micro optimizations in allocating hashes and arrays

[dnesteryuk]

There were a few places where hashes could be mutated without side effects, they already work with copies of original hashes. So, merge! replaced some of merge calls.

Also, Grape::Util::StackableValues, Grape::Util::ReverseStackableValues and Grape::Util::InheritableValues got a base class which keeps common methods
to avoid duplication. Besides, that a [] and keys methods were optimized to allocate less arrays when it is possible.

There is an additional benchmark which measures more complex scenario. Unfortunately, there isn't noticeable difference in performance between the master and this change. However, I also profiled the same endpoint as the benchmark has.

Before this change:

 Measure Mode: memory
 Thread ID: 47216803207600
 Fiber ID: 47216805628400
 Total: 40536.000000
 Sort by: self_time

 %self      total      self      wait     child     calls  name
 13.85   6848.000  5616.000     0.000  1232.000       15  *Hash#each_pair
 9.73    3944.000  3944.000     0.000     0.000       17   Hash#merge
 8.80    5088.000  3568.000     0.000  1520.000        8  *Array#map
 8.57   11720.000  3472.000     0.000  8248.000       27  *Class#new
 4.54    1840.000  1840.000     0.000     0.000       46   Symbol#to_s
 4.44    1800.000  1800.000     0.000     0.000       20  *Grape::Util::StackableValues#[]
 3.36    1440.000  1360.000     0.000    80.000        5   Grape::Endpoint#run_filters
 2.88    3040.000  1168.000     0.000  1872.000        9  *Hash#each
 2.37    1200.000   960.000     0.000   240.000        6   <Class::Grape::Router>#normalize_path
 2.29    1088.000   928.000     0.000   160.000        4   ActiveSupport::HashWithIndifferentAccess#[]=
 2.21    1128.000   896.000     0.000   232.000        8  *Kernel#dup
 2.07     840.000   840.000     0.000     0.000        9   String#split
 1.64   12008.000   664.000     0.000 11344.000        1   Grape::Endpoint#run_validators

After this change:

 Measure Mode: memory
 Thread ID: 47390769740200
 Fiber ID: 47390772121400
 Total: 37296.000000
 Sort by: self_time

 %self      total      self      wait     child     calls  name
 15.06   6848.000  5616.000     0.000  1232.000       15  *Hash#each_pair
 9.57    5088.000  3568.000     0.000  1520.000        8  *Array#map
 9.31   11720.000  3472.000     0.000  8248.000       27  *Class#new
 4.93    1840.000  1840.000     0.000     0.000       46   Symbol#to_s
 4.35    1624.000  1624.000     0.000     0.000        7   Hash#merge
 3.65    1440.000  1360.000     0.000    80.000        5   Grape::Endpoint#run_filters
 3.13    3040.000  1168.000     0.000  1872.000        9  *Hash#each
 2.57    1200.000   960.000     0.000   240.000        6   <Class::Grape::Router>#normalize_path
 2.49    1088.000   928.000     0.000   160.000        4   ActiveSupport::HashWithIndifferentAccess#[]=
 2.40    1128.000   896.000     0.000   232.000        8  *Kernel#dup
 2.25     840.000   840.000     0.000     0.000        9   String#split
 2.15    1400.000   800.000     0.000   600.000       20  *Grape::Util::StackableValues#[]
 1.78   11768.000   664.000     0.000 11104.000        1   Grape::Endpoint#run_validators

• The total memory usage went down from 40536 to 37296 bytes.
• Hash#merge moved from second position to 5th.
• Grape::Util::StackableValues#[] moved from 6th to 12th.

[grape-bot]

	1 Warning
⚠️	There’re library changes, but not tests. That’s OK as long as you’re refactoring existing code.

Generated by 🚫 danger

[dblock]

👍