-- luatex-cn-hori-pipeline.lua -- Node pipeline for the horizontal (clreq) backend. -- -- pre_linebreak_filter (H1): walks each paragraph and inserts, at every -- CJK-involved character boundary, the penalty / adjustable glue decided by -- hori-spacing.lua; Western word spaces get tagged with the western_word -- adjustment class (clreq 挤压第 2 级 / 拉伸第 1 级). TeX's line breaker then -- handles break search globally (禁则 = penalty, 可调空间 = glue). -- -- post_linebreak_filter (H2/H3): re-distributes each line's surplus or -- shortfall by clreq priority via hori-adjust-line.lua (overriding TeX's -- proportional distribution, reclaiming the line-final punctuation blank), -- then draws the inter-line marks (专名号/书名号甲式/着重号) via -- hori-linemark.lua. -- -- Standalone: depends only on tex/shared/ and the hori/ siblings (never on -- the vertical engine's core/). local spacing = require("hori.luatex-cn-hori-spacing") local adjust_line = require("hori.luatex-cn-hori-adjust-line") local linemark = require("hori.luatex-cn-hori-linemark") local punct_table = require("shared.luatex-cn-punct-table") local punct_anchors = require("shared.luatex-cn-punct-anchors") local pipeline = {} local D = node.direct local GLYPH = node.id("glyph") local GLUE = node.id("glue") local KERN = node.id("kern") local PENALTY = node.id("penalty") local WHATSIT = node.id("whatsit") local MATH = node.id("math") -- Adjustment-class attribute for the H2 redistribution pass local ATTR_ADJUST_CLASS = luatexbase.attributes.cnhoriadjustclass or luatexbase.new_attribute("cnhoriadjustclass") pipeline.ATTR_ADJUST_CLASS = ATTR_ADJUST_CLASS -- Runtime options (set via setup) local opts = { style = "mainland", level = "basic", cjk_latin_space = true, inter_cjk_stretch = 0.05, line_adjust = true, -- H2 priority redistribution on/off line_end_punct = "compress", -- "compress" | "natural" quote_style = "keep", -- "keep" | "auto" | "curly" | "corner" hanging_punct = false, -- 行尾点号悬挂 (opt-in) avoid_orphan_char = true, -- H5 段末孤字避免 (clreq: 前行借字) last_line = "left", -- H5 段末行对齐: left|center|right|justify adjacent_punct = "1.5", -- 连续标点缩减: "1.5"|"1"|"natural" line_start_bracket = "trim", -- 行首开始夹注符号: trim|natural } -- Resolved quote conversion target ("curly"/"corner"/nil). clreq 引号体例: -- 简体横排用弯引号（先双后单），台湾用传统引号（先单后双）。默认 keep -- （不改动来稿用字，体例转换须用户显式开启）；auto 按 style 选择。 local quote_target = nil local function resolve_quote_target() local q = opts.quote_style if q == "curly" or q == "corner" then quote_target = q elseif q == "keep" then quote_target = nil else -- auto if opts.style == "taiwan" then quote_target = "corner" elseif opts.style == "mainland" then quote_target = "curly" else quote_target = nil end end end --- Configure the pipeline. -- @param o (table) { style, level, cjk_latin_space, inter_cjk_stretch, -- line_adjust, line_end_punct, quote_style, hanging_punct } function pipeline.setup(o) if not o then return end if o.style ~= nil then opts.style = o.style end if o.level ~= nil then opts.level = o.level end if o.cjk_latin_space ~= nil then opts.cjk_latin_space = o.cjk_latin_space end if o.inter_cjk_stretch ~= nil then opts.inter_cjk_stretch = o.inter_cjk_stretch end if o.line_adjust ~= nil then opts.line_adjust = o.line_adjust end if o.line_end_punct ~= nil then opts.line_end_punct = o.line_end_punct end if o.quote_style ~= nil then opts.quote_style = o.quote_style end if o.hanging_punct ~= nil then opts.hanging_punct = o.hanging_punct end if o.avoid_orphan_char ~= nil then opts.avoid_orphan_char = o.avoid_orphan_char end if o.last_line ~= nil then opts.last_line = o.last_line end if o.adjacent_punct ~= nil then opts.adjacent_punct = o.adjacent_punct end if o.line_start_bracket ~= nil then opts.line_start_bracket = o.line_start_bracket end resolve_quote_target() end -- Interword space glue subtypes (LuaTeX: spaceskip / xspaceskip; ordinary -- font spaces are emitted as spaceskip) local SPACE_SUBTYPES = { [13] = true, [14] = true } local function em_size(glyph_d) local fid = D.getfield(glyph_d, "font") local f = fid and font.getfont(fid) return (f and f.size) or 655360 -- fallback 10pt end local function make_penalty(value) local p = D.new(PENALTY) D.setfield(p, "penalty", value) return p end local function make_glue(width_sp, stretch_sp, shrink_sp, class_name) local g = D.new(GLUE) D.setfield(g, "width", width_sp) D.setfield(g, "stretch", stretch_sp) D.setfield(g, "shrink", shrink_sp) local code = spacing.ADJUST_CLASS_CODES[class_name] if code then D.set_attribute(g, ATTR_ADJUST_CLASS, code) end return g end -- Nodes transparent when walking for the previous/next glyph local TRANSPARENT = { [KERN] = true, [GLUE] = true, [PENALTY] = true, [WHATSIT] = true, } local function prev_glyph_of(n) local p = D.getprev(n) while p do local id = D.getid(p) if id == GLYPH then return p end if not TRANSPARENT[id] then return nil end p = D.getprev(p) end return nil end -- 段末孤字避免（clreq H5: 段落最后一行不宜只剩一个汉字，孤字可带 -- 后随标点）：在段末「内容字」前的断点补 penalty 10000。TeX 的全局 -- 断行随之把前一行的字借到末行（前行借字），末行至少两字。 local function protect_paragraph_end(head_d) -- 最后一个字形（跳过段尾的 glue/penalty 等） local n = D.tail(head_d) while n and D.getid(n) ~= GLYPH do if not TRANSPARENT[D.getid(n)] then return end n = D.getprev(n) end if not n then return end -- 跳过末尾标点串，得到段末内容字 local content = n while content do local c = D.getfield(content, "char") if not c or spacing.kind(c) ~= "cjk_punct" then break end content = prev_glyph_of(content) end if not content then return end local cc = D.getfield(content, "char") if not cc or spacing.kind(cc) ~= "cjk" then return end -- 内容字之前最近的断点 glue；其前已有禁断 penalty 则天然受保护 local g = D.getprev(content) while g and (D.getid(g) == KERN or D.getid(g) == WHATSIT) do g = D.getprev(g) end if not g or D.getid(g) ~= GLUE then return end local before = D.getprev(g) if before and D.getid(before) == PENALTY and D.getfield(before, "penalty") >= 10000 then return end -- 须有前字可借 if not prev_glyph_of(g) then return end D.insert_before(head_d, g, make_penalty(10000)) end --- Process one node list: insert boundary nodes between adjacent glyphs. -- Boundaries are only considered between glyphs separated by nothing or by -- font kerns; an existing glue or penalty between two glyphs means the -- document (or TeX) already decided that boundary — leave it alone. Math is -- skipped wholesale; boxes/discretionaries reset the boundary context. -- -- Insertion is done by relinking between the current node's predecessor and -- the current node; since a boundary always has a preceding glyph, the list -- head never changes. -- @param head_d (direct node) list head -- @return (direct node) list head (unchanged) -- clreq 字面分布（度量锚定，共享层 punct-anchors）：把点号/中点类的墨迹 -- 挪到本风格的规范位置——中国大陆式靠左下（GB 惯例）、台湾式居中。字体把 -- 墨迹画在哪是字体的设计惯例（TW-Kai 两向居中、思源宋体左下），排版 -- 风格不应随字体漂移：中国大陆式文档用 TW-Kai 时点号也应落到左下，台湾式 -- 文档用思源宋体时也应居中。xoffset/yoffset 只挪字面不动 advance， -- 对断行、间隙与 H2 二次分配均无影响。style=none / 无 bbox 时不动。 local function apply_ink_anchor(g) local c = D.getfield(g, "char") if not c then return end -- 先查锚点表（一次哈希查找），绝大多数字形（汉字/西文）在此返回， -- 不触碰 font.getfont if not punct_anchors.anchor(c, opts.style, "horizontal") then return end local fid = D.getfield(g, "font") local f = fid and font.getfont(fid) local desc = f and f.descriptions and f.descriptions[c] local bb = desc and desc.boundingbox local dx, dy = punct_anchors.offsets(c, opts.style, "horizontal", bb, (f and f.units_per_em) or 1000, f and f.size) if dx then D.setfield(g, "xoffset", dx) -- 绝对写入：重复处理幂等 D.setfield(g, "yoffset", dy) end end pipeline.apply_ink_anchor = apply_ink_anchor function pipeline.process(head_d) local prev_glyph = nil local prev_node = nil -- node immediately before curr in the walk local blocked = false -- an intervening glue/penalty blocks insertion local math_level = 0 local curr = head_d while curr do local id = D.getid(curr) if id == MATH then -- subtype 0 = math on, 1 = math off if D.getsubtype(curr) == 0 then math_level = math_level + 1 else math_level = math_level - 1 if math_level < 0 then math_level = 0 end end prev_glyph = nil elseif math_level > 0 then -- inside math: ignore everything elseif id == GLYPH then -- Opt-in quote style conversion (clreq 引号体例). Depth/role are -- preserved by the shared 1:1 map, so boundary decisions below -- see the converted character. if quote_target then local c0 = D.getfield(curr, "char") local conv = c0 and punct_table.quote_convert(c0, quote_target) if conv then D.setfield(curr, "char", conv) end end apply_ink_anchor(curr) if prev_glyph and not blocked then local a = D.getfield(prev_glyph, "char") local c = D.getfield(curr, "char") if a and c then local action = spacing.boundary(a, c, opts) if action and action.glue then local em = em_size(prev_glyph) local g = make_glue( math.floor(action.glue.width * em + 0.5), math.floor(action.glue.stretch * em + 0.5), math.floor(action.glue.shrink * em + 0.5), action.glue.class) -- prev_node → [penalty] → glue → curr if action.penalty then local p = make_penalty(action.penalty) D.setlink(prev_node, p) D.setlink(p, g) else D.setlink(prev_node, g) end D.setlink(g, curr) end end end prev_glyph = curr blocked = false elseif id == KERN or id == WHATSIT then -- transparent for boundary purposes (font kerns, marks) elseif id == GLUE or id == PENALTY then -- The boundary already has spacing/break semantics; skip it. -- Word spaces (from source blanks) get the western_word class so -- the H2 pass manages them at clreq 挤压第 2 级 / 拉伸第 1 级. -- clreq 数值界限按汉字宽夹紧：西文词距最小可挤到 1/4 汉字宽、 -- 最大可拉到半个汉字宽（字体自身弹性越界时收窄）。 if id == GLUE and SPACE_SUBTYPES[D.getsubtype(curr)] and D.getfield(curr, "width") > 0 and not D.get_attribute(curr, ATTR_ADJUST_CLASS) then D.set_attribute(curr, ATTR_ADJUST_CLASS, spacing.ADJUST_CLASS_CODES.western_word) if prev_glyph then local em = em_size(prev_glyph) local w = D.getfield(curr, "width") local floor_w = math.min(w, math.floor(0.25 * em)) if w - D.getfield(curr, "shrink") < floor_w then D.setfield(curr, "shrink", w - floor_w) end local ceil_w = math.max(w, math.floor(0.5 * em)) if w + D.getfield(curr, "stretch") > ceil_w then D.setfield(curr, "stretch", ceil_w - w) end end end blocked = true else -- boxes, discretionaries, rules, dirs, ...: reset context prev_glyph = nil blocked = false end prev_node = curr curr = D.getnext(curr) end return head_d end -- ============================================================================ -- Callback registration -- ============================================================================ local CALLBACK_NAME = "luatexcn.hori.pre_linebreak" local POST_CALLBACK_NAME = "luatexcn.hori.post_linebreak" local enabled = false local function pre_linebreak(head, groupcode) local d = D.todirect(head) d = pipeline.process(d) if opts.avoid_orphan_char then protect_paragraph_end(d) end return D.tonode(d) end -- Exposed for unit tests pipeline.protect_paragraph_end = protect_paragraph_end local function post_linebreak(head, groupcode) local d = D.todirect(head) if opts.line_adjust then d = adjust_line.process_lines(d, ATTR_ADJUST_CLASS, opts) end d = linemark.decorate(d) return D.tonode(d) end --- Register the pre/post_linebreak_filter pair (idempotent). function pipeline.enable() if enabled then return end luatexbase.add_to_callback("pre_linebreak_filter", pre_linebreak, CALLBACK_NAME) luatexbase.add_to_callback("post_linebreak_filter", post_linebreak, POST_CALLBACK_NAME) enabled = true end --- Remove the callbacks (for tests / package unloading). function pipeline.disable() if not enabled then return end luatexbase.remove_from_callback("pre_linebreak_filter", CALLBACK_NAME) luatexbase.remove_from_callback("post_linebreak_filter", POST_CALLBACK_NAME) enabled = false end return pipeline