lib/hexapdf/parser.rb in hexapdf-0.12.3 vs lib/hexapdf/parser.rb in hexapdf-0.13.0

@@ -57,10 +57,11 @@
     def initialize(io, document)
       @io = io
       @tokenizer = Tokenizer.new(io)
       @document = document
       @object_stream_data = {}
+      @reconstructed_revision = nil
       retrieve_pdf_header_offset_and_version
     end
 
     # Loads the indirect (potentially compressed) object specified by the given cross-reference
     # entry.
@@ -84,10 +85,12 @@
         raise_malformed("The oid,gen (#{oid},#{gen}) values of the indirect object don't match " \
           "the values (#{xref_entry.oid},#{xref_entry.gen}) from the xref")
       end
 
       @document.wrap(obj, oid: oid, gen: gen, stream: stream)
+    rescue HexaPDF::MalformedPDFError
+      reconstructed_revision.object(xref_entry)
     end
 
     # Parses the indirect object at the specified offset.
     #
     # This method is used by a PDF Document to load objects. It should **not** be used by any
@@ -233,18 +236,18 @@
         end
 
         @tokenizer.skip_whitespace
         start.upto(start + number_of_entries - 1) do |oid|
           pos, gen, type = @tokenizer.next_xref_entry do |matched_size|
-            maybe_raise("Invalid cross-reference subsection entry", pos: @tokenizer.pos,
-                        force: matched_size == 20)
+            maybe_raise("Invalid cross-reference entry", pos: @tokenizer.pos,
+                        force: !matched_size)
           end
           if xref.entry?(oid)
             next
           elsif type == 'n'
             if pos == 0 || gen > 65535
-              maybe_raise("Invalid in use cross-reference entry in cross-reference section",
+              maybe_raise("Invalid in use cross-reference entry",
                           pos: @tokenizer.pos)
               xref.add_free_entry(oid, gen)
             else
               xref.add_in_use_entry(oid, gen, pos)
             end
@@ -311,10 +314,15 @@
       end
 
       @startxref_offset = lines[eof_index - 1].to_i
     end
 
+    # Returns the reconstructed revision.
+    def reconstructed_revision
+      @reconstructed_revision ||= reconstruct_revision
+    end
+
     # Returns the PDF version number that is stored in the file header.
     #
     # See: PDF1.7 s7.5.2
     def file_header_version
       unless @header_version
@@ -334,9 +342,63 @@
     # See: PDF1.7 s7.5.2, ADB1.7 sH.3-3.4.1
     def retrieve_pdf_header_offset_and_version
       @io.seek(0)
       @header_offset = (@io.read(1024) || '').index(/%PDF-(\d\.\d)/) || 0
       @header_version = $1
+    end
+
+    # Tries to reconstruct the PDF document's main cross-reference table by serially parsing the
+    # file and returning a Revision object for loading the found objects.
+    #
+    # If the file contains multiple cross-reference sections, all objects will be put into a single
+    # cross-reference table, later objects overwriting prior ones.
+    def reconstruct_revision
+      raise unless @document.config['parser.try_xref_reconstruction']
+      msg = "#{$!} - trying cross-reference table reconstruction"
+      @document.config['parser.on_correctable_error'].call(@document, msg, @tokenizer.pos)
+
+      xref = XRefSection.new
+      @tokenizer.pos = 0
+      while true
+        pos = @tokenizer.pos
+        @tokenizer.scan_until(/(\n|\r\n?)+|\z/)
+        next_new_line_pos = @tokenizer.pos
+        @tokenizer.pos = pos
+
+        token = @tokenizer.next_token rescue nil
+        if token.kind_of?(Integer)
+          gen = @tokenizer.next_token rescue nil
+          tok = @tokenizer.next_token rescue nil
+          if @tokenizer.pos > next_new_line_pos
+            @tokenizer.pos = next_new_line_pos
+          elsif gen.kind_of?(Integer) && tok.kind_of?(Tokenizer::Token) && tok == 'obj'
+            xref.add_in_use_entry(token, gen, pos)
+            @tokenizer.scan_until(/(?:\n|\r\n?)endobj\b/)
+          end
+        elsif token.kind_of?(Tokenizer::Token) && token == 'trailer'
+          obj = @tokenizer.next_object rescue nil
+          # Use last trailer found in case of multiple revisions but use first trailer in case of
+          # linearized file.
+          trailer = obj if obj.kind_of?(Hash) && (obj.key?(:Prev) || trailer.nil?)
+        elsif token == Tokenizer::NO_MORE_TOKENS
+          break
+        else
+          @tokenizer.pos = next_new_line_pos
+        end
+      end
+
+      trailer&.delete(:Prev) # no need for this and may wreak havoc
+      if !trailer || trailer.empty?
+        raise_malformed("Could not reconstruct malformed PDF because trailer was not found", pos: 0)
+      end
+
+      loader = lambda do |xref_entry|
+        obj, oid, gen, stream = parse_indirect_object(xref_entry.pos)
+        @document.wrap(obj, oid: oid, gen: gen, stream: stream)
+      end
+
+      Revision.new(@document.wrap(trailer, type: :XXTrailer), xref_section: xref,
+                   loader: loader)
     end
 
     # Raises a HexaPDF::MalformedPDFError with the given message and source position.
     def raise_malformed(msg, pos: nil)
       raise HexaPDF::MalformedPDFError.new(msg, pos: pos)