discourse/spec/components/html_to_markdown_spec.rb

457 lines
16 KiB
Ruby

# frozen_string_literal: true
require 'rails_helper'
require 'html_to_markdown'
describe HtmlToMarkdown do
def html_to_markdown(html, opts = {})
HtmlToMarkdown.new(html, opts).to_markdown
end
it "remove whitespaces" do
html = <<-HTML
<div dir="auto">Hello,
<div dir="auto"><br></div>
<div dir="auto">&nbsp; &nbsp; This is the 1st paragraph.&nbsp; &nbsp; </div>
<div dir="auto"><br></div>
<div dir="auto">
&nbsp; &nbsp; &nbsp; &nbsp; This is another paragraph
</div>
</div>
HTML
expect(html_to_markdown(html)).to eq("Hello,\n\nThis is the 1st paragraph.\n\nThis is another paragraph")
html = <<~HTML
<body text="#000000" bgcolor="#FFFFFF">
<p>Let me see if it happens by answering your message through
Thunderbird.</p>
<p>Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1
Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1
Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1
Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1
Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1
Long sentence 1
</p>
</body>
HTML
markdown = <<~MD
Let me see if it happens by answering your message through Thunderbird.
Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1 Long sentence 1
MD
expect(html_to_markdown(html)).to eq(markdown.strip)
html = <<~HTML
<p> This post
has lots<br> of
space
</p>
<pre> This space was left untouched !</pre>
HTML
markdown = <<~MD
This post has lots
of space
```
This space was left untouched !
```
MD
expect(html_to_markdown(html)).to eq(markdown.strip)
end
it "doesn't error on non-inline elements like (aside, section)" do
html = <<~HTML
<aside class="quote no-group">
<blockquote>
<p>Hello,<br>is it me you're looking for?</p>
</blockquote>
<br>
</aside>
HTML
markdown = <<~MD
> Hello,
> is it me you're looking for?
MD
expect(html_to_markdown(html)).to eq(markdown.strip)
end
it "skips hidden tags" do
expect(html_to_markdown(%Q{<p>Hello <span hidden>cruel </span>World!</p>})).to eq("Hello World!")
end
it "converts <strong>" do
expect(html_to_markdown("<strong>Strong</strong>")).to eq("**Strong**")
expect(html_to_markdown("<strong>Str*ng</strong>")).to eq("__Str*ng__")
end
it "converts <b>" do
expect(html_to_markdown("<b>Bold</b>")).to eq("**Bold**")
expect(html_to_markdown("<b>B*ld</b>")).to eq("__B*ld__")
html = <<~HTML
Before
<p><b>Bold
<br>
<br>
</b>
</p>
After
HTML
expect(html_to_markdown(html)).to eq("Before\n\n**Bold**\n\nAfter")
end
it "converts <em>" do
expect(html_to_markdown("<em>Emphasis</em>")).to eq("*Emphasis*")
expect(html_to_markdown("<em>Emph*sis</em>")).to eq("_Emph*sis_")
end
it "converts <i>" do
expect(html_to_markdown("<i>Italic</i>")).to eq("*Italic*")
expect(html_to_markdown("<i>It*lic</i>")).to eq("_It*lic_")
end
it "converts <a>" do
expect(html_to_markdown(%Q{<a href="https://www.discourse.org">Discourse</a>})).to eq("[Discourse](https://www.discourse.org)")
end
it "supports SiteSetting.allowed_href_schemes" do
SiteSetting.allowed_href_schemes = "tel|steam"
expect(html_to_markdown(%Q{<a href="steam://store/48000">LIMBO</a>})).to eq("[LIMBO](steam://store/48000)")
end
it "removes empty & invalid <a>" do
expect(html_to_markdown(%Q{<a>Discourse</a>})).to eq("Discourse")
expect(html_to_markdown(%Q{<a href="">Discourse</a>})).to eq("Discourse")
expect(html_to_markdown(%Q{<a href="foo.bar">Discourse</a>})).to eq("Discourse")
end
HTML_WITH_IMG ||= %Q{<img src="https://www.discourse.org/logo.svg" alt="Discourse Logo">}
HTML_WITH_CID_IMG ||= %Q{<img src="cid:ii_1525434659ddb4cb" title="Discourse Logo">}
it "converts <img>" do
expect(html_to_markdown(HTML_WITH_IMG)).to eq("![Discourse Logo](https://www.discourse.org/logo.svg)")
end
it "keeps <img> with 'keep_img_tags'" do
expect(html_to_markdown(HTML_WITH_IMG, keep_img_tags: true)).to eq(HTML_WITH_IMG)
end
it "removes empty & invalid <img>" do
expect(html_to_markdown(%Q{<img>})).to eq("")
expect(html_to_markdown(%Q{<img src="">})).to eq("")
expect(html_to_markdown(%Q{<img src="foo.bar">})).to eq("")
end
it "keeps <img> with src='cid:' with 'keep_cid_imgs'" do
expect(html_to_markdown(HTML_WITH_CID_IMG, keep_cid_imgs: true)).to eq(HTML_WITH_CID_IMG)
end
it "skips hidden <img>" do
expect(html_to_markdown(%Q{<img src="https://www.discourse.org/logo.svg" width=0>})).to eq("")
expect(html_to_markdown(%Q{<img src="https://www.discourse.org/logo.svg" height="0">})).to eq("")
end
it "supports width/height on <img>" do
expect(html_to_markdown(%Q{<img src="https://www.discourse.org/logo.svg" height=100>})).to eq("![](https://www.discourse.org/logo.svg)")
expect(html_to_markdown(%Q{<img src="https://www.discourse.org/logo.svg" width=200>})).to eq("![](https://www.discourse.org/logo.svg)")
expect(html_to_markdown(%Q{<img src="https://www.discourse.org/logo.svg" height=100 width=200>})).to eq("![|200x100](https://www.discourse.org/logo.svg)")
end
(1..6).each do |n|
it "converts <h#{n}>" do
expect(html_to_markdown("<h#{n}>Header #{n}</h#{n}>")).to eq("#" * n + " Header #{n}")
end
end
it "converts <br>" do
expect(html_to_markdown("Before<br>Inside<br>After")).to eq("Before\nInside\nAfter")
end
it "skips <br> inside <p> if next character is \n" do
expect(html_to_markdown("<p>Before<br>\nInside<br>After</p>")).to eq("Before\nInside\nAfter")
end
it "converts <hr>" do
expect(html_to_markdown("Before<hr>Inside<hr>After")).to eq("Before\n\n---\n\nInside\n\n---\n\nAfter")
end
it "converts <tt>" do
expect(html_to_markdown("<tt>Teletype</tt>")).to eq("`Teletype`")
end
it "converts <code>" do
expect(html_to_markdown("<code>Code</code>")).to eq("`Code`")
end
it "supports <ins>" do
expect(html_to_markdown("This is an <ins>insertion</ins>")).to eq("This is an <ins>insertion</ins>")
end
it "supports <del>" do
expect(html_to_markdown("This is a <del>deletion</del>")).to eq("This is a <del>deletion</del>")
end
it "supports <sub>" do
expect(html_to_markdown("H<sub>2</sub>O")).to eq("H<sub>2</sub>O")
end
it "supports <mark>" do
expect(html_to_markdown("<mark>This is highlighted!</mark>")).to eq("<mark>This is highlighted!</mark>")
end
it "supports <sup>" do
expect(html_to_markdown("<sup>Super Script!</sup>")).to eq("<sup>Super Script!</sup>")
end
it "supports <small>" do
expect(html_to_markdown("<small>Small</small>")).to eq("<small>Small</small>")
end
it "supports <kbd>" do
expect(html_to_markdown("<kbd>CTRL</kbd>+<kbd>C</kbd>")).to eq("<kbd>CTRL</kbd>+<kbd>C</kbd>")
end
it "supports <abbr>" do
expect(html_to_markdown(%Q{<abbr title="Civilized Discourse Construction Kit, Inc.">CDCK</abbr>})).to eq(%Q{<abbr title="Civilized Discourse Construction Kit, Inc.">CDCK</abbr>})
end
it "supports <s>" do
expect(html_to_markdown("<s>Strike Through</s>")).to eq("~~Strike Through~~")
end
it "supports <strike>" do
expect(html_to_markdown("<strike>Strike Through</strike>")).to eq("~~Strike Through~~")
end
it "supports <blockquote>" do
expect(html_to_markdown("<blockquote>Quote</blockquote>")).to eq("> Quote")
end
it "supports <ul>" do
expect(html_to_markdown("<ul><li>🍏</li><li>🍐</li><li>🍌</li></ul>")).to eq("- 🍏\n- 🍐\n- 🍌")
expect(html_to_markdown("<ul>\n<li>🍏</li>\n<li>🍐</li>\n<li>🍌</li>\n</ul>")).to eq("- 🍏\n- 🍐\n- 🍌")
end
it "supports <ol>" do
expect(html_to_markdown("<ol><li>🍆</li><li>🍅</li><li>🍄</li></ol>")).to eq("1. 🍆\n1. 🍅\n1. 🍄")
end
it "supports <p> inside <li>" do
expect(html_to_markdown("<ul><li><p>🍏</p></li><li><p>🍐</p></li><li><p>🍌</p></li></ul>")).to eq("- 🍏\n\n- 🍐\n\n- 🍌")
end
it "supports <ul> inside <ul>" do
expect(html_to_markdown(<<-HTML
<ul>
<li>Fruits
<ul>
<li>🍏</li>
<li>🍐</li>
<li>🍌</li>
</ul>
</li>
<li>Vegetables
<ul>
<li>🍆</li>
<li>🍅</li>
<li>🍄</li>
</ul>
</li>
</ul>
HTML
)).to eq("- Fruits\n - 🍏\n - 🍐\n - 🍌\n- Vegetables\n - 🍆\n - 🍅\n - 🍄")
end
it "supports bare <li>" do
expect(html_to_markdown("<li>I'm alone</li>")).to eq("- I'm alone")
end
it "supports <pre>" do
expect(html_to_markdown("<pre>var foo = 'bar';</pre>")).to eq("```\nvar foo = 'bar';\n```")
expect(html_to_markdown("<pre><code>var foo = 'bar';</code></pre>")).to eq("```\nvar foo = 'bar';\n```")
expect(html_to_markdown(%Q{<pre><code class="lang-javascript">var foo = 'bar';</code></pre>})).to eq("```javascript\nvar foo = 'bar';\n```")
expect(html_to_markdown("<pre> function f() {\n console.log('Hello world!');\n }</pre>")).to eq("```\n function f() {\n console.log('Hello world!');\n }\n```")
end
it "supports <pre> inside <blockquote>" do
expect(html_to_markdown("<blockquote><pre><code>var foo = 'bar';</code></pre></blockquote>")).to eq("> ```\n> var foo = 'bar';\n> ```")
end
it "works" do
expect(html_to_markdown("<ul><li><p>A list item with a blockquote:</p><blockquote><p>This is a <strong>blockquote</strong><br>inside a list item.</p></blockquote></li></ul>")).to eq("- A list item with a blockquote:\n\n > This is a **blockquote**\n > inside a list item.")
end
it "supports html document" do
expect(html_to_markdown("<html><body>Hello<div>World</div></body></html>")).to eq("Hello\nWorld")
end
it "handles <p>" do
expect(html_to_markdown("<p>1st paragraph</p><p>2nd paragraph</p>")).to eq("1st paragraph\n\n2nd paragraph")
expect(html_to_markdown("<body><p>1st paragraph</p>\n <p> 2nd paragraph\n 2nd paragraph</p>\n<p>3rd paragraph</p></body>")).to eq("1st paragraph\n\n2nd paragraph 2nd paragraph\n\n3rd paragraph")
end
it "handles <div>" do
expect(html_to_markdown("<div>1st div</div><div>2nd div</div>")).to eq("1st div\n2nd div")
end
it "swallows <span>" do
expect(html_to_markdown("<span>Span</span>")).to eq("Span")
end
it "swallows <u>" do
expect(html_to_markdown("<u>Underline</u>")).to eq("Underline")
end
it "removes <script>" do
expect(html_to_markdown("<script>var foo = 'bar'</script>")).to eq("")
end
it "removes <style>" do
expect(html_to_markdown("<style>* { margin: 0 }</style>")).to eq("")
end
it "handles <p> and <div> within <span>" do
html = "<div>1st paragraph<span><div>2nd paragraph</div><p>3rd paragraph</p></span></div>"
expect(html_to_markdown(html)).to eq("1st paragraph\n2nd paragraph\n\n3rd paragraph")
end
it "handles <p> and <div> within <font>" do
html = "<font>1st paragraph<br><span>2nd paragraph</span><div>3rd paragraph</div><p>4th paragraph</p></font>"
expect(html_to_markdown(html)).to eq("1st paragraph\n2nd paragraph\n3rd paragraph\n\n4th paragraph")
end
context "with an oddly placed <br>" do
it "handles <strong>" do
expect(html_to_markdown("Hello <strong><br>Bold</strong> World")).to eq("Hello\n**Bold** World")
expect(html_to_markdown("Hello <strong>Bold<br></strong> World")).to eq("Hello **Bold**\nWorld")
expect(html_to_markdown("Hello <strong>Bold<br>text</strong> World")).to eq("Hello **Bold**\n**text** World")
end
it "handles <em>" do
expect(html_to_markdown("Hello <em><br>Italic</em> World")).to eq("Hello\n*Italic* World")
expect(html_to_markdown("Hello <em>Italic<br></em> World")).to eq("Hello *Italic*\nWorld")
expect(html_to_markdown("Hello <em>Italic<br>text</em> World")).to eq("Hello *Italic*\n*text* World")
end
it "works" do
expect(html_to_markdown("<div>A <b> B <i> C <br> D </i> E <br> F </b> G</div>")).to eq("A __B *C*__\n__*D* E__\n**F** G")
end
end
context "with an empty tag" do
it "handles <strong>" do
expect(html_to_markdown("<strong></strong>")).to eq("")
expect(html_to_markdown("<strong> </strong>")).to eq("")
expect(html_to_markdown("Some<strong> </strong>text")).to eq("Some text")
expect(html_to_markdown("Some<strong> </strong>text")).to eq("Some text")
end
it "handles <em>" do
expect(html_to_markdown("<em></em>")).to eq("")
expect(html_to_markdown("<em> </em>")).to eq("")
expect(html_to_markdown("Some<em> </em>text")).to eq("Some text")
expect(html_to_markdown("Some<em> </em>text")).to eq("Some text")
end
end
context "with spaces around text" do
it "handles <strong>" do
expect(html_to_markdown("<strong> Bold</strong>")).to eq("**Bold**")
expect(html_to_markdown("<strong> Bold</strong>")).to eq("**Bold**")
expect(html_to_markdown("<strong>Bold </strong>")).to eq("**Bold**")
expect(html_to_markdown("<strong>Bold </strong>")).to eq("**Bold**")
expect(html_to_markdown("Some<strong> bold</strong> text")).to eq("Some **bold** text")
expect(html_to_markdown("Some<strong> bold</strong> text")).to eq("Some **bold** text")
expect(html_to_markdown("Some <strong>bold </strong>text")).to eq("Some **bold** text")
expect(html_to_markdown("Some <strong>bold </strong>text")).to eq("Some **bold** text")
end
it "handles <em>" do
expect(html_to_markdown("<em> Italic</em>")).to eq("*Italic*")
expect(html_to_markdown("<em> Italic</em>")).to eq("*Italic*")
expect(html_to_markdown("<em>Italic </em>")).to eq("*Italic*")
expect(html_to_markdown("<em>Italic </em>")).to eq("*Italic*")
expect(html_to_markdown("Some<em> italic</em> text")).to eq("Some *italic* text")
expect(html_to_markdown("Some<em> italic</em> text")).to eq("Some *italic* text")
expect(html_to_markdown("Some <em>italic </em>text")).to eq("Some *italic* text")
expect(html_to_markdown("Some <em>italic </em>text")).to eq("Some *italic* text")
end
end
it "supports <table>" do
html = <<~HTML
<table>
<thead>
<tr>
<th>This</th>
<th>is</th>
<th>the</th>
<th><i>headers</i></th>
</tr>
</thead>
<tbody>
<tr>
<td>I am</td>
<td>the</td>
<td><b>first</b></td>
<td>row</td>
</tr>
<tr>
<td>And this</td>
<td>is the</td>
<td>2<sup>nd</sup></td>
<td>line</td>
</tr>
</tbody>
</table>
HTML
markdown = <<~MD
| This | is | the | *headers* |
| - | - | - | - |
| I am | the | **first** | row |
| And this | is the | 2<sup>nd</sup> | line |
MD
expect(html_to_markdown(html)).to eq(markdown.strip)
expect(html_to_markdown("<table><tr><td>Hello</td><td>World</td></tr></table>")).to eq("| Hello | World |\n| - | - |")
end
it "doesn't swallow badly formatted <table>" do
html = <<~HTML
<table>
<tr>
<th>1</th>
<th>2</th>
<th>3</th>
<th>4</th>
</tr>
<tr>
<td>One</td>
<td>Two</td>
<td>Three</td>
</tr>
</table>
HTML
expect(html_to_markdown(html)).to eq("1 2 3 4 \nOne Two Three")
end
end