rexml Encoding logic creates unnecessary singleton classes, aliases at runtime
The logic for switching a given parser to a new encoding currently does the following:
- Reopens the Encoding module (only on first use of the encoding)
- Defines two new methods for the encoding being applied (only on first use)
- Singletonizes the parser object and aliases the encoding-specific methods to "encode" and "decode"
This last phase is unnecessary, but happens at runtime whenever an encoding is applied to a parser. The resulting singletonizing and method aliasing causes cache flushes.
I have created a patch that does not use this singleton logic and which will not cause any singletons or method table updates at steady-state.
I know 1.8.7 is only in security patch mode, but I thought I'd throw this out there. JRuby will probably apply it to our fork of 1.8.7 stdlib.