Investigate alternative higher-performance XML processors

[hartsock]

This is a low priority wishlist item from an IRC conversation in #rbvmomi-dev on freenode.

Based on pain points in nokogiri, around memory use, performance, and installation problems it's been suggested that we consider a different XML library. XML parser performance problems are not trivial and are not necessarily easily solved by simply swapping out a library. Some problems are endemic to a library's internal design and others are caused by poor a understanding of what performance characteristics we need to design our utility code for.

Background Research:

• http://www.danneu.com/posts/16-xml-parsing-benchmark-ox-vs-nokogiri-sax-vs-dom-traversal/
• http://stackoverflow.com/questions/4036659/what-are-fast-xml-parsers-for-ruby

Possible candidates:

• http://yorickpeterse.com/articles/oga-a-new-xml-and-html-parser-for-ruby/
• http://www.ohler.com/ox/

Possible courses of action:

• Continue with Nokogiri but alter code so it is more efficient
• Implement an alternative XML parser in a branch and bench mark it against the master
• Alter the codebase to allow people to supply their own XML parsing module based on their performance needs
• Do NOTHING. Because, really... is it worth the effort?

TODO:

• Before doing anything: create unit tests to automatically and deterministically cover all XML document use-cases this establishes rbVmomi's current expected behavior. Do nothing before doing this as this base-test work allows us to evaluate fairly alternative implementations.
• create performance metrics to evaluate XML processing speed fairly
• develop multiple candidates for consideration in their own branches

NOTE:

• This is a large change and should not be attempted lightly (it may be best to not even attempt the effort but we can at least document things here)
• This library feeds a large ecosystem and many quirks may have become embedded as expectations and features
• If someone really feels strongly that they need to do this then we can discuss it and coordinate efforts with other core developers.