Switch to using XMLParser for parsing RDF/XML

[njh]

This Pull Request is work in progress for fixing #350

• Write test to demonstrate the problem
• Add a test for parsing non-standard namespaces - don't currently seem to test this
• Switch to using XMLReader API
• Optimise code a bit - splitURI() isn't needed anymore, the XML parse can do this
• Check performance of new parser - check it isn't slower than the old implementation

[k00ni]

Its good that you take care of this.

[njh]

I have just committed my work in progress on switching to the XMLReader class.

Annoying it it nearly works but not quite. Some tests are currently failing and I keep getting a very difficult to interpret error message:

EasyRdf\Parser\Exception: startElementHandler() called at state 3 in http://xmlns.com/foaf/0.1/primaryTopic

I am not quite sure what still needs changing/fixing from moving over to the new parser.

[njh]

I am going to re-factor this code to use the XMLParser added in #357

[njh]

I think I may have worked out why this isn't working.

I created this script:
https://gist.github.com/njh/b5ae7a4399b7c89765c81bc4dbb9df2e

And this is the output:

>>> \EasyRdf\XMLParser
  start: root
    start: one
      text: one
    end: one
    start: two
      text: two
    end: two
    start: three
  end: root
<<< \EasyRdf\XMLParser


>>> xml_parser_create_ns()
  start: root
    start: one
      text: one
    end: one
    start: two
      text: two
    end: two
    start: three
    end: three
  end: root
<<< xml_parser_create_ns()

So I think it might relate to the handling of empty XML elements.

[njh]

Tests passing and build fixed! 🎉

Sorry that took so long to work out - the inner workings of the RDF/XML parser code is quite hard to understand.

Little bit of tidying up and it should be ready to merge.
Also want to check that the performance hasn't been made worse.

[k00ni]

Good job so far 👍

[k00ni]

Can we type-hint $parser here? That might make it more clear what it is.

[njh]

Yup, good idea.
It is a bit annoying that the callback has to be public.
But at least @ignore should remove it from the API documentation.

[k00ni]

That should not be in a loop, maybe make it a class property?

[njh]

Really close to giving up on this 😢

1. Now that it sort-of work, I tested the performance and it is a little bit slower than the old implementation
2. When parsing a largish RDF/XML file from dbpedia, it returned a larger number of triples than the old implementation and rapper. So there is still more work to be done fixing bugs. And writing tests.

It is possible that if I fix [2], then the performance will improve.

But I have moved this to 1.2.0 milestone and will back-port the original PR fix.

[k00ni]

Don't give up!

Also, I rather have more readable code than better performance.

[njh]

Might pick this branch up in the future but closing this PR - it has been replaced by #370