AS::Callbacks: Refactor :per_key option

[bogdan]

I suppose that :per_key option was originally designed to improve performance of (before/after/around)_filter in ActionController. Maybe someone who is in Rails core team since 2009 can explain why this option appeared.

But now when overall performance issue was fixed, we can rewrite it with 3 lines of code using other code we have.

Performance test: https://gist.github.com/1626009

19% performance improvement on set_callback
19% performance improvement on run_callbacks with :per_key option
15% performance improvement on skip_callback

Checked tests for AR, AM and AP to ensure everything is fine.

Rehearsal ----------------------------------------------------
New set_callback   0.050000   0.000000   0.050000 (  0.045225)
Old set_callback   0.050000   0.010000   0.060000 (  0.056748)
------------------------------------------- total: 0.110000sec

                       user     system      total        real
New set_callback   0.040000   0.000000   0.040000 (  0.043966)
Old set_callback   0.050000   0.000000   0.050000 (  0.053711)
************************************************************************************************************************
Rehearsal --------------------------------------------------------
New define_callbacks   0.020000   0.000000   0.020000 (  0.019756)
Old define_callbacks   0.020000   0.000000   0.020000 (  0.019723)
----------------------------------------------- total: 0.040000sec

                           user     system      total        real
New define_callbacks   0.010000   0.010000   0.020000 (  0.019719)
Old define_callbacks   0.020000   0.010000   0.030000 (  0.019841)
************************************************************************************************************************
Rehearsal -----------------------------------------------------
New run_callbacks   0.000000   0.000000   0.000000 (  0.004363)
Old run_callbacks   0.010000   0.000000   0.010000 (  0.005427)
-------------------------------------------- total: 0.010000sec

                        user     system      total        real
New run_callbacks   0.000000   0.000000   0.000000 (  0.004433)
Old run_callbacks   0.010000   0.000000   0.010000 (  0.005051)
************************************************************************************************************************
Rehearsal --------------------------------------------------------------
New run_callbacks with key   0.000000   0.000000   0.000000 (  0.000471)
Old run_callbacks with key   0.000000   0.000000   0.000000 (  0.001329)
----------------------------------------------------- total: 0.000000sec

                                 user     system      total        real
New run_callbacks with key   0.000000   0.000000   0.000000 (  0.000473)
Old run_callbacks with key   0.000000   0.000000   0.000000 (  0.000570)
************************************************************************************************************************
Rehearsal -----------------------------------------------------
New skip_callback   0.020000   0.010000   0.030000 (  0.029028)
Old skip_callback   0.030000   0.000000   0.030000 (  0.033500)
-------------------------------------------- total: 0.060000sec

                        user     system      total        real
New skip_callback   0.030000   0.000000   0.030000 (  0.028263)
Old skip_callback   0.030000   0.000000   0.030000 (  0.033339)

[josevalim]

/cc @wycats

[tenderlove]

I think you can use Array#concat in this case to avoid creating more arrays. Probably doesn't matter though. :-/

[bogdan]

There are many tiny improvement that should be done.

But I just want to keep focus on problems that will give a significant performance boost and/or reduce complexity.
I will move to details as soon all main problems will be fixed,

[tenderlove]

This patch seems good to me.