Fix show_backtraces: false leaking Exception#cause

This code was likely fine back when it was written against Ruby 1.9.3,
but since then `Kernel#raise` now automatically attach the currently rescued error
as the new exception `cause`.

Since this code was re-raising the same instance over and once a `cause` was
attached it would never be de-associated, which in some contrived scenario could
leak to information leak across requests / test etc.

I ran a benchmark and the fastest way to raise exception I could find is:

```ruby
raise ErrorClass, "message".freeze, EMPTY_ARRAY, cause: nil
```

lib/memcached/exceptions.rb

@@ -45,21 +45,7 @@ class Memcached
 * Memcached::WriteFailure
 
 =end
-  class Error < RuntimeError
-    attr_accessor :no_backtrace
-
-    def initialize(*args)
-      super(*args)
-    end
-
-    def set_backtrace(*args)
-      no_backtrace ? [] : super
-    end
-
-    def backtrace(*args)
-      no_backtrace ? [] : super
-    end
-  end
+  Error = Class.new(RuntimeError)
 
 #:stopdoc:
 

lib/memcached/memcached.rb

@@ -3,6 +3,7 @@
 =end
 class Memcached
   FLAGS = 0x0
+  EMPTY_ARRAY = [].freeze
 
   DEFAULTS = {
     :hash => :fnv1_32,
@@ -174,17 +175,7 @@ def initialize(servers = nil, opts = {})
     set_servers(servers)
 
     # Not found exceptions
-    if options[:show_backtraces]
-      # Raise classes for full context
-      @not_found = NotFound
-      @not_stored = NotStored
-    else
-      # Raise fast context-free exception instances
-      @not_found = NotFound.new
-      @not_found.no_backtrace = true
-      @not_stored = NotStored.new
-      @not_stored.no_backtrace = true
-    end
+    @show_backtraces = options[:show_backtraces] ? nil : EMPTY_ARRAY
   end
 
   # Set the server list.
@@ -597,9 +588,9 @@ def check_return_code(ret, key = nil) #:doc:
     if ret == 0 # Lib::MEMCACHED_SUCCESS
     elsif ret == 32 # Lib::MEMCACHED_BUFFERED
     elsif ret == 16
-      raise @not_found # Lib::MEMCACHED_NOTFOUND
+      raise NotFound, "NOTFOUND".freeze, @show_backtraces, cause: nil # Lib::MEMCACHED_NOTFOUND
     elsif ret == 14
-      raise @not_stored # Lib::MEMCACHED_NOTSTORED
+      raise NotStored, "NOTSTORED".freeze, @show_backtraces, cause: nil # Lib::MEMCACHED_NOTSTORED
     else
       reraise(key, ret)
     end

test/unit/memcached_test.rb

@@ -192,6 +192,33 @@ def test_initialize_without_backtraces
     end
   end
 
+  def test_without_backtrace_no_cause
+    cache = Memcached.new @servers, show_backtraces: false
+
+    # show_backtraces: false was implemented by always re-raising the same instance.
+    # This was likely fine back in Ruby 1.9.3. But since then now Ruby automatically
+    # attach a `cause` if you raise from a `rescue` or `ensure` and it's never cleared
+    # from the instance causing information to leak.
+
+    rescued = false
+
+    begin
+      raise "foo"
+    rescue RuntimeError => error
+      begin
+        cache.get key
+      rescue Memcached::NotFound
+        rescued = true
+      end
+    end
+    assert rescued
+
+    error = assert_raises Memcached::NotFound do
+      cache.get key
+    end
+    assert_nil error.cause
+  end
+
   def test_initialize_with_backtraces
     cache = Memcached.new @servers, :show_backtraces => true
     cache.delete key rescue nil