Prevent data corruption when decoding gzip stream, fixes #284

This fixes an issue where a gzip stream chunked into smaller pieces
would result in a buffer overrun when decoding the headers.

Previously the position where the headers end in `compressed` was
determined by the delta of `@pos` before and after reading the
headers. This didn't take into account the fact that `@data` may already
contain data, but not enough to have advanced `@pos` on previous
iterations.

Author: boxofrad

lib/em-http/decoders.rb

@@ -127,7 +127,6 @@ def eof?
 
     def extract_stream(compressed)
       @data << compressed
-      pos = @pos
 
       while !eof? && !finished?
         buffer = ""
@@ -208,7 +207,7 @@ def extract_stream(compressed)
       end
 
       if finished?
-        compressed[(@pos - pos)..-1]
+        compressed[(@pos - (@data.length - compressed.length))..-1]
       else
         ""
       end

spec/gzip_spec.rb

@@ -56,6 +56,29 @@
     decompressed.size.should eq(32907)
   end
 
+  it "should not care how many chunks the file is split up into" do
+    examples = [
+      ["\x1F", "\x8B", "\b", "\x00", "\x00", "\x00", "\x00", "\x00", "\x00", "\x00", "\xF3", "\xCB", "/", "Q", "p\xCB/\xCDK\x01\x00M\x8Ct\xB1\t\x00\x00\x00"],
+      ["\x1F", "\x8B", "\b", "\x00", "\x00", "\x00", "\x00", "\x00\x00\x00\xF3\xCB/Qp\xCB/\xCDK\x01\x00M\x8Ct\xB1\t\x00\x00\x00"]
+    ]
+
+    examples.each do |example|
+      decompressed = ""
+
+      gz = EventMachine::HttpDecoders::GZip.new do |data|
+        decompressed << data
+      end
+
+      example.each do |bytes|
+        gz << bytes
+      end
+
+      gz.finalize!
+
+      decompressed.should eq("Not Found")
+    end
+  end
+
   it "should fail with a DecoderError if not a gzip file" do
     not_a_gzip = ["1f8c08089668a650000"].pack("H*")
     header = EventMachine::HttpDecoders::GZipHeader.new