# This benchmark arose from rspec/rspec-support#199 where we experimented with # faster ways of generating / capturing a backtrace and whether it made sense # to lazily generate it using `raise` to capture the backtrace via an exception. # See also rspec/rspec-mocks#937 require 'benchmark/ips' def use_raise_to_capture_caller use_raise_lazily.backtrace end def use_raise_lazily raise "nope" rescue StandardError => exception return exception end def create_stack_trace(n, &block) return create_stack_trace(n - 1, &block) if n > 0 yield end [10, 50, 100].each do |frames| puts "-" * 80 puts "With #{frames} extra stack frames" puts "-" * 80 create_stack_trace(frames) do Benchmark.ips do |x| x.report("caller() ") { caller } x.report("caller_locations() ") { caller_locations } x.report("raise with backtrace ") { use_raise_to_capture_caller } x.report("raise and store (lazy)") { use_raise_lazily } x.report("caller(1, 2) ") { caller(1, 2) } x.report("caller_locations(1, 2)") { caller_locations(1, 2) } x.compare! end end end __END__ -------------------------------------------------------------------------------- With 10 extra stack frames -------------------------------------------------------------------------------- Calculating ------------------------------------- caller() 5.583k i/100ms caller_locations() 14.540k i/100ms raise with backtrace 4.544k i/100ms raise and store (lazy) 27.028k i/100ms caller(1, 2) 25.739k i/100ms caller_locations(1, 2) 48.848k i/100ms ------------------------------------------------- caller() 61.386k (±11.6%) i/s - 307.065k caller_locations() 176.033k (±12.8%) i/s - 872.400k raise with backtrace 48.348k (±10.5%) i/s - 240.832k raise and store (lazy) 425.768k (±10.7%) i/s - 2.108M caller(1, 2) 368.142k (±18.9%) i/s - 1.776M caller_locations(1, 2) 834.431k (±17.8%) i/s - 4.054M Comparison: caller_locations(1, 2): 834431.2 i/s raise and store (lazy): 425767.6 i/s - 1.96x slower caller(1, 2) : 368142.0 i/s - 2.27x slower caller_locations() : 176032.6 i/s - 4.74x slower caller() : 61386.0 i/s - 13.59x slower raise with backtrace : 48348.2 i/s - 17.26x slower -------------------------------------------------------------------------------- With 50 extra stack frames -------------------------------------------------------------------------------- Calculating ------------------------------------- caller() 2.282k i/100ms caller_locations() 6.446k i/100ms raise with backtrace 2.138k i/100ms raise and store (lazy) 23.649k i/100ms caller(1, 2) 22.113k i/100ms caller_locations(1, 2) 36.586k i/100ms ------------------------------------------------- caller() 24.105k (± 9.9%) i/s - 120.946k caller_locations() 68.610k (± 7.9%) i/s - 341.638k raise with backtrace 21.458k (± 9.6%) i/s - 106.900k raise and store (lazy) 341.152k (± 8.1%) i/s - 1.703M caller(1, 2) 297.805k (±12.5%) i/s - 1.482M caller_locations(1, 2) 557.278k (±16.6%) i/s - 2.744M Comparison: caller_locations(1, 2): 557278.2 i/s raise and store (lazy): 341151.6 i/s - 1.63x slower caller(1, 2) : 297804.8 i/s - 1.87x slower caller_locations() : 68610.3 i/s - 8.12x slower caller() : 24105.5 i/s - 23.12x slower raise with backtrace : 21458.2 i/s - 25.97x slower -------------------------------------------------------------------------------- With 100 extra stack frames -------------------------------------------------------------------------------- Calculating ------------------------------------- caller() 1.327k i/100ms caller_locations() 3.773k i/100ms raise with backtrace 1.235k i/100ms raise and store (lazy) 19.990k i/100ms caller(1, 2) 18.269k i/100ms caller_locations(1, 2) 29.668k i/100ms ------------------------------------------------- caller() 13.879k (± 9.9%) i/s - 69.004k caller_locations() 39.070k (± 7.6%) i/s - 196.196k raise with backtrace 12.703k (±12.7%) i/s - 62.985k raise and store (lazy) 243.959k (± 8.3%) i/s - 1.219M caller(1, 2) 230.289k (± 8.2%) i/s - 1.151M caller_locations(1, 2) 406.804k (± 8.8%) i/s - 2.047M Comparison: caller_locations(1, 2): 406804.3 i/s raise and store (lazy): 243958.7 i/s - 1.67x slower caller(1, 2) : 230288.9 i/s - 1.77x slower caller_locations() : 39069.8 i/s - 10.41x slower caller() : 13879.4 i/s - 29.31x slower raise with backtrace : 12702.9 i/s - 32.02x slower