[debt] make XML::RelaxNG.read_memory an alias for XML::RelaxNG.new

[flavorjones]

The method XML::RelaxNG.read_memory is intended to be a factory method for creating a new schema from the string contents of the schema file.

But ... this is exactly what XML::Schema.new does. And under the hood, libxml2 is still calling the same parsing code, so I would be surprised if it was any faster.

This duplication introduces debt to the codebase because it's another path we need to test (note: we're not doing a good job of this already), and the fact that we've begun to have to add workarounds for libxml2 misbehavior (see #1985) and it's not clear if we need to add it to both code paths or what.

Anyway, here's what I want:

1. benchmark the two paths, make sure there isn't a performance reason to keep read_memory around
2. remove read_memory and make it an alias for XML::Schema.new

[flavorjones]

Benchmark:

#! /usr/bin/env ruby

require "nokogiri"
require "benchmark/ips"

ASSETS_DIR          = File.expand_path File.join(File.dirname(__FILE__), 'test', 'files')
PO_SCHEMA_FILE      = File.join(ASSETS_DIR, 'po.xsd')

xsd = File.read(PO_SCHEMA_FILE)

Benchmark.ips do |b|
  b.time = 10
  b.warmup = 2

  b.report("Schema.new") do
    Nokogiri::XML::Schema.new(xsd)
  end

  b.report("Schema.from_document") do
    Nokogiri::XML::Schema.from_document(Nokogiri::XML::Document.parse(xsd))
  end

  b.report("Schema.read_memory") do
    Nokogiri::XML::Schema.read_memory(xsd)
  end

  b.compare!
end

Results:

Warming up --------------------------------------
          Schema.new   928.000  i/100ms
Schema.from_document   954.000  i/100ms
  Schema.read_memory     1.002k i/100ms
Calculating -------------------------------------
          Schema.new      8.430k (± 9.8%) i/s -     84.448k in  10.118964s
Schema.from_document      8.040k (±13.1%) i/s -     79.182k in  10.055691s
  Schema.read_memory      9.334k (±17.4%) i/s -     90.180k in  10.051572s

Comparison:
  Schema.read_memory:     9333.8 i/s
          Schema.new:     8429.7 i/s - same-ish: difference falls within error
Schema.from_document:     8040.4 i/s - same-ish: difference falls within error

Generally, read_memory is modestly faster though shows greater variation in performance. I don't think the modest performance justifies the presence of a second implementation. @tenderlove WDYT?

[flavorjones]

I'll also add that #2115 documents that read_memory doesn't work (and has probably not worked for a long time) in JRuby.