mirror of
https://github.com/discourse/discourse.git
synced 2024-12-13 17:34:17 +08:00
2d51833ca9
It used to insert block Oneboxes inside paragraphs which resulted in
invalid HTML. This needed an additional parsing for removal of empty
paragraphs and the resulting HTML could still be invalid.
This commit ensure that block Oneboxes are inserted correctly, by
splitting the paragraph containing the link and putting the block
between the two. Paragraphs left with nothing but whitespaces will
be removed.
Follow up to 7f3a30d79f
.
98 lines
4.6 KiB
Ruby
98 lines
4.6 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
require "rails_helper"
|
|
require "excerpt_parser"
|
|
|
|
describe ExcerptParser do
|
|
|
|
it "handles nested <details> blocks" do
|
|
html = <<~HTML.strip
|
|
<details>
|
|
<summary>
|
|
FOO</summary>
|
|
<details>
|
|
<summary>
|
|
BAR</summary>
|
|
<p>Lorem ipsum dolor sit amet, consectetur adipiscing elit. Fusce ultrices, ex bibendum vestibulum vestibulum, mi velit pulvinar risus, sed consequat eros libero in eros. Fusce luctus mattis mauris, vitae semper lorem sodales quis. Donec pellentesque lacus ac ante aliquam, tincidunt iaculis risus interdum. In ullamcorper cursus massa ut lacinia. Donec quis diam finibus, rutrum odio eu, maximus leo. Nulla facilisi. Nullam suscipit quam et bibendum sagittis. Praesent sollicitudin neque at luctus ornare. Maecenas tristique dapibus risus, ac dictum ipsum gravida aliquam. Phasellus vehicula eu arcu sed imperdiet. Vestibulum ornare eros a nisi faucibus vehicula. Quisque congue placerat nulla, nec finibus nulla ultrices vitae. Quisque ac mi sem. Curabitur eu porttitor justo. Etiam dignissim in orci iaculis congue. Donec tempus cursus orci, a placerat elit varius nec.</p>
|
|
</details>
|
|
</details>
|
|
HTML
|
|
|
|
expect(ExcerptParser.get_excerpt(html, 50, {})).to match_html <<~HTML
|
|
<details><summary>FOO</summary>BAR
|
|
Lorem ipsum dolor sit amet, consectetur adi…</details>
|
|
HTML
|
|
|
|
expect(ExcerptParser.get_excerpt(html, 6, {})).to match_html('<details><summary>FOO</summary>BAR…</details>')
|
|
expect(ExcerptParser.get_excerpt(html, 3, {})).to match_html('<details class="disabled"><summary>FOO</summary></details>')
|
|
end
|
|
|
|
it "respects length parameter for <details> block" do
|
|
html = '<details><summary>foo</summary><p>bar</p></details>'
|
|
expect(ExcerptParser.get_excerpt(html, 100, {})).to match_html('<details><summary>foo</summary>bar</details>')
|
|
expect(ExcerptParser.get_excerpt(html, 5, {})).to match_html('<details><summary>foo</summary>ba…</details>')
|
|
expect(ExcerptParser.get_excerpt(html, 3, {})).to match_html('<details class="disabled"><summary>foo</summary></details>')
|
|
expect(ExcerptParser.get_excerpt(html, 2, {})).to match_html('<details class="disabled"><summary>fo…</summary></details>')
|
|
end
|
|
|
|
describe "keep_onebox_body parameter" do
|
|
it "keeps the body content for external oneboxes" do
|
|
html = <<~HTML.strip
|
|
<aside class="onebox">
|
|
<header class="source">
|
|
<img src="https://github.githubassets.com/favicon.ico" class="site-icon" width="32" height="32">
|
|
<a href="https://github.com/discourse/discourse" target="_blank">GitHub</a>
|
|
</header>
|
|
<article class="onebox-body">
|
|
<img src="/uploads/default/original/1X/10c0f1565ee5b6ca3fe43f3183529bc0afd26003.jpeg" class="thumbnail">
|
|
<h3>
|
|
<a href="https://github.com/discourse/discourse" target="_blank">discourse/discourse</a>
|
|
</h3>
|
|
<p>A platform for community discussion. Free, open, simple. - discourse/discourse</p>
|
|
</article>
|
|
</aside>
|
|
HTML
|
|
expect(ExcerptParser.get_excerpt(html, 100, keep_onebox_body: true)).to eq(<<~HTML.strip)
|
|
[image]
|
|
|
|
<a href="https://github.com/discourse/discourse" target="_blank">discourse/discourse</a>
|
|
|
|
A platform for community discussion. Free, o…
|
|
HTML
|
|
end
|
|
|
|
it "keeps the content for internal oneboxes" do
|
|
html = <<~HTML.strip
|
|
<aside class="quote" data-post="1" data-topic="8">
|
|
<div class="title">
|
|
<div class="quote-controls"></div>
|
|
<img width="20" height="20" src="/user_avatar/localhost/system/40/2_2.png" class="avatar">
|
|
<a href="/t/welcome-to-discourse/8/1">Welcome to Discourse</a>
|
|
</div>
|
|
<blockquote>The first paragraph of this pinned topic will be visible as a welcome message to all new visitors on your homepage.</blockquote>
|
|
</aside>
|
|
HTML
|
|
expect(ExcerptParser.get_excerpt(html, 100, keep_onebox_body: true)).to eq(<<~HTML.strip)
|
|
[image]
|
|
|
|
<a href="/t/welcome-to-discourse/8/1">Welcome to Discourse</a>
|
|
|
|
The first paragraph of this pinned topic will be …
|
|
HTML
|
|
end
|
|
end
|
|
|
|
describe "keep_quotes parameter" do
|
|
it "should keep the quoted content in html" do
|
|
html = <<~HTML.strip
|
|
<aside class="quote">
|
|
<blockquote>
|
|
This is a quoted text.
|
|
</blockquote>
|
|
</aside>
|
|
HTML
|
|
expect(ExcerptParser.get_excerpt(html, 100, keep_quotes: true)).to eq("This is a quoted text.")
|
|
end
|
|
end
|
|
end
|