epub2md/pandoc/book.lua

--[[
book.lua – Pandoc filter for book files from epub2md_merge.py (HTML output, e.g. PDF via weasyprint).

Pandoc's --toc always puts the table of contents (and its own title block) before the whole text,
i.e. before the book's cover, title page and imprint. epub2md_merge.py marks the right places:

    <div id="epub2md-titlepage"></div>   after the cover, only if the book has no title page of its own
    <div id="epub2md-toc"></div>         after the front matter (cover, title page, imprint …)

This filter, for HTML output:
  * builds the table of contents at the placeholder (use it INSTEAD of --toc; --toc-depth is honoured),
  * drops Pandoc's title block when the book has its own title page, or moves it to the placeholder,
  * shows the cover (first chapter with nothing but one picture) without the caption "Cover",
  * turns the chapter anchors <a id="001-…"></a> into <div class="chapter-start"> so that print.css can
    start every chapter – including cover, title page and imprint – on a new page.

For LaTeX (PDF route md -> html -> pdf, filter in BOTH steps, no --toc) the table of contents becomes
\tableofcontents at the same place, every chapter starts with \clearpage and there is no \maketitle
before the cover. For other formats it only removes the placeholders (EPUB, DOCX, ODT: use --toc there).

Usage:
    pandoc buch.md -s --lua-filter /pfad/zu/epub2md/pandoc/book.lua \
      --lua-filter /pfad/zu/epub2md/pandoc/callouts.lua --css /pfad/zu/epub2md/pandoc/print.css \
      --pdf-engine=weasyprint -o buch.pdf
]]

local TOC_ID, TITLE_ID = "epub2md-toc", "epub2md-titlepage"
local TOC_TITLE = { de = "Inhaltsverzeichnis", en = "Contents" }

local function is_placeholder(block, id)
  return block.t == "Div" and block.identifier == id and #block.content == 0
end

local function esc(s)
  return (s:gsub("&", "&amp;"):gsub("<", "&lt;"):gsub(">", "&gt;"))
end

-- <header id="title-block-header"> as Pandoc's HTML template writes it.
local function title_block(meta)
  local html = { '<header id="title-block-header">', '<h1 class="title">' .. esc(pandoc.utils.stringify(meta.title)) .. "</h1>" }
  if meta.subtitle then
    html[#html + 1] = '<p class="subtitle">' .. esc(pandoc.utils.stringify(meta.subtitle)) .. "</p>"
  end
  local authors = meta.author
  if authors and pandoc.utils.type(authors) ~= "List" then
    authors = { authors }
  end
  for _, a in ipairs(authors or {}) do
    html[#html + 1] = '<p class="author">' .. esc(pandoc.utils.stringify(a)) .. "</p>"
  end
  if meta.date then
    html[#html + 1] = '<p class="date">' .. esc(pandoc.utils.stringify(meta.date)) .. "</p>"
  end
  html[#html + 1] = "</header>"
  return pandoc.RawBlock("html", table.concat(html, "\n"))
end

-- Paragraph that only holds chapter start anchors (<a id="001-cover"></a>) → id of the first one.
local function chapter_anchor(block)
  if block.t ~= "Para" and block.t ~= "Plain" then
    return nil
  end
  local id = nil
  for _, inl in ipairs(block.content) do
    if inl.t ~= "RawInline" then
      return nil
    end
    local found = inl.text:match('^<a id="(%d%d%d%-[^"]*)">$') or inl.text:match('^<a id="(%d%d%d%-[^"]*)"></a>$')
    if found then
      id = id or found
    elseif inl.text ~= "</a>" then
      return nil
    end
  end
  return id
end

-- Characters in running text that even DejaVu lacks: typeset as math symbols ("↩︎": footnote back-links).
local LATEX_SYMBOLS = { ["↩"] = "$\\hookleftarrow$", ["ℓ"] = "$\\ell$", ["\u{2329}"] = "$\\langle$",
  ["\u{232A}"] = "$\\rangle$", ["\u{3008}"] = "$\\langle$", ["\u{3009}"] = "$\\rangle$",
  ["∖"] = "$\\setminus$", ["⋯"] = "$\\cdots$", ["\u{FE0E}"] = "" }
local LATEX_FONTS = { mainfont = "DejaVu Serif", monofont = "DejaVu Sans Mono", mathfont = "DejaVu Math TeX Gyre" }

local function font_installed(name)
  local pipe = io.popen("fc-list -q " .. string.format("%q", name) .. " && echo yes")
  local out = pipe and pipe:read("a") or ""
  if pipe then
    pipe:close()
  end
  return out:match("yes") ~= nil
end

local function latex_symbols(str)
  local found = false
  for _, c in utf8.codes(str.text) do
    if LATEX_SYMBOLS[utf8.char(c)] then
      found = true
      break
    end
  end
  if not found then
    return nil
  end
  local out, buf = pandoc.Inlines({}), {}
  for _, c in utf8.codes(str.text) do
    local ch = utf8.char(c)
    local tex = LATEX_SYMBOLS[ch]
    if tex then
      if #buf > 0 then
        out:insert(pandoc.Str(table.concat(buf)))
        buf = {}
      end
      if tex ~= "" then
        out:insert(pandoc.RawInline("latex", tex))
      end
    else
      buf[#buf + 1] = ch
    end
  end
  if #buf > 0 then
    out:insert(pandoc.Str(table.concat(buf)))
  end
  return out
end

-- LaTeX (2nd step of the PDF route md -> html -> pdf): the HTML from step 1 already holds the table of
-- contents (div#TOC.toc), the chapter starts (div.chapter-start) and, if moved, Pandoc's title block.
local function latex_book(doc)
  local found = false
  doc.blocks:walk({ Div = function(d)
    if d.identifier == "TOC" and d.classes:includes("toc") or d.classes:includes("chapter-start") then
      found = true
    end
  end })
  if not found then
    return nil
  end
  local blocks = pandoc.Blocks({})
  for _, block in ipairs(doc.blocks) do
    if block.t == "Div" and block.identifier == "TOC" and block.classes:includes("toc") then
      blocks:insert(pandoc.RawBlock("latex", "\\clearpage\n\\tableofcontents\n\\clearpage"))
    elseif block.t == "Div" and block.classes:includes("chapter-start") then
      blocks:insert(pandoc.RawBlock("latex", "\\clearpage"))
      blocks:insert(block)
    elseif block.t == "Div" and block.classes:includes("cover") then
      blocks:insert(block:walk({ Image = function(img) -- cover as large as the page allows
        img.attributes.width = nil
        img.attributes.height = "20cm"
        return img
      end }))
    elseif block.t == "Div" and block.identifier == "title-block-header" then
      blocks:insert(pandoc.RawBlock("latex", "\\clearpage\n\\begin{center}\n\\vspace*{3cm}"))
      for _, b in ipairs(block.content) do
        if b.t == "Header" then
          local title = pandoc.Inlines({ pandoc.RawInline("latex", "{\\Huge\\bfseries ") })
          title:extend(b.content)
          title:insert(pandoc.RawInline("latex", "}"))
          blocks:insert(pandoc.Para(title))
        else
          blocks:insert(b)
        end
      end
      blocks:insert(pandoc.RawBlock("latex", "\\end{center}"))
    else
      blocks:insert(block)
    end
  end
  doc.blocks = blocks:walk({ Str = latex_symbols })
  -- fonts with a wide Unicode coverage (∈ λ ≠ ≥ □, thin spaces …), unless the user chose fonts himself
  -- A4 with moderate margins, like print.css (Pandoc's LaTeX default: US letter, wide margins)
  doc.meta.papersize = doc.meta.papersize or "a4"
  doc.meta.geometry = doc.meta.geometry or "margin=2cm"
  for var, font in pairs(LATEX_FONTS) do
    if doc.meta[var] == nil and font_installed(font) then
      doc.meta[var] = font
    end
  end
  -- no \maketitle at the very top (before the cover); keep the PDF metadata
  if doc.meta.title then
    doc.meta["title-meta"] = pandoc.utils.stringify(doc.meta.title)
    doc.meta.title = nil
  end
  if doc.meta.author then
    doc.meta["author-meta"] = pandoc.utils.stringify(doc.meta.author)
    doc.meta.author = nil
  end
  doc.meta.date = nil
  return doc
end

function Pandoc(doc)
  if FORMAT:match("latex") then
    return latex_book(doc)
  end
  local html = FORMAT:match("html") and not FORMAT:match("epub")
  local toc_at, title_at = nil, nil
  for i, block in ipairs(doc.blocks) do
    if is_placeholder(block, TOC_ID) then
      toc_at = i
    elseif is_placeholder(block, TITLE_ID) then
      title_at = i
    end
  end
  if not toc_at and not title_at then
    return nil -- not a file from epub2md_merge.py
  end
  if not html then -- placeholders are top-level blocks: just drop them
    local kept = pandoc.Blocks({})
    for _, block in ipairs(doc.blocks) do
      if not is_placeholder(block, TOC_ID) and not is_placeholder(block, TITLE_ID) then
        kept:insert(block)
      end
    end
    doc.blocks = kept
    return doc
  end

  local blocks = pandoc.Blocks({})
  local seen_chapter = false
  for i, block in ipairs(doc.blocks) do
    local anchor = chapter_anchor(block)
    if i == toc_at then
      if PANDOC_WRITER_OPTIONS.table_of_contents then
        -- --toc given as well: Pandoc's own table of contents stays at the top, no second one
      else
        local lang = doc.meta.lang and pandoc.utils.stringify(doc.meta.lang):sub(1, 2):lower() or "de"
        local toc = pandoc.structure.table_of_contents(doc.blocks, PANDOC_WRITER_OPTIONS)
        blocks:insert(pandoc.Div({
          pandoc.Header(1, TOC_TITLE[lang] or TOC_TITLE.de, pandoc.Attr("toc-title", { "unnumbered", "unlisted" })),
          toc,
        }, pandoc.Attr("TOC", { "toc" })))
      end
    elseif i == title_at then
      if doc.meta.title then
        blocks:insert(title_block(doc.meta))
      end
    elseif anchor then
      blocks:insert(pandoc.Div({}, pandoc.Attr(anchor, { "chapter-start" })))
      -- cover: the first chapter, if it holds nothing but one picture ('Cover', '[Titelinfo]', 'Deckblatt' …);
      -- shown without the caption "Cover", centred on its own page
      local nxt, after = doc.blocks[i + 1], doc.blocks[i + 2]
      local ends = after == nil or chapter_anchor(after) ~= nil or is_placeholder(after, TOC_ID)
        or is_placeholder(after, TITLE_ID)
      if not seen_chapter and nxt and (nxt.t == "Figure" or nxt.t == "Para") and ends then
        local imgs, text = {}, false
        nxt:walk({ Image = function(img) imgs[#imgs + 1] = img end })
        if nxt.t == "Para" then
          for _, inl in ipairs(nxt.content) do
            if inl.t ~= "Image" and inl.t ~= "Space" and inl.t ~= "SoftBreak" then
              text = true
            end
          end
        end
        if #imgs == 1 and not text then
          doc.blocks[i + 1] = pandoc.Div({ pandoc.Plain({ imgs[1] }) }, pandoc.Attr("", { "cover" }))
        end
      end
      seen_chapter = true
    else
      blocks:insert(block)
    end
  end
  doc.blocks = blocks
  -- The book's own title page (or the moved block) replaces Pandoc's title block at the very top.
  if doc.meta.title then
    doc.meta.pagetitle = doc.meta.pagetitle or pandoc.utils.stringify(doc.meta.title)
    doc.meta.title = nil
  end
  return doc
end