Correctly render glyphs with x or y offsets

These offsets are generated by fontkit when laying out glyphs, for example accents which should appear above other glyphs. To render them correctly in PDF, we have to break out of the TJ command and move the text position manually.

lib/mixins/text.coffee

@@ -222,7 +222,7 @@ module.exports =
     @addContent "BT"
 
     # text position
-    @addContent "#{x} #{y} Td"
+    @addContent "1 0 0 1 #{x} #{y} Tm"
 
     # font and font size
     @addContent "/#{@_font.id} #{@_fontSize} Tf"
@@ -244,39 +244,66 @@ module.exports =
       wordSpacing *= 1000 / @_fontSize
       
       encoded = []
-      advances = []
+      positions = []
       for word in words
-        [encodedWord, advancesWord] = @_font.encode(word, options.features)
-        unless advancesWord
-          advancesWord = (0 for i in [0...encodedWord.length])
-          
+        [encodedWord, positionsWord] = @_font.encode(word, options.features)          
         encoded.push encodedWord...
-        advances.push advancesWord...
+        positions.push positionsWord...
         
-        # add thw word spacing to the end of the word
-        advances[advances.length - 1] += wordSpacing
+        # add the word spacing to the end of the word
+        positions[positions.length - 1].xAdvance += wordSpacing
     else
-      [encoded, advances] = @_font.encode(text, options.features)
+      [encoded, positions] = @_font.encode(text, options.features)
       
-    # if we have an array of advances (e.g kerning data),
-    # use the TJ operator to control spacing
-    if advances
-      commands = []
-      last = 0
-      for a, i in advances
-        # group consecutive letters together whose advance adjustments are 0
-        if a is 0 and i < advances.length - 1
-          continue
+    scale = @_fontSize / 1000
+    commands = []
+    last = 0
+    hadOffset = no
+    
+    # Adds a segment of text to the TJ command buffer
+    addSegment = (cur) =>
+      if last < cur
+        hex = encoded.slice(last, cur).join ''
+        advance = positions[cur - 1].xAdvance - positions[cur - 1].advanceWidth
+        commands.push "<#{hex}> #{-advance}"
+      
+      last = cur
+    
+    # Flushes the current TJ commands to the output stream
+    flush = (i) =>
+      addSegment i
+      
+      if commands.length > 0
+        @addContent "[#{commands.join ' '}] TJ"
+        commands.length = 0
+    
+    for pos, i in positions
+      # If we have an x or y offset, we have to break out of the current TJ command
+      # so we can move the text position.
+      if pos.xOffset or pos.yOffset
+        # Flush the current buffer
+        flush i
         
-        h = encoded.slice(last, i + 1).join ''
-        commands.push "<#{h}> #{-a}"
-        last = i + 1
+        # Move the text position and flush just the current character
+        @addContent "1 0 0 1 #{x + pos.xOffset * scale} #{y + pos.yOffset * scale} Tm"
+        flush i + 1
+        
+        hadOffset = yes
+      else
+        # If the last character had an offset, reset the text position
+        if hadOffset
+          @addContent "1 0 0 1 #{x} #{y} Tm"
+          hadOffset = no
+        
+        # Group segments that don't have any advance adjustments
+        unless pos.xAdvance - pos.advanceWidth is 0
+          addSegment i + 1
+      
+      x += pos.xAdvance * scale
+    
+    # Flush any remaining commands
+    flush i
       
-      @addContent "[#{commands.join ' '}] TJ"
-    else
-      # otherwise, just use the normal Tj operator
-      @addContent "<#{encoded.join ''}> Tj"
-
     # end the text object
     @addContent "ET"