IronRuby performance issue while using Variables

Question

Here is code of very simple expression evaluator using IronRuby

public class BasicRubyExpressionEvaluator
{
 ScriptEngine engine;
 ScriptScope scope;
 public Exception LastException
 {
  get; set;
 }
 private static readonly Dictionary<string, ScriptSource> parserCache = new Dictionary<string, ScriptSource>();
 public BasicRubyExpressionEvaluator()
 {
  engine = Ruby.CreateEngine();
  scope = engine.CreateScope();

 }

 public object Evaluate(string expression, DataRow context)
 {
  ScriptSource source;
  parserCache.TryGetValue(expression, out source);
  if (source == null)
  {
   source = engine.CreateScriptSourceFromString(expression, SourceCodeKind.SingleStatement);
   parserCache.Add(expression, source);
  }

  var result = source.Execute(scope);
  return result;
 }
    public void SetVariable(string variableName, object value)
 {
  scope.SetVariable(variableName, value);
 }
}

and here is problem.

var evaluator = new BasicRubyExpressionEvaluator();
evaluator.SetVariable("a", 10);
evaluator.SetVariable("b", 1 );
evaluator.Evaluate("a+b+2", null);

vs

var evaluator = new BasicRubyExpressionEvaluator();
evaluator.Evaluate("10+1+2", null);

First Is 25 times slower than second. Any suggestions? String.Replace is not a solution for me.

Answer 1

you can of course first build the string something like

evaluator.Evaluate(string.format("a={0}; b={1}; a + b + 2", 10, 1))

Or you can make it a method

if instead of your script you return a method then you should be able to use it like a regular C# Func object.

var script = @"

def self.addition(a, b)
  a + b + 2
end
"

engine.ExecuteScript(script);
var = func = scope.GetVariable<Func<object,object,object>>("addition");    
func(10,1)

This is probably not a working snippet but it shows the general idea.

Answer 2

I do not think the performance you are seeing is due to variable setting; the first execution of IronRuby in a program is always going to be slower than the second, regardless of what you're doing, since most of the compiler isn't loaded in until code is actually run (for startup performance reasons). Please try that example again, maybe running each version of your code in a loop, and you'll see the performance is roughly equivalent; the variable-version does have some overhead of method-dispatch to get the variables, but that should be negligible if you run it enough.

Also, in your hosting code, how come you are holding onto ScriptScopes in a dictionary? I would hold onto CompiledCode (result of engine.CreateScriptSourceFromString(...).Compile()) instead -- as that will help a lot more in repeat runs.