XML: improve the DOM interface

[ysbaddaden]

There are a number of shortcomings in the XML DOM interface:

1. Something annoying is that every method to navigate the DOM tree return a nilable: #first_element_child, #next, #parent, ... This is in stark contrast with the rest of the stdlib API, where methods expect the value to exist and raise otherwise, with a question mark alternative method to return a nilable, for example #parent vs #parent?.

   That would be a breaking change, but it would make the API easier and in par with the rest of the stdlib.

2. There are no means to mutate the DOM, apart for the isolated #unlink... that follows the libxml naming instead of the DOM spec by-the-way (it should be named #remove). We should be able to create, if not whole documents, at least nodes and fragments, and to append and insert nodes wherever.

[ysbaddaden]

Because of the changes in #15906 we might want to implement basic mutations of the DOM tree, because custom extensions that directly call to libxml will become dangerous (e.g. every XML::Node will have to always have a strong reference to their owner document, otherwise you risk segfaults).

Possible methods to create nodes:

• XML::Document#create_element(name)
• XML::Document#create_text_node(content)
• XML::Document#create_cdata_section(content)
• XML::Document#create_comment(content)
• XML::Document#create_processing_instruction(target, data)

Possible methods to mutate the tree:

• XML::Node#prepend(*nodes)
• XML::Node#append(*nodes)
• XML::Node#insert(*nodes, *, before: child) => child.before(*nodes)
• XML::Node#insert(*nodes, *, after: child) => child.after(*nodes)
• XML::Node#before(*nodes)
• XML::Node#after(*nodes)
• XML::Node#replace_with(*nodes)
• XML::Node#remove instead of #unlink
• XML::Node#clone