Summary

All string-based eval variants (eval, instance_eval, class_eval, binding.eval) are significantly slower in JRuby 10 compared to JRuby 1.7. Block-based eval is unaffected.

Benchmark

require 'benchmark'

n = 50_000
obj = Object.new
klass = Class.new
b = binding

Benchmark.bm(20) do |x|
  x.report("eval string")       { n.times { eval("1 + 1") } }
  x.report("instance_eval str") { n.times { obj.instance_eval("1 + 1") } }
  x.report("instance_eval blk") { n.times { obj.instance_eval { 1 + 1 } } }
  x.report("class_eval str")    { n.times { klass.class_eval("1 + 1") } }
  x.report("class_eval blk")    { n.times { klass.class_eval { 1 + 1 } } }
  x.report("binding.eval")      { n.times { b.eval("1 + 1") } }
end

Results

Key observations:

• String-based eval: 7-13x regression from JRuby 1.7 to JRuby 10
• Block-based eval: No regression, actually improved
• JRuby 1.7 was actually faster than MRI for string eval, JRuby 10 is slower

Impact

This affects code using dynamic evaluation with strings, including:

• DSLs that compile strings
• Template engines
• Dynamic method generation
• Legacy code using string eval

Environment

• JRuby 10.0.2.0 on Java 21 (Temurin 21.0.9)
• JRuby 1.7.27 on Java 8 (Corretto 8.472)
• MRI 3.4.7
• macOS Darwin 25.1.0