fix(tocObj): skip permalink symbol

[curbengh]

Fixes #174 Closes hexojs/hexo-renderer-markdown-it#102

Permalink (in this context) is a single non-word character, word = [a-Z0-9], and wrapped in <a>.
It may be wrapped in whitespace(s)

Scenario	Input	Output (text)	Explanation
1	<h1 id="foo"><a>#</a>bar</h1> <h1 id="foo">bar<a>#</a></h1> <h1 id="foo"><a>#</a>bar<a>#</a></h1>	bar bar bar	# is treated as a permalink and skipped
2	<h1>号码<a>牌</a></h1>	号码	牌 is treated as a permalink
3	<h1><a>#</a></h1>	#	If there is only permalink, parse it
4	<h1><a>b</a> two</h1>	b two	b is not a permalink

---

returns empty string '' instead of null. This is following convention. htmlparser2 DomUtils.getText(elNoText) and camaro camaro.transform(xml, {foo:'invalid'}) also return empty string. This is compatible with toc, as empty string is still falsy (same as null and undefined).

[coveralls]

Coverage increased (+0.03%) to 97.017% when pulling 38a0e5f on curbengh:tocobj-child into 7e5633a on hexojs:master.

[SukkaW]

Suggested change

	'<h1 id="foo"><a>#</a>bar<a>#</a></h1>'
	'<h1 id="foo"><a>#</a>bar<a>#</a></h1>',
	'<h1 id="foo"><a>#</a><span>bar</span><a>#</a></h1>'

[curbengh]

<h1 id="foo"><a>#</a><span>bar</span><a>#</a></h1> would result in #bar# though since bar is wrapped in a children element.

[curbengh]

A workaround can be if (element.type === 'text' || element.type === 'span').

<h1><a>#</a><span>bar</span><a>#</a></h1> -> bar

But the workaround is not foolproof,

<h1><a>#</a>hey<span>bar</span><a>#</a></h1> -> hey
<h1><a>#</a><span>bar</span>hey<a>#</a></h1> -> bar

Since for..of doesn't necessarily follow order, the result can be erratic.

[SukkaW]

The children tag could be anything: <span>, <a> even <p>.

[SukkaW]

LGTM!

[curbengh]

<h2>foo<i>bar</i>baz</h2> only results in foo. I think tocObj should just avoid <a>.

[SukkaW]

<h2>foo<i>bar</i>baz</h2> only results in foo. I think tocObj should just avoid <a>.

@curbengh So what happened if <h2><a href="">foobar</a></h2> or <h2><a>foo</a> bar</h2>?

[curbengh]

with the new workaround (will commit in a minute),

<h2><a href="">foobar</a></h2> -> foobar
<h2><a>foo</a> bar</h2> -> bar (foo is assumed to be a permalink symbol)

I think 2nd case is rare, if a user wants to add a link, 1st case would be used.

---

I just thought of another approach, skip <a> with a single symbol.

[SukkaW]

@curbengh I have a better idea. Using replace to remove the symbol after getText. The symbol should be passed as a parameter.

[curbengh]

The symbol should be passed as a parameter

parameter of which function?

[SukkaW]

Current implementation LGTM!

[curbengh]

rebased to include #181.

[noraj]

@curbengh Awesome, this will be released in 1.9.0? 1.8.2?

[curbengh]

should be part of 1.9.0