upgrade pg_bench

[SamSaffron]

I think you are missing some of the goal of the perf test. the idea here is to generate a giant string as efficiently as possible not lots of arrays.

This ensures every db adapter does exactly the same thing and measured against the same goal.

In general web apps tend to create one giant string from a bunch of db queries

[ermolaev]

In general web apps tend to create one giant string from a bunch of db queries

yes, but not always, and creating a large string we add a measurement error in the benchmark, testing not only the database library

float = 1.0
int = 1

Benchmark.ips do |x|
  x.report("integer") do |times|
    i = 0
    while i < times
      int + int
      i += 1
    end
  end
  x.report("float") do |times|
    i = 0
    while i < times
      float + float
      i += 1
    end
  end

  x.compare!
end

# Comparison:
#              integer: 35359450.3 i/s
#                float: 29289130.9 i/s - 1.21x  slower

Benchmark.ips do |x|
  x.report("integer") do |times|
    i = 0
    s = +''
    while i < times
      int + int
      s << 1.to_s
      i += 1
    end
  end
  x.report("float") do |times|
    i = 0
    s = +''
    while i < times
      float + float
      s << 1.to_s
      i += 1
    end
  end
  x.compare!
end

# Comparison:
#              integer:  4273034.2 i/s
#                float:  4082119.9 i/s - same-ish: difference falls within error

[SamSaffron]

I am fine with creating a few other benchmarks then, as long as we leave the existing ones as is