gistfile1.txt

fast: through middleware inserted at 0
slwr: through middleware inserted via @use@
rotr: through endpoint sent via the router
bare: through ActionController::Metal with self.response_body
cntr: through ActionController::Base with self.response_body
text: through ActionController::Base with render :text
tmpl: through ActionController::Base with simple render :template
layt: through ActionController::Base with render :template with layout

Note: These tests are being run without ActiveRecord, which currently
      has an extremely slow connection pool implementation. It might
      be impossible to improve this implementation, but straight
      framework benchmarks should compare reasonably with frameworks
      without ORM support

Before any work (with ActiveRecord):

           real   rps
fast   0.004780   2900
slwr   0.358510   1300
rotr   0.351757   1200
bare   0.476070   1105
cntr   0.745560    695
text   0.978853    570
tmpl   1.110000    539
layt   2.080000    340

Before any work (without ActiveRecord):
* The rest of the optimization work is done without AR

           real   rps
fast   0.004488   2900
slwr   0.282390   1450
rotr   0.301546   1380
bare   0.367800   1281
cntr   0.639768    800
text   0.860590    650
tmpl   0.973746    610
layt   1.900397    360


After first round:
* Memoize commonly used request methods
* Use faster run_callbacks method
* Eliminate the need to check superclass for callbacks changes each time
* Improve log subscriber by remembering list of loggers
* Small improvement to performance of MessageVerifier
* Use simple Request methods in Logger middleware because the more advanced
  versions aren't available yet but are slower anyway
* Stop the flash middleware from pulling in the session even if it's never used

           real   rps
fast   0.004857   2900
slwr   0.067485   2150
rotr   0.087929   2000
bare   0.154795   1720
cntr   0.462846    950
text   0.675570    800
tmpl   0.742683    725
layt   1.855277    375

Second round:
* Eliminate redundant unescaping
* Slightly improve performance of #unescape

           real   rps
fast   0.004365   2900
slwr   0.068906   2200
rotr   0.088535   2000
bare   0.118635   1800
cntr   0.377927   1035
text   0.622412    820
tmpl   0.690456    740
layt   1.799501    375

Third round:
* Improve LookupContext, more to come
* Avoid constantly going through the whole PathSet coercion logic
* Add AS::Dependencies.ref, which returns an object you can call #get on to get
  the real object
  
           real   rps
fast   0.004271   2900
slwr   0.067029   2200
rotr   0.088085   2000
bare   0.103868   1900
cntr   0.355898   1070
text   0.557127    825
tmpl   0.639581    765
layt   1.678789    375

Exceptional work. 550 rps more for Sinatra apps plugged into the router, and 175 rps more for simple actions like AJAX methods that return text :-) Great stuff! Look forward to seeing more :-)