LaTeX Generator Roadmap

Custom LaTeX Generator Roadmap

Last updated: 2026-03-10 (status snapshot) Owner: Generator + NowType integration

Companion analysis:

• pdf-generator-gap-analysis.md

Status legend:

• [x] done
• [ ] _(in progress)_ started but incomplete
• [ ] not started / pending

Goal

Make the Hugo + NowType pipeline behave like a practical LaTeX book/article toolchain:

• predictable pagination
• stable counters and references
• high-quality print/PDF typography
• editable source-of-truth Markdown

Current Baseline (already in flight)

• PDF-style in-page editing mode (single + spread)
• A4 paper mode and print controls
• Book/article mode toggle
• header/footer and ToC UI support
• chapter/part/book scope editing + print flow

Missing Concepts (prioritized)

Phase 0: Core Pagination Correctness (highest priority)

• Replace fixed pixel slicing with explicit break offsets (computeNowtypePdfPageBreaks()).
• Break only at legal points (block boundaries + line-box boundaries).
• Never break inside a line box. (in progress)
• Enforce no-overlap invariant: page i+1 starts exactly at page i end.
• Add widow/orphan guards (minimum lines at top/bottom) with line-cut penalties.
• Keep headings with next block when possible.
• Keep figure/table blocks together where possible.
• Fix list-container overcapture so ul / ol no-break ranges do not absorb following paragraphs.
• Allow structurally legal list breaks between items when a whole list will not fit on one page.

Acceptance:

• No repeated lines across adjacent pages.
• No cut titles/headings at page bottom.
• Bullet lists never “own” following non-list paragraphs.
• Bullet-list pages do not collapse into artificial half pages because of false no-break ranges.
• Stable page content under hover/click/scroll.

Phase 1: Counter + Cross-Ref Model

• Unified counter registry: part, chapter, section, figure, table, equation, footnote.
• Configurable numbering schemes:
  • equations: (n) / (chapter.n) / (part.chapter.n) (in progress)
  • figures/tables by chapter.
• Deterministic counter reset rules (chapter/part transitions).
• Multi-pass label resolution for \label/\ref/\eqref/\pageref.
• Forward-reference placeholder + second-pass fixup.

Acceptance:

• Same equation/reference numbers in view mode, PDF mode, and print output.
• No numbering drift when switching chapter/part/book scopes.

Phase 2: Float Placement Engine

• Introduce float classes: inline, top, bottom, page-float, margin-float. (in progress)
• Parse place / placement / float hints into browser paginator penalties and hard breaks.
• Placement policy with penalties (h/t/b/p-style behavior). (partial: top and page-float now survive into committed compose-root page starts; queue-feasible bottom floats can now pull earlier prep breaks and land materially lower in the page; a bounded handoff path can now honor the common top -> bottom -> page-float sequence when legal breakpoints exist between those floats; there is still no true global deferred float queue yet.)
• Figure/table keep-together + caption keep-with-parent rules (best-effort in paginator/CSS).
• Smart shrink-to-fit policy before defer-to-next-page. (in progress)
• Float queue diagnostics in dev mode. (partial: committed pagination now records float queue outcomes in the smoke API, including honored vs deferred status and simple blocked-by-earlier-float attribution.)

Acceptance:

• Large figures/tables do not split or duplicate across pages.
• Placement remains stable during edits unless content truly changes.

Phase 3: Footnotes, Margin Notes, Citations

• Footnote extraction and per-page placement model. (in progress: pagination now records per-page note-load diagnostics and penalizes candidate breaks that would overflow the note lane, and committed note assignments can now spill onto later pages in screen/print preview. The runtime also now repairs footnote refs back into the canonical markdown buffer during nt_changed sync, resolves misparsed live previewcite anchors against source footnote definitions, and can render committed footnotes into a real bottom deck in both spread/print page shells and single-page PDF viewport mode; the live editor DOM still lacks native .footnotes, and pagination still does not build line/page breaks jointly with footnote height the way TeX does.)
• Margin-note lane with collision avoidance and leader markers. (in progress: the compositor now estimates per-page note load and exposes overflow/fill telemetry in the smoke API, with stable spill-forward note ownership across committed pages, but it still lacks leader markers and true note reflow.)
• Optional full citation expansion in margin mode.
• Citation style layer (numeric/author-year) from one source model.
• Bibliography pass with deterministic sort/order.

Acceptance:

• Margin notes mode does not overlap body text.
• Footnotes and citations match print/PDF output consistently.

Phase 4: Typography + Composition Quality

• Hyphenation dictionaries and soft-hyphen pipeline.
• Better justification strategy (decent badness metric).
• Micro-typography options (protrusion/spacing approximations).
• Better math baseline and display spacing harmonization. (in progress)
• Consistent print font sizing against LaTeX targets. (in progress)

Acceptance:

• Visual rhythm close to LaTeX reference pages for same content.
• Reduced rivers/overfull lines in dense text.

Phase 5: Tables and Multi-Page Structures

• Long-table pagination with repeated headers.
• Keep row integrity (avoid split row fragments).
• Column width balancing per page.
• Better handling for wide tables (scale/rotate/fallback policy). (in progress)

Acceptance:

• Long tables flow over pages without cutoff or duplicated chunks.

Phase 6: Book Semantics + Front/Back Matter

• Recto/verso rules and optional blank-page insertion. (in progress)
• Frontmatter modes: title page, abstract, ToC, LoF, LoT. (in progress)
• Running heads from section/chapter marks.
• Backmatter conventions: bibliography, appendix, index.

Acceptance:

• Book mode follows expected recto/verso + first-page behavior.

Phase 7: Performance and Responsiveness

• Incremental pagination (reflow affected region only).
• Background worker for break scoring where feasible.
• Debounced recomposition in spread mode.
• Instrumentation: pagination time, DOM nodes, repaint hotspots.

Acceptance:

• Spread mode interaction responsiveness close to single-page mode.

Phase 8: Test Harness (must-have regression gate)

• Browser smoke tests:
  • click-to-type focus
  • --- section/page-break behavior
  • spread navigation and editing parity (in progress)
  • no duplicate figure/table between pages
  • equation numbering format correctness
• Snapshot tests for key chapter fixtures.
• Debug overlay toggle for breakpoints and penalties.

Acceptance:

• Regressions are caught before shipping editor changes.

Implementation Notes (architecture constraints)

• Keep Hugo as render authority.
• Keep Markdown as source of truth.
• Keep NowType as editing surface; do not duplicate Hugo render hooks in client JS.
• Reuse existing pipeline entry points:
  • layouts/partials/preprocess.html
  • layouts/_default/_markup/render-passthrough.html
  • cdn/custom/toggleMarkdown.js
  • cdn/editor/src/*.js + rebuild to nowtype.min.js

Immediate Next 3 Tasks

1. Finish strict no-line split behavior (hard guarantee, not just scored preference).
2. Add smoke checks for duplicate figure/table and equation numbering format.
3. Implement spread editing parity validation (edit on either page and verify canonical root sync).

Progress Log

• 2026-03-10: Roadmap created.
• 2026-03-10: Status snapshot added (done/in-progress/pending) against current implementation.
• 2026-03-10: Paginator hardened to avoid raw-pixel fallback breaks; break selection now snaps to legal candidates and avoids inside-line cuts.
• 2026-03-10: Smoke checks added for equation numbering format and cross-page block splits (headings/figures/tables).
• 2026-03-10: Smoke check added for spread edit parity (activate spread side + type into canonical root).
• 2026-03-10: Smoke runner hardened with explicit stage logs and hard timeouts around firefox.launch, browser.newContext, context.newPage, and browser.close to fail fast in degraded browser-worker conditions.
• 2026-03-27: Heading keep-with-next no longer captures an entire ul / ol; list-following headings now anchor to the first li so page breaks can still occur between later list items.