Roadmap for API Changes

overhaul serialize/pretty printing API

• #530 XHTML formatting can't be turned off

• #415 XML formatting should be no formatting

overhaul and optimize the SAX parsing

• see fairy wing throwdown - SAX parsing is wicked slow.

Node should not be Enumerable; and should have a better attributes API

• #679 Mixing in Enumerable has some unintended consequences; plus we want to improve the attributes API

• Some ideas for a better attributes API?

  • (closed) #666
  • #765

improve CSS query parsing

• #528 support :not() with a nontrivial argument, like :not(div p.c)

• #451 chained :not pseudoselectors

• better jQuery selector and CSS pseudo-selector support:

  • #621
  • #342
  • #628
  • #652
  • #688

• #394 nth-of-type is wrong, and possibly other selectors as well

• #309 incorrect query being executed

• #350 :has is wrong?

DocumentFragment

• there are a few tickets about searches not working properly if you use or do not use the context node as part of the search.
  • #213
  • #370
  • #454
  • #572 could we fix this by making DocumentFragment be a subclass of NodeSet?

Better Syntax for custom XPath function handler

• PR#464

Better Syntax around Node#xpath and NodeSet#xpath

• look at those methods, and use of Node#extract_params in Node#{css,search}
  • we should standardize on a hash of options for these and other calls
• what should NodeSet#xpath return?
  • #656

Encoding

We have a lot of issues open around encoding. How bad are things? Somebody who knows encoding well should head this up.

• Extract EncodingReader as a real object that can be injected https://groups.google.com/forum/#!msg/nokogiri-talk/arJeAtMqvkg/tGihB-iBRSAJ

Reader

It's fundamentally broken, in that we can't stop people from crashing their application if they want to use object reference unsafely.

Class methods that require Document

There are a few methods, like Nokogiri::XML::Comment.new that require a Document object.

We should probably make Document instance methods to wrap this, since it's a non-obvious expectation and thus fails as a convention.

So, instead, let's make alternative methods like Nokogiri::XML::Document#new_comment, and recommend those as the proper convention.

collect_namespaces is just broken

collect_namespaces is returning a hash, which means it can't return namespaces with the same prefix. See this issue for background:

#885

Do we care? This seems like a useless method, but then again I hate XML, so what do I know?

Overhaul ParseOptions

Currently we mirror libxml2's parse options, and then retrofit those options on top of Xerces-J for JRuby.

• I'd like to identify which options work across both parsers,
• And overhaul the parse methods so that these options are easier to use.

By "easier to use" I mean:

• it's unwieldy to create a block to set/unset parse options
• it's unwieldy to create a constant like MY_PARSE_OPTIONS = Nokogiri::XML::ParseOptions::STRICT | Nokogiri::XML::ParseOptions::RECOVER ...
• some options are named dangerously poorly, like NOENT which does the opposite of what it says
• semantically some options should be set/unset together, specifically "this is a trusted document" or "this is an untrusted document" should flip the senses of NONET and NOENT and DTDLOAD together.
• we need the ability to invent new parse options, like the one suggested in #1582 that would allow local entities but not external entities.