fix odd convert from string to integer and float

[kaisecheng]

Fix convert to covert hexadecimal notation and scientific notation string to float and integer

Add support for hexadecimal float notation (0x1.edcdp6) and scientific
notation (1.230000e+02) in convert option for integer and float. Includes manual
sign parsing for JRuby compatibility and BigDecimal for precision.

Fix: #174

JRuby cannot parse negative hex string to float Float("-0x7b") and uppsercase hex Float("0X1A"), but CRuby can.
JRuby 9.4.5.0 Logstash 8.13 also cannot parse floating point hex string Float("0x1.edcdp6")
This commit adds parsing for signed hex strings for Logstash version >= 8.14. For older versions, type conversions fall back to the old implementation.

[kaisecheng]

benchmark Integer(Float(value)) VS Integer(BigDecimal(value))

require 'benchmark/ips'
require 'bigdecimal'

test_cases = {
  'small_integer' => '123',
  'large_integer' => '9007199254740993',
  'decimal' => '123.456',
  'large_decimal' => '1234567.89012345',
  'scientific' => '1.23e2',
  'scientific_large' => '1.23e10'
}

test_cases.each do |name, value|
  clean_value = value.delete(',')

  puts "Testing: #{name} (#{value})"
  puts "-" * 80

  Benchmark.ips do |x|
    x.config(time: 5, warmup: 2)

    x.report("Integer(Float(value))") do
      Integer(Float(clean_value))
    end

    x.report("Integer(BigDecimal(value))") do
      Integer(BigDecimal(clean_value))
    end

    x.compare!
  end

  puts
end

test suite	BigDecimal slower
small_integer	2.29x
large_integer	1.25x
decimal	10.92x
large_decimal	11.66x
scientific	1.41x
scientific_large	1.44x

[kaisecheng]

created an issue upstream for the difference of jruby and cruby jruby/jruby#9044

[jsvd]

At this point we're only left with decimals and exponentials, and I think that to_i can handle anything that is not exponential, and Float handle the exponentials?

Suggested change

	# floating point number. BigDecimal() can't parse hex string
	return Integer(BigDecimal(value)) if value.count('.') == 1
	return Float(value).to_i if value.include?("e")
	return value.to_i

This change makes the performance hit less significant:

Testing: small_integer (123)
Comparison:
       Test Original:  2117612.5 i/s
         Test Change:  1971480.7 i/s - 1.07x  slower

Testing: large_integer (9007199254740993)
Comparison:
       Test Original:  1780093.5 i/s
         Test Change:  1616786.2 i/s - 1.10x  slower

Testing: decimal (123.456)
Comparison:
       Test Original:  2104084.0 i/s
         Test Change:  1956469.1 i/s - 1.08x  slower

Testing: large_decimal (1234567.89012345)
Comparison:
       Test Original:  2085646.4 i/s
         Test Change:  1905504.3 i/s - 1.09x  slower

Testing: scientific (1.23e2)
Comparison:
       Test Original:  2004532.4 i/s
         Test Change:  1687208.7 i/s - 1.19x  slower

Testing: scientific_large (1.23e10)
Comparison:
       Test Original:  2115260.8 i/s
         Test Change:  1680841.1 i/s - 1.26x  slower

[jsvd]

LGTM