require_relative "gbconvert" require "gb_agencies" require_relative "gbcleanup" require_relative "metadata" require "fileutils" module IsoDoc module Gb module BaseConvert def extract_fonts(options) b = options[:bodyfont] || "Arial" h = options[:headerfont] || "Arial" m = options[:monospacefont] || "Courier" t = options[:titlefont] || "Arial" "$bodyfont: #{b};\n$headerfont: #{h};\n$monospacefont: #{m};\n$titlefont: #{t};\n" end def metadata_init(lang, script, labels) unless ["en", "zh"].include? lang lang = "zh" script = "Hans" end @meta = Metadata.new(lang, script, labels) @meta.set(:standardclassimg, @standardclassimg) @common.meta = @meta end def cleanup(docxml) @cleanup = Cleanup.new(@script, @deprecated_lbl) super @cleanup.cleanup(docxml) docxml end def example_cleanup(docxml) super @cleanup.example_cleanup(docxml) end def i18n_init(lang, script) super y = if lang == "en" YAML.load_file(File.join(File.dirname(__FILE__), "i18n-en.yaml")) elsif lang == "zh" && script == "Hans" YAML.load_file(File.join(File.dirname(__FILE__), "i18n-zh-Hans.yaml")) else YAML.load_file(File.join(File.dirname(__FILE__), "i18n-zh-Hans.yaml")) end @labels = @labels.merge(y) end def omit_docid_prefix(prefix) super || prefix == "Chinese Standard" end def formula_parse(node, out) out.div **attr_code(id: node["id"], class: "formula") do |div| insert_tab(div, 1) parse(node.at(ns("./stem")), out) lbl = anchor(node['id'], :label, false) unless lbl.nil? insert_tab(div, 1) div << "(#{lbl})" end end formula_where(node.at(ns("./dl")), out) end def formula_where(dl, out) return unless dl out.p { |p| p << @where_lbl } formula_dl_parse(dl, out) end def formula_dl_parse(node, out) out.table **{ class: "dl" } do |v| node.elements.each_slice(2) do |dt, dd| v.tr do |tr| tr.td **{ style: "vertical-align:top;text-align:left;" } do |term| dt_parse(dt, term) end tr.td(**{ style: "vertical-align:top;" }) { |td| td << "—" } tr.td **{ style: "vertical-align:top;" } do |listitem| dd.children.each { |n| parse(n, listitem) } end end end end end EXAMPLE_TBL_ATTR = { class: "example_label", style: "padding:2pt 2pt 2pt 2pt;vertical-align:top;" }.freeze def example_label(node) l10n(super + ":") end def note_parse(node, out) @note = true out.table **attr_code(id: node["id"], class: "Note") do |t| t.tr do |tr| @libdir = File.dirname(__FILE__) tr.td **EXAMPLE_TBL_ATTR do |td| td << l10n(note_label(node) + ":") end tr.td **{ style: "vertical-align:top;", class: "Note" } do |td| node.children.each { |n| parse(n, td) } end end end @note = false end def termnote_parse(node, out) @note = true out.table **attr_code(id: node["id"], class: "Note") do |t| t.tr do |tr| tr.td **EXAMPLE_TBL_ATTR do |td| td << l10n("#{anchor(node['id'], :label)}:") end tr.td **{ style: "vertical-align:top;", class: "Note" } do |td| node.children.each { |n| parse(n, td) } end end end @note = false end def middle(isoxml, out) super end_line(isoxml, out) end def end_line(_isoxml, out) out.hr **{ width: "25%" } end def error_parse(node, out) # catch elements not defined in ISO case node.name when "string" then string_parse(node, out) else super end end def string_parse(node, out) if node["script"] == "Hant" out.span **{ class: "Hant" } do |s| node.children.each { |c| parse(c, s) } end else node.children.each { |c| parse(c, out) } end end def deprecated_term_parse(node, out) out.p **{ class: "DeprecatedTerms" } do |p| p << l10n("#{@deprecated_lbl}: ") node.children.each { |c| parse(c, p) } end end def termref_render(x) x.sub!(%r{\s*\[MODIFICATION\]\s*$}m, l10n(", #{@modified_lbl}")) parts = x.split(%r{(\s*\[MODIFICATION\]|,)}m) parts[1] = l10n(", #{@source_lbl}") if parts.size > 1 && parts[1] == "," && !/^\s*#{@modified_lbl}/.match(parts[2]) parts.map do |p| /\s*\[MODIFICATION\]/.match(p) ? l10n(", #{@modified_lbl} — ") : p end.join.sub(/\A\s*/m, l10n("[")).sub(/\s*\z/m, l10n("]")) end def termref_resolve(docxml) docxml.split(%r{(\[TERMREF\]|\[/TERMREF\])}).each_slice(4). map do |a| a.size < 3 ? a[0] : a[0] + termref_render(a[2]) end.join end def foreword(isoxml, out) f = isoxml.at(ns("//foreword")) || return page_break(out) out.div do |s| s.h1 **{ class: "ForewordTitle" } do |h1| h1 << "#{@foreword_lbl} " # insert_tab(h1, 1) end f.elements.each { |e| parse(e, s) unless e.name == "title" } end end def clausedelimspace(out) out << " " end def clause_name(num, title, div, header_class) header_class = {} if header_class.nil? div.h1 **attr_code(header_class) do |h1| if num && !@suppressheadingnumbers h1 << "#{num}." h1 << " " end title.is_a?(String) ? h1 << title : title&.children&.each { |c2| parse(c2, h1) } end div.parent.at(".//h1") end def example_span_label(node, div, name) n = get_anchors[node["id"]] div.span **{ class: "example_label" } do |p| lbl = (n.nil? || n[:label].nil? || n[:label].empty?) ? @example_lbl : l10n("#{@example_lbl} #{n[:label]}") p << l10n(lbl + ":") name and !lbl.nil? and p << " — " name and name.children.each { |n| parse(n, div) } end end def example_p_parse(node, div) name = node&.at(ns("./name"))&.remove div.p do |p| example_span_label(node, p, name) insert_tab(p, 1) node.first_element_child.children.each { |n| parse(n, p) } end node.element_children[1..-1].each { |n| parse(n, div) } end def example_parse1(node, div) div.p do |p| example_span_label(node, p, node.at(ns("./name"))) insert_tab(p, 1) end node.children.each { |n| parse(n, div) unless n.name == "name" } end def node_begins_with_para(node) node.elements.each do |e| next if e.name == "name" return true if e.name == "p" return false end false end def example_parse(node, out) out.div **{ id: node["id"], class: "example" } do |div| if node_begins_with_para(node) example_p_parse(node, div) else example_parse1(node, div) end end end def textcleanup(docxml) termref_resolve(docxml) end end end end