TokenSink trait forces interior mutability/RefCell usage

[LunNova]

html5ever/html5ever/src/tokenizer/interface.rs

Lines 83 to 99 in 4641184

	pub trait TokenSink {
	type Handle;
	/// Process a token.
	fn process_token(&self, token: Token, line_number: u64) -> TokenSinkResult<Self::Handle>;
	// Signal sink that tokenization reached the end.
	fn end(&self) {}
	/// Used in the markup declaration open state. By default, this always
	/// returns false and thus all CDATA sections are tokenized as bogus
	/// comments.
	/// <https://html.spec.whatwg.org/multipage/#markup-declaration-open-state>
	fn adjusted_current_node_present_but_not_in_html_namespace(&self) -> bool {
	false
	}
	}

The TokenSink trait's process_token method takes &self

Many usecases for a TokenSink are going to involve mutability. This includes multiple files in this repo such as tree_builder.

I think TokenSink would be better like this:

pub trait TokenSink {
    type Handle;
    fn process_token(self, token: Token, line_number: u64) -> TokenSinkResult<Self::Handle>;
    fn end(self);
    fn adjusted_current_node_present_but_not_in_html_namespace(self);
}

Usecases that need mutability could then be impl TokenSink for &mut MyTokenSink, and those that don't like the printer examples can be impl TokenSink for &MyType.
I haven't tried implementing this refactor, it might not actually work out well.

[nicoburns]

The requirement for interior mutability is actually a fairly recent change to html5ever (v0.29). It was introduced in #548 with the motivation being to support reentrancy in the parser. There is some hope that it might be possible to eliminate the interior mutability requirement in #590

[simonwuelker]

The proposal above sounds interesting on its own though, assuming it works.