Optimize _is_collection? method

[moberegger]

Saw the _is_collection? helper method showing up as a hotspot in profiles for some of our larger templates.

Running the .all? is 3.75x slower than simply doing a couple of respond_to? comparisons. Comparing the two _is_collection? method implementations directly:

ruby 3.4.2 (2025-02-15 revision d2930f8e7a) +YJIT +PRISM [arm64-darwin24]
Warming up --------------------------------------
                 new     2.287M i/100ms
                 old   875.646k i/100ms
Calculating -------------------------------------
                 new     39.004M (± 1.3%) i/s   (25.64 ns/i) -    196.665M in   5.043030s
                 old     10.398M (± 3.9%) i/s   (96.18 ns/i) -     52.539M in   5.062959s

Comparison:
                 new: 39003805.0 i/s
                 old: 10397648.9 i/s - 3.75x  slower

As a side bonus, this also prevents an additional array allocation from when _object_respond_to? splats the provided method symbols, which adds up for larger responses that use multiple partials.

Calculating -------------------------------------
                 new     0.000  memsize (     0.000  retained)
                         0.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)
                 old    40.000  memsize (     0.000  retained)
                         1.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)

Comparison:
                 new:          0 allocated
                 old:         40 allocated - Infx more

A simple benchmark to compare results through the jbuilder API itself. This is the most minimalistic way I found to exercise the condition.

some_array = Array.new(10) { |i| { id: i, name: "name_#{i}" } }
json = Jbuilder.new

Benchmark.ips do |x|
  x.report('old') do
    json.set! :foo, some_array, :id, :name
  end

  x.report('new') do
    json.set! :foo, some_array, :id, :name
  end

  x.hold! 'temp_results'
  x.compare!
end

ruby 3.4.2 (2025-02-15 revision d2930f8e7a) +YJIT +PRISM [arm64-darwin24]
Warming up --------------------------------------
                 old    18.514k i/100ms
                 new    19.520k i/100ms
Calculating -------------------------------------
                 old    189.515k (± 1.8%) i/s    (5.28 μs/i) -    962.728k in   5.081742s
                 new    198.868k (± 2.5%) i/s    (5.03 μs/i) -    995.520k in   5.009296s

Comparison:
                 new:   198867.8 i/s
                 old:   189515.1 i/s - 1.05x  slower

So ~5% faster, although performance characteristics will depend on how often your template exercises the condition.

Already filed upstream in rails#590

[moberegger]

As an aside, you see a lot of this sort of arg splatting throughout jbuilder, which really adds up in terms of memory allocations. Some are necessary to support things like json.foo @foo, :id, :name, :etc, but others like this can be optimized away.