the #78 and #65 stack.

Cause

collapsedEntries run-length collapsed adjacent identical paragraph formats into a single tableParaStyle entry at offset 0. That reads like a reasonable optimization, but Keynote treats each entry as a paragraph boundary marker, so collapsing left later paragraphs unstyled.

The archive stayed valid and nothing errored — the defect was only visible by opening a slide, which is why it survived #37's render verification (that deck is one paragraph).

What the fixture says

A human-authored 5-paragraph body storage in build_action_B.key:

STORAGE id=2651751  text=Body Level One\nBody Level Two\n…
  paraEntries=5
    char=0  -> 2651127
    char=15 -> 0
    char=30 -> 0
    char=47 -> 0
    char=63 -> 0

One entry per paragraph. The style rides entry 0; every later boundary carries identifier 0 = "same as the preceding entry". The entry must exist even though it references no record.

Fix

paragraphEntries emits one entry per paragraph, writing identifier 0 where a format repeats. Fork records are still deduped — the dedupe is on records, not entries. applyParagraphStyles filters identifier-0 entries out of the header references, since 0 isn't a record.

Two existing tests changed

itemWideAlignment and dedupeCollapsesEntries asserted the collapsing behavior — they encoded the bug. Both now assert the per-paragraph shape. Flagging explicitly since changing a green test to make a fix pass deserves scrutiny; the fixture evidence above is why I'm confident the new assertion is the correct one.

Everything uses three or more paragraphs

Deliberate. A two-paragraph case exercises only the first and last and passes while interior paragraphs stay broken — which is exactly how the per-span variant of this bug survived the first #64 probe round, where iiii and 1111 rendered correctly and MMMM didn't.

Verification

• swift test — 70 green, including a new MultiParagraphFormattingTests suite
• LINT_MODE=STRICT ./Scripts/lint.sh — exit 0
• swift run AcceptanceDecks — multi_paragraph_formatting.key writes and self-checks

Structural only. This bug was invisible to structural checks by definition, so the render pass is the real gate.

Render checklist (yours)

Deck generated at /Users/leo/Downloads/para-fix/multi_paragraph_formatting.key.

1. Slide 1 — all three lines Menlo, equal width, large. Before the fix only iiii was.
2. Slide 2 — all five lines bold and red.
3. Slide 3 — the middle line matches its neighbours. This is the case per-span styling missed.
4. Slide 4 — left / centre / right / left / left alignment still honored, i.e. the fix didn't regress TextBox layout & list styling: plain-by-default, list styles, alignment/indent, rotation, columns #51.

If this renders green, #66 unblocks and I'd also restore the multi-line code sample in #78's background_fill slide 1.

Fixes #81

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: c1215832-86bc-4718-aed8-a9ea37053635

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[leogdion]

Author — the crash was mine, and it's fixed. Thanks for catching it; a crash on open is a worse failure than the unstyled render this PR set out to fix.

What happened

The per-paragraph entries were the right idea. Writing them was not.

For a repeat entry I set entry.object.identifier = 0. In swift-protobuf that materializes the optional object message as a present-but-empty TSP.Reference. Keynote resolves it to nil and crashes — the same id-0 trap already recorded twice in agent-notes.md, once for RecordCloner remaps and once for plain character spans.

I saw the template's repeats reported "identifier 0" and generalized from that without checking how they actually serialize. The bytes are unambiguous:

template repeat:  [08 0f]         <- characterIndex only, no object field
what I wrote:     [08 0f 12 00]   <- present-but-empty reference

Reading the identifier back cannot distinguish them — absent and zeroed both report 0. Only hasObject does. That is also why my original test passed: #expect(entries[1].object.identifier == 0) is true in both the correct and the crashing shape.

The fix

1. applyParagraphStyles assigns object only for real identifiers.
2. UUIDMapVerifier rule 7 — no storage object-attribute entry may carry a present-but-empty reference, checked across tableParaStyle, tableCharStyle, and tableListStyle. Every deck goes through the verifier on write, so this class of crash can't ship again from any code path, not just this one.
3. A regression test asserting hasObject is false and the entry serializes to exactly [08 04].

I confirmed the new test fails against the crashing version before restoring the fix — a test that can't fail would have been worthless here, and the previous one proved that.

Verification

• 71 tests green, STRICT lint exit 0
• All 12 acceptance decks regenerate clean under rule 7
• Audited the two other object.identifier = assignments in the surgeon: both always receive real identifiers (character runs mint a style per the Support mixed formatting (runs) within a single text box #40 directive, list styles are always real), so neither can hit this

Re-render, please

Fresh deck at /Users/leo/Downloads/para-fix2/multi_paragraph_formatting.key.

1. It should open at all — that's the new bar
2. Slide 1 — all three lines Menlo, equal width, large
3. Slide 2 — all five lines bold red
4. Slide 3 — the middle line matches its neighbours
5. Slide 4 — left / centre / right / left / left still honored

Recorded in research/findings/text_formatting.md and as a standing directive in agent-notes.md, since the distinction is invisible to the obvious assertion.