Commits on May 1, 2015

1. Reduce unnecessary allocations in Rule creation …

   The pattern of doing `[var].flatten` to coerce values to Arrays makes
   three Array allocations every time. The first is for the array enclosing
   var, the other two are in the C implementation of `flatten` to perform a
   depth-first traversal over any nested arrays found within the outer-most
   array[1]. Rules are allocated a *lot* during the typical execution of an
   application using CanCanCan for authorization. Since the parameters to
   Ability#can are expected to not be deeply nested, we can cut out the
   extra allocations by using Kernel#Array. This benchmark:
   
   [Array#flatten] Mem diff: 18.40625 MB
   [Kernel#Array] Mem diff: 1.01953125 MB
   
   Shows a savings of 17MB of allocations when instantiating 10,000
   Rules[2].
   This translates to a 2x speed improvement in Rule instantiation time:
   
   Calculating -------------------------------------
          Array#flatten    16.147k i/100ms
           Kernel#Array    34.565k i/100ms
   -------------------------------------------------
          Array#flatten    218.381k (± 9.7%) i/s -      1.098M
           Kernel#Array    460.892k (±30.7%) i/s -      2.108M
   
   Comparison:
           Kernel#Array:   460892.2 i/s
          Array#flatten:   218380.6 i/s - 2.11x slower
   
   [1] The first array used by flatten is a stack to push sub-arrays onto
   while traversing, the second is for the result array.
   
   [2] Memory figures are RSS, which doesn't consider shared memory, but we
   have none for this simple benchmark. GC was disabled during the
   benchmark eliminate the effects of unpredictable GC activity. Most of
   the intermediary arrays are immediately garbage, but the act of
   allocating them increases the work the garbage collector has to do.
   Benchmark/ips was used with GC enabled in the second benchmark.
   Benchmark script used is available here:
   https://gist.github.com/timraymond/8d7014e0c7804f0fe508

   

   timraymond committed May 1, 2015
   Configuration menu

   • View commit details
   • Copy full SHA for c81cf44
   • Browse repository at this point

   Copy the full SHA

   c81cf44 View commit details

   Browse the repository at this point in the history