fontisan 0.4.47 → 0.4.48

data/lib/fontisan/reference/table_digests.rb

@@ -0,0 +1,251 @@
+# frozen_string_literal: true
+
+require "digest"
+
+module Fontisan
+  module Reference
+    # Canonical digests of parsed table content, computed from the raw table
+    # bytes with the OpenType layouts written out here (not through the
+    # table classes), so another implementation can reproduce them from the
+    # same bytes. Every function returns JSON-ready data with String keys;
+    # Fixed 16.16 values are raw integers. A structure that runs past the
+    # end of its table raises {Truncated}.
+    #
+    # See docs/reference-facade.md ("Parsed table digests").
+    module TableDigests
+      class Truncated < StandardError; end
+
+      module_function
+
+      # SHA-256 over the concatenation of the 32-byte digests of each part.
+      def digest_of_parts(parts)
+        Digest::SHA256.hexdigest(parts.map { |p| Digest::SHA256.digest(p) }.join)
+      end
+
+      def sha(bytes)
+        Digest::SHA256.hexdigest(bytes)
+      end
+
+      def slice(data, offset, length)
+        raise Truncated, "#{length} bytes at #{offset} past #{data.bytesize}" if
+          offset.negative? || length.negative? || offset + length > data.bytesize
+
+        data.byteslice(offset, length)
+      end
+
+      def u8(data, off) = slice(data, off, 1).unpack1("C")
+      def u16(data, off) = slice(data, off, 2).unpack1("n")
+      def u32(data, off) = slice(data, off, 4).unpack1("N")
+      def i32(data, off) = slice(data, off, 4).unpack1("l>")
+
+      # gvar: header fields, the shared tuples, and one digest over every
+      # glyph's GlyphVariationData byte range (located by the offset array).
+      def gvar(data)
+        axis_count = u16(data, 4)
+        shared_count = u16(data, 6)
+        shared_offset = u32(data, 8)
+        glyph_count = u16(data, 12)
+        flags = u16(data, 14)
+        array_offset = u32(data, 16)
+        long = flags.anybits?(1)
+        step = long ? 4 : 2
+        offsets = Array.new(glyph_count + 1) do |i|
+          value = long ? u32(data, 20 + (i * step)) : u16(data, 20 + (i * step)) * 2
+          array_offset + value
+        end
+        parts = offsets.each_cons(2).map { |a, b| slice(data, a, b - a) }
+        {
+          "axis_count" => axis_count,
+          "flags" => flags,
+          "glyph_count" => glyph_count,
+          "glyphs_with_data" => parts.count { |p| !p.empty? },
+          "glyphs_digest" => digest_of_parts(parts),
+          "shared_tuple_count" => shared_count,
+          "shared_tuples_sha256" => sha(slice(data, shared_offset, shared_count * axis_count * 2)),
+          "version" => [u16(data, 0), u16(data, 2)],
+        }
+      end
+
+      # HVAR / VVAR share a layout: header, ItemVariationStore, and up to
+      # three DeltaSetIndexMaps.
+      def hvar(data)
+        maps = { "advance_mapping" => 8, "lsb_mapping" => 12, "rsb_mapping" => 16 }
+        result = { "version" => [u16(data, 0), u16(data, 2)] }
+        result["item_variation_store"] = item_variation_store(data, u32(data, 4))
+        maps.each do |name, at|
+          offset = u32(data, at)
+          result[name] = offset.zero? ? nil : delta_set_index_map(data, offset)
+        end
+        result
+      end
+
+      # MVAR: value records (tag, outer, inner) and the ItemVariationStore
+      # (Offset16 from the table start).
+      def mvar(data)
+        record_size = u16(data, 6)
+        count = u16(data, 8)
+        store_offset = u16(data, 10)
+        {
+          "version" => [u16(data, 0), u16(data, 2)],
+          "value_records" => Array.new(count) do |i|
+            at = 12 + (i * record_size)
+            { "tag" => slice(data, at, 4), "outer" => u16(data, at + 4), "inner" => u16(data, at + 6) }
+          end,
+          "item_variation_store" => store_offset.zero? ? nil : item_variation_store(data, store_offset),
+        }
+      end
+
+      # Exact byte ranges: header+offsets, region list, each
+      # ItemVariationData. Digests are per range, then one over all.
+      def item_variation_store(data, base)
+        region_offset = u32(data, base + 2)
+        data_count = u16(data, base + 6)
+        region_axes = u16(data, base + region_offset)
+        region_count = u16(data, base + region_offset + 2)
+        region_len = 4 + (region_count * region_axes * 6)
+        subtables = Array.new(data_count) do |i|
+          start = base + u32(data, base + 8 + (i * 4))
+          items = u16(data, start)
+          words = u16(data, start + 2)
+          indexes = u16(data, start + 4)
+          long = words.anybits?(0x8000)
+          word_count = words & 0x7FFF
+          row = long ? (word_count * 4) + ((indexes - word_count) * 2) : word_count + indexes
+          slice(data, start, 6 + (indexes * 2) + (items * row))
+        end
+        {
+          "region_count" => region_count,
+          "item_data_count" => data_count,
+          "regions_sha256" => sha(slice(data, base + region_offset, region_len)),
+          "data_digest" => digest_of_parts(subtables),
+          "data_sha256" => subtables.map { |s| sha(s) },
+        }
+      end
+
+      # DeltaSetIndexMap: format 0 (u16 count) or 1 (u32 count).
+      def delta_set_index_map(data, offset)
+        format = u8(data, offset)
+        entry_format = u8(data, offset + 1)
+        count = format.zero? ? u16(data, offset + 2) : u32(data, offset + 2)
+        header = format.zero? ? 4 : 6
+        size = ((entry_format >> 4) & 3) + 1
+        {
+          "format" => format,
+          "entry_format" => entry_format,
+          "map_count" => count,
+          "sha256" => sha(slice(data, offset, header + (count * size))),
+        }
+      end
+
+      # STAT: design axes and every axis value record, Fixed as raw ints.
+      def stat(data)
+        axis_size = u16(data, 4)
+        axis_count = u16(data, 6)
+        axes_offset = u32(data, 8)
+        value_count = u16(data, 12)
+        values_offset = u32(data, 14)
+        minor = u16(data, 2)
+        {
+          "version" => [u16(data, 0), minor],
+          "elided_fallback_name_id" => minor >= 1 ? u16(data, 18) : nil,
+          "axes" => Array.new(axis_count) do |i|
+            at = axes_offset + (i * axis_size)
+            { "tag" => slice(data, at, 4), "name_id" => u16(data, at + 4), "ordering" => u16(data, at + 6) }
+          end,
+          "axis_values" => Array.new(value_count) do |i|
+            axis_value(data, values_offset + u16(data, values_offset + (i * 2)))
+          end,
+        }
+      end
+
+      def axis_value(data, pos)
+        format = u16(data, pos)
+        case format
+        when 1
+          fixed_value(data, pos, format, 8, %w[value])
+        when 2
+          fixed_value(data, pos, format, 8, %w[nominal range_min range_max])
+        when 3
+          fixed_value(data, pos, format, 8, %w[value linked_value])
+        when 4
+          n = u16(data, pos + 2)
+          { "format" => 4, "flags" => u16(data, pos + 4), "value_name_id" => u16(data, pos + 6),
+            "axis_values" => Array.new(n) do |j|
+              { "axis_index" => u16(data, pos + 8 + (j * 6)), "value" => i32(data, pos + 10 + (j * 6)) }
+            end }
+        else
+          { "format" => format, "unsupported" => true }
+        end
+      end
+
+      def fixed_value(data, pos, format, first, names)
+        out = { "format" => format, "axis_index" => u16(data, pos + 2), "flags" => u16(data, pos + 4),
+                "value_name_id" => u16(data, pos + 6) }
+        names.each_with_index { |n, i| out["#{n}_raw"] = i32(data, pos + first + (i * 4)) }
+        out
+      end
+
+      # CFF2: CharStrings INDEX located through Top DICT operator 17.
+      # Returns the per-glyph charstring byte slices.
+      def cff2_charstrings(data)
+        header_size = u8(data, 2)
+        top_length = u16(data, 3)
+        offset = dict_operand(slice(data, header_size, top_length), 17)
+        return [] unless offset
+
+        count = u32(data, offset)
+        return [] if count.zero?
+
+        off_size = u8(data, offset + 4)
+        # The offset array must fit in the table; a fuzzed count of 2^32
+        # would otherwise build billions of slices (same bound as the Rust
+        # engine's INDEX reader).
+        raise Truncated, "invalid CFF2 offSize #{off_size}" unless (1..4).cover?(off_size)
+
+        slice(data, offset + 5, (count + 1) * off_size)
+        read = ->(i) { slice(data, offset + 5 + (i * off_size), off_size).bytes.inject(0) { |a, b| (a << 8) | b } }
+        base = offset + 5 + ((count + 1) * off_size) - 1
+        Array.new(count) { |i| slice(data, base + read.call(i), read.call(i + 1) - read.call(i)) }
+      end
+
+      # The last operand before `operator` in a DICT, or nil. Only the
+      # operand encodings a CharStrings offset can use are decoded; real
+      # numbers are skipped (pushed as 0).
+      def dict_operand(dict, operator)
+        stack = []
+        i = 0
+        while i < dict.bytesize
+          lead = dict.getbyte(i)
+          if lead <= 21
+            op = lead == 12 ? (i += 1) && (1200 + dict.getbyte(i)) : lead
+            return stack.last if op == operator
+
+            stack.clear
+          else
+            value, i = dict_number(dict, i, lead)
+            stack << value
+          end
+          i += 1
+        end
+        nil
+      end
+
+      # Decode the operand starting at `i`; returns [value, index of its
+      # last byte].
+      def dict_number(dict, idx, lead)
+        case lead
+        when 28 then [dict.byteslice(idx + 1, 2).unpack1("s>"), idx + 2]
+        when 29 then [dict.byteslice(idx + 1, 4).unpack1("l>"), idx + 4]
+        when 30
+          idx += 1
+          idx += 1 until idx >= dict.bytesize || dict.getbyte(idx) >> 4 == 0xF || dict.getbyte(idx) & 0xF == 0xF
+          [0, idx]
+        when 32..246 then [lead - 139, idx]
+        when 247..250 then [((lead - 247) * 256) + dict.getbyte(idx + 1) + 108, idx + 1]
+        when 251..254 then [-((lead - 251) * 256) - dict.getbyte(idx + 1) - 108, idx + 1]
+        else [0, idx]
+        end
+      end
+    end
+  end
+end