In general, I find that it's a bad idea to destroy information. Keep all input intact as long as possible, so if you ever do encounter a problem, you don't need to reconstruct the original data. You can always process data into the correct form at runtime.
Of course, there are performance implications to this, but those can be dealt with. Encoding strings right away for performance reasons is definitely a premature optimization.
Of course, there are performance implications to this, but those can be dealt with. Encoding strings right away for performance reasons is definitely a premature optimization.