RDoc::Markup::ToHtml uses RDoc::RubyLex.tokenize directly and calls RDoc::TokenStream.to_html with returned tokens.

On the one hand, RDoc::Parser::RubyTools#get_tk collects up TkSYMBEG and trailing tokens inside the symbol with tokens by RDoc::RubyLex.

I moved logic what collects up tokens for a symbol from RDoc::Parser::RubyTools#get_tk to RDoc::RubyLex. So RDoc::TokenStream.to_html becomes to handle symbols as value.

Before, each symbols are separeted TkSYMBEG and trailing tokens:

<pre class="ruby">
<span class="ruby-identifier">a</span> = :<span class="ruby-identifier">hoge</span>
</pre>

I think what : and hoge with ruby-identifier attribute are separeted is not correct because symbol is one literal and one value. It's not identifier.

After, each symbols are treated as literal with ruby-value attribute.

<pre class="ruby">
<span class="ruby-identifier">a</span> = <span class="ruby-value">:hoge</span>
</pre>