Feed entries are usually ordered from most to least recent.
Processing older entries first ensures that their creation timestamp
is lower than that of newer entries.
This is useful when we order by creation, because then we get a
consistent timeline.
* in some cases, what the scraper got is only a landing page, user can use scraper rules to extract the link of the landing page and follow it
* it also fix the wrong scrape rule apply when the server redirects it to another host
Trim space around CDATA elements before extracting the CharData.
This problem was discovered when reading https://www.sethvargo.com/feed.xml.
Title and Summary fields have newlines and space between the <title>
element and the CDATA element. e.g.
<title>
<![CDATA[Entry title here]]>
</title>
This meant the title of the feed was coming into MiniFlux as,
<![CDATA[Entry title here]]>
Some articles (especially the recent year-in-review ones) include a Youtube
video. The server-side rendered articles do not include the Youtube iframe,
but they do have a script that looks like
<script type="text/javascript" data-reactid="6">
window.__APOLLO_STATE__ = {
...
youtube_id: "9uASADiYe_8",
We add a reformatting function that tries to detect obvious JavaScript code
that has a field or variable called youtube_id that has an 11-character
double-quoted value, and adds the referenced Youtube videos in the beginning of
the article. This is slightly more general than needed for Quanta, in the hope
that it could be useful for similar sites.
This is a somewhat complex React site so the rules could be a little fragile.
Text content seems to be always inside .outer--content, and most h6 elements
are fluff like "read later" or pointers to other articles. However, h6.byline
and h6.post__title__kicker are relevant to the current article.
Figure captions are sometimes inside both figure and div.outer--content
elements, sometimes only inside figure, so take both and remove the
intersection.
The figure elements sometimes contain multiple copies of images or
videos, and we just take them all. Math articles seem to use Mathjax,
which we don't add.
