[moberegger]

Switches _blank? from BLANK == value to Blank === value.

Originally _blank? did a Ruby-level dispatch to Blank#==, which itself calls super (for an identity check) and then Blank === other. We can short circuit this to simply be Blank === other. This gives us the same result for less work because Blank#== would always return true for any Blank instance. We don't actually need to an identity check.

blank1 = Jbuilder::Blank.new
blank2 = Jbuilder::Blank.new

puts blank1 == blank2 # true
puts blank2 == blank1 # true

Below is an IPS benchmark against a simple builder that exercises _blank a few times.

# frozen_string_literal: true

require 'benchmark/ips'
require_relative 'lib/jbuilder'

json = Jbuilder.new

Benchmark.ips do |x|
  x.report('old') do |n|
    i = 0
    while i < n
      i += 1

      json.set! :foo do
        json.set! :bar, 123
      end
    end
  end

  x.report('new') do |n|
    i = 0
    while i < n
      i += 1

      json.set! :foo do
        json.set! :bar, 123
      end
    end
  end

  x.hold! 'ips_results'
  x.compare! order: :baseline
end

Without YJIT

ruby 4.0.7 (2026-09-15 revision 229531a6cf) +PRISM [arm64-darwin23]
Warming up --------------------------------------
                 old    55.923k i/100ms
                 new    70.938k i/100ms
Calculating -------------------------------------
                 old    563.384k (± 1.4%) i/s    (1.77 μs/i) -      2.852M in   5.062397s
                 new    716.586k (± 1.4%) i/s    (1.40 μs/i) -      3.618M in   5.048717s

Comparison:
old:   563383.9 i/s
new:   716585.6 i/s - 1.27x  faster

With YJIT

ruby 4.0.7 (2026-09-15 revision 229531a6cf) +YJIT +PRISM [arm64-darwin23]
Warming up --------------------------------------
                 old   169.546k i/100ms
                 new   178.359k i/100ms
Calculating -------------------------------------
                 old      1.712M (± 2.3%) i/s  (584.09 ns/i) -      8.647M in   5.050534s
Calculating -------------------------------------
                 new      1.783M (± 3.3%) i/s  (560.71 ns/i) -      8.918M in   5.000406s

Comparison:
old:  1712065.7 i/s
new:  1783445.2 i/s - same-ish: difference falls within error

[moberegger]

Not necessary for the optimization, but figured I'd remove it since it is no longer running in the gem.

I suppose it's possible that a user is doing something with Jbuilder::Blank directly that may require this, but I believe this class was intended to be an implementation detail.

[seanpdoyle]

Since this is technically a change to the public API (Jbuilder::Blank is not marked with # :nodoc:), I think it'd be best (and increase the likelihood of this PR being reviewed and merged) to maintain the method and open a separate PR to deprecate it.

[moberegger]

Yup, I can put it back no problem!

To clarify:

but I believe this class was intended to be an implementation detail.

I failed to mention that my only rationale there was that the Readme doesn't mention Blank.

[seanpdoyle]

Since this is technically a change to the public API (Jbuilder::Blank is not marked with # :nodoc:), I think it'd be best (and increase the likelihood of this PR being reviewed and merged) to maintain the method and open a separate PR to deprecate it.

[seanpdoyle]

Could you share the source for the IPS benchmark mentioned in the pull request description?

[moberegger]

@seanpdoyle Updated the PR description to add the bench source. Also updated the results for Ruby 4.0.7 (originally they were run on 4.0.6). Also included a run with YJIT enabled (which was missing before).

YJIT is a tad faster, but still falls within the margin of error according to the benchmark/ips tool.

[moberegger]

Interestingly, when initializing a new Jbuilder object in each pass

Benchmark.ips do |x|
  x.report('old') do |n|
    i = 0
    while i < n
      i += 1

      json = Jbuilder.new
      json.set! :foo do
        json.set! :bar, 123
      end
    end
  end

  x.report('new') do |n|
    i = 0
    while i < n
      i += 1

      json = Jbuilder.new
      json.set! :foo do
        json.set! :bar, 123
      end
    end
  end

  x.hold! 'ips_results'
  x.compare! order: :baseline
end

ruby 4.0.7 (2026-09-15 revision 229531a6cf) +YJIT +PRISM [arm64-darwin23]
Warming up --------------------------------------
                 old   193.640k i/100ms
                 new   195.000k i/100ms
Calculating -------------------------------------
                 old      2.228M (± 4.2%) i/s  (448.84 ns/i) -     11.231M in   5.040934s
                 new      2.405M (± 2.8%) i/s  (415.75 ns/i) -     12.090M in   5.026368s

Comparison:
old:  2227983.9 i/s
new:  2405315.3 i/s - 1.08x  faster