Creating shards

In this section, we'll view the blog posts we've ingested in the previous section, and make some improvements to our ingester. We'll add some customizations to the visual theme, and make sure videos are downloaded for offline viewing too. Then we will package the hatch for distribution, in an archive format called a "shard", and create a second shard that determines how our content is organized in the app.

We'll use Hatch Previewer to view the blog posts. If you haven't installed it yet, check how to do so in the Install section.

Viewing the blog posts

Running the ingester in the last section should have created a directory with a name like hatch_cc-blog_20180215_104946 (where the numbers are the date and time that the ingester was run.) Start Hatch Previewer, either by clicking its icon on the desktop, or from the command line:

flatpak run com.endlessm.HatchPreviewer

You'll be shown a dialog box where you select a directory. Navigate to the hatch directory and select it.

NOTE: You can also pass the hatch directory on the command line, but you have to give an absolute path due to a limitation in Flatpak.

You should see a mostly blank window with a list of the ingested blog posts on the left. Click on one, and you should see something like this:

Screenshot of Hatch Previewer

You can see the page as it looks when rendered in the article format, and press the "Flip" button to view the raw HTML. You can also see the metadata on the right-hand side.

Browse through the posts, and take note of anything that looks odd. You can also compare them to the posts on their original website in your browser, to make sure nothing is missing.

The first thing you might notice is that there are some blank spaces in the article format we used. These will display metadata, but haven't been filled in yet. There's a space after the dash at the top right, for the date that the post was published. This isn't pre-filled because we want to display it in whatever date format the user has selected in their settings, and that's impossible to know at this point.

There's also a space at the bottom of the post where the tags will go. We won't determine what tags will be used in the app until later, so there's nothing to display here yet.

Customizing the fonts and colors

Perhaps the most obvious thing to refine is the use of default fonts and colors in the rendered posts. We want to change these; in our case to better reflect the style of the original site.

The way we do this is with the set_custom_scss method of the BlogArticle format. Each article format has its own default CSS theme. You can customize certain variables in the theme, while largely sticking to the default.

Here's what the custom SCSS looks like for customizing the font:

$title-font: 'Source Sans Variable';
$body-font: 'Source Sans Variable';
$context-font: 'Source Sans Variable';
$support-font: 'Source Sans Variable';

Each of these variables governs a font used for a particular purpose in the rendered asset. In our case, we'll stick with Source Sans as the font, just as the original blog does.

We also want to customize the color scheme to look more like the original blog. Here's some SCSS to do that:

$primary-light-color: #fb7928;
$primary-medium-color: #ee5b32;
$accent-light-color: #049bce;
$accent-dark-color: #464646;
$background-light-color: white;
$background-dark-color: #e9e9e9;

Just as with the fonts, these are variables specific to the BlogArticle's default theme.

Putting it all together, we add the following lines to our ingester in the same place where we set all the metadata on the post asset:

postAsset.set_custom_scss(`
    $title-font: 'Source Sans Variable';
    $body-font: 'Source Sans Variable';
    $context-font: 'Source Sans Variable';
    $support-font: 'Source Sans Variable';
    $primary-light-color: #fb7928;
    $primary-medium-color: #ee5b32;
    $accent-light-color: #049bce;
    $accent-dark-color: #464646
    $background-light-color: white;
    $background-dark-color: #e9e9e9;
    @import '_default';
`);

The @import '_default'; brings in the default theme for BlogArticle. We can add custom SCSS rules to do further customization after the import line, but we're not going to do that in this example. You can also leave out the import line if you want to start from scratch and write your own theme.

Re-run the ingester, and view the hatch with Hatch Previewer; you should see the effects of our customizations in the rendered HTML, and also in the embedded CSS when you click the "Flip" button.

Fixing the missing images

Another thing you might notice while browsing through the hatch is that some images are not showing up. At the time of this walkthrough's writing, this happens in the articles "5 Awesome Organisations Working to Protect and Expand the Public Domain" and "Thank you to Private Internet Access, lead sponsor of CC’s Global Summit", though of course if you are doing this walkthrough later, it may happen in other articles, or not at all. (If you want to reproduce these results exactly, you can set your ingester to page back to a longer time ago.)

We use the "Flip" button in Hatch Previewer to figure out why this is happening. The raw HTML for the paragraph with the missing image looks like this:

<p>
  <a href="https://www.eff.org/copyrightweek">
    <img data-image-title="copyright_week_2018" data-image-description="" src="https://d15omoko64skxi.cloudfront.net/wp-content/uploads/2018/01/copyright_week_2018-1024x512.png" alt="" width="840" height="420" sizes="(max-width: 709px) 85vw, (max-width: 909px) 67vw, (max-width: 1362px) 62vw, 840px">
  </a>
  <i>
    <span>We’re taking part in </span>
  </i>
  <a href="https://www.eff.org/copyrightweek">
    <i>
      <span>Copyright Week</span>
    </i>
  </a>
  <i>
    <span>, a series of actions and discussions supporting key principles that should guide copyright policy. Every day this week, various groups are taking on different elements of the law, and addressing what’s at stake, and what we need to do to make sure that copyright promotes creativity and innovation.</span>
  </i>
</p>

The cause is apparent: the ingester isn't turning the image into an image asset. The image's location still points to an online address, which Hatch Previewer can't access because it runs sandboxed and doesn't have network privileges. Looking back at our ingester code, we only looked for images inside a <figure> tag.

The BlogArticle format works best with <figure> images anyway, and not inline <img> elements, so instead of changing our image-to-asset code to look for the latter, we'll transmute them into the former, using Cheerio to manipulate the DOM.

$('p img').each(function () {
    const figure = $('<figure></figure>');
    const enclosingPara = $(this).parents('p')[0];
    figure.append($(this));
    figure.insertBefore(enclosingPara);
});

Cleaning unused elements

We also see from the HTML example above that there are useless <span> elements that we can remove to save even more space. Doing too much of this kind of cleanup can yield diminishing returns, but in this case the fix is quick to do with Cheerio, so we will not refrain.

$('span').filter(function () {
    return Object.keys(this.attribs).length === 0;
}).each(function () {
    $(this).replaceWith($(this).html());
});

Downloading linked media

As a last tweak to our ingester, it seems that one of the articles we are ingesting has an embedded video: "Art and the Every Day with Mike Winkelmann (AKA beeple)". (Again, if you are doing this walkthrough later, that article might not be in the batch you ingested.) We want to download this video and save it as an asset as well.

NOTE: This works a bit differently from saving images as assets. The video doesn't get put into the hatch directly; instead, a placeholder is created, and the video is downloaded later in the next processing step. This keeps the hatch creation time quick. It also allows transcoding the video in order to use a free codec or a different resolution.

NOTE: This section nearly doubles the size of the ingester code and requires a Vimeo account. If you aren't interested in downloading videos, you can safely skip it, or save it for later.

As might be expected from the Creative Commons blog, the video has a license that makes it freely distributable, and it is available for download from Vimeo. We'll add some code to our ingester that looks for embedded videos and saves their download location into the hatch.

Run the following command to install the necessary packages:

npm install --save vimeo youtube-dl

We use the vimeo package to access the Vimeo API, in order to get the video's metadata, and to make sure that it is OK to download. The youtube-dl package is used to get a download URL for the video. (Despite its name, it's not only for YouTube.)

First of all, since we are accessing the Vimeo API, we'll need a Vimeo account and some credentials. Create an account if you don't have one, then go to https://developer.vimeo.com, log in, and click the "Create App" button in the top right corner. Fill in the required information and submit the form. After that, your newly-created app should show up under "My Apps". Click on it and then click on the "Authentication" tab to see the credentials. You need the client identifier and the client secret. Fill them into the appropriate places in the following code:

const {Vimeo} = require('vimeo');

const vimeoClientID = '(fill in client ID here)';
const vimeoClientSecret = '(fill in client secret here)';

const ensureVimeoClient = (function () {
    let vimeo;
    return async function ensureVimeoClient() {
        if (vimeo)
            return vimeo;

        vimeo = new Vimeo(vimeoClientID, vimeoClientSecret);
        vimeo.generateClientCredentials =
            util.promisify(vimeo.generateClientCredentials.bind(vimeo));
        vimeo.request = util.promisify(vimeo.request.bind(vimeo));

        const {access_token: accessToken} =
            await vimeo.generateClientCredentials(['public']);
        vimeo.setAccessToken(accessToken);
        return vimeo;
    };
})();

This makes an IIFE called ensureVimeoClient() with a private vimeo variable representing the Vimeo client object. We do it this way so that the ingester doesn't need to authenticate to Vimeo if there are no articles with videos to be ingested in the current batch.

Next, we note the structure of the embedded videos:

<div class="jetpack-video-wrapper">
  <div class="embed-vimeo">
    <iframe src="https://player.vimeo.com/video/..." ...></iframe>
  </div>
</div>

This whole thing is discarded by Fathom when cleaning up the DOM, so we need to add a Fathom rule that considers an embedded video part of the cluster of paragraph-like elements:

rule(dom('.jetpack-video-wrapper'), props(() => ({
    score: 100,
    note: {length: 1},
})).type('paragraphish')),

Then, we add some code to fix this up so that the videos are nicely in a <figure> element:

const videosToProcess = $('.jetpack-video-wrapper')
.map(function () {
    const iframe = $('.embed-vimeo iframe', this).first();
    let figure;
    if (iframe) {
        figure = $('<figure></figure>');
        figure.append(iframe);
        figure = figure.insertAfter(this);
    }
    $(this).remove();
    return figure;
})
.get().filter(figure => !!figure);

In videosToProcess we now have a list of <figure> elements each containing an <iframe> which refers to a video that should be downloaded.

In order to download each video, we first check the Vimeo API for all the metadata, such as the title and license of the video, and the tags that it is tagged with. We also use the Vimeo API to check that downloading is allowed. For performance, but also as a courtesy to Vimeo, we request only the fields that we actually use.

Subsequently, we use youtube-dl to get a download URL, and create a video asset using the libingester API function Libingester.util.get_embedded_video_asset().

This function replaces the <iframe> with a special placeholder element. Videos aren't actually downloaded in libingester. This is so that you can implement your own compression or transcoding if need be. For this reason, you can't play the video in Hatch Previewer. However, we'll download the video in the section below with Basin.

await Promise.all(videosToProcess.map(async figure => {
    const iframe = figure.find('iframe');
    const {host, pathname} = url.parse(iframe.attr('src'));
    if (host !== 'player.vimeo.com') {
        $(iframe).remove();
        return;
    }

    const [,, vimeoID] = pathname.split('/');
    const {
        description, license, link, name, pictures, privacy, tags,
        release_time: releaseTime,
        modified_time: modifiedTime,
    } = await getVideoInfo(vimeoID);

    const freeLicense = licenseFromVimeoLicense(license);
    if (!privacy.download || !freeLicense) {
        $(iframe).remove();
        return;
    }

    const {url: downloadURL} = await Youtubedl.getInfo(link, [
        '--prefer-free-formats',
        '--format=worst',
    ], {
        maxBuffer: 500 * 1024,  // JSON info is big!
    });
    const video = Libingester.util.get_embedded_video_asset(iframe,
        downloadURL);
    video.set_title(name);
    video.set_synopsis(description);
    video.set_canonical_uri(link);
    video.set_last_modified_date(modifiedTime);
    video.set_date_published(releaseTime);
    video.set_license(freeLicense);
    video.set_tags(tagsFromVimeoTags(tags));

    const posterFrame = pictures.sizes.pop();
    const poster = Libingester.util.download_image(posterFrame.link);
    video.set_thumbnail(poster);

    hatch.save_asset(video);
    hatch.save_asset(poster);
}));

The functions getVideoInfo(), licenseFromVimeoLicense(), and tagsFromVimeoTags() are not given here, but you can see them in the full code.

NOTE: Make sure to put this code after the code that cleans up the class attributes on all elements, because Libingester.util.get_embedded_video_asset() adds some new class attributes that are necessary in the final result.

If you run this code and view the result in Hatch Previewer, you should be able to see the article with the placeholders where the videos were.

Creating a shard with Basin

The tool that we use to turn a hatch into a shard is called Basin. Basin is available as part of the Endless SDK. The easiest way to use it is inside the Flatpak sandbox of the Endless SDK. If you haven't installed it yet, check how to do so in the Install section.

Basin is actually a suite of related tools for creating shards from content on disk, not only from hatches. The tool we'll use is called basin-hatch. It also downloads the videos that we didn't download in the previous section.

First start a command shell in the Flatpak sandbox and give it permission to access the internet (to download the videos) and the ingester directory (to write the shard):

flatpak run --command=bash --share=network --filesystem=$(pwd) com.endlessm.apps.Sdk

NOTE: Use exit to get out of the shell when you're done with it.

Next, create a file inside the hatch directory, called hatch_sets.json, with the following content:

{
    "ignore-unlisted": true
}

This file is needed because we want to categorize our content ourselves, instead of letting basin-hatch do it automatically based on the Wordpress tags from the ingester. We'll do this in the next section.

Then create a directory to hold the output, and run basin-hatch on the hatch directory:

mkdir -p content
basin-hatch hatch_cc-blog_20180226_082238 content

You should see the videos being downloaded, and finally the message Successfully created content/output.shard.

NOTE: If you included the code from the previous section in your ingester, you have to do this step quickly after creating the hatch. The Vimeo download links will expire after some time.

Looking at the content/ directory, there are actually two files in it:

  • output.shard is a compressed, binary version of the hatch with all the posts, images, videos, and their metadata.
  • manifest.json tells information about the output.shard file.

These files are the form in which the content is ultimately given to the user! We'll learn what to do with them in the next section of the walkthrough.

Determining the content structure

There's one last thing we have to do before being ready to use the content in an app. Most kinds of app experiences that we can build with the tools offered by the modular framework rely on organizing the content into some kind of categories. As promised in a previous part of the walkthrough, it's time to learn more about how categories are represented in this toolkit's database format.

From the user's point of view, there are only collections of content, which we call sets. Sets have metadata just like the assets that we've stored in the hatch so far: for example a title, or a thumbnail image. The biggest difference is that sets aren't stored in hatches and aren't ingested — instead, they are created by the app author. Sets can be displayed in many different ways depending on their metadata and the app's UI. Sets can also be part of other sets, and they can be marked as featured, although the meaning of featured depends on how the app's UI interprets it.

From the database's point of view, each asset (set, image, blog post, or other content) is tagged with any number of string identifiers. These identifiers are called tags, and aren't visible to the user.

Sets have a metadata property called childTags, which is an array of tag strings. Whichever tags are named in that property are considered to be part of the set. So, on the UI side of things, a set may contain articles from one or more tags, and tags may be in one or more sets, or not used in any set at all.

NOTE "Categories", "tags", and even "sets" are overloaded terms — they can mean a lot of different things in different contexts. The way they're used here is quite different from the way they're used in Wordpress, for example.

It's best to think of the modular framework's sets and tags as a starting point, not as an ending point or a particular categorization scheme that your app is forced into. As a tool, and combined with different UI modules, it's flexible enough to implement many different kinds of categorization schemes.

Set assets are usually provided in a separate shard from the rest of the content. We call this a permanent shard. (It's a common use case in apps to rotate content out in an update when it becomes old, and at the same time rotate new content in, and these shards are called in contrast ephemeral shards.)

The categorization scheme for our app is going to work like this:

  • We'll have sets in a sidebar menu in the app. These sets will be the most general ones, corresponding to what Wordpress calls "categories" (usually predefined for a whole blog, and maximum one for each post.) These are the sets we'll mark as featured.
  • Other sets, which we will not mark as featured, will not be directly clickable in a menu in the UI. These will correspond to what Wordpress calls "tags" (free-form, more than one for each post.) These sets will be accessible at the bottom of each blog post; you can click on it to see which other posts included that tag.

Now, in Hatch Previewer, we can scroll the left panel all the way down to see a list of tags, along with how many articles use them. (Remember when we first wrote the ingester, we prefixed the tags of the latter kind with tag: so we could tell them apart later.)

We'll merge featured sets with only one post into one generic "Blog Posts" set for the purposes of this walkthrough, and create a featured set from a few free-form tags that have lots of posts associated with them. Normally we would not do this, since we'd have more content and the number of posts in each set will change anyway as new content rotates in and out, but for this purpose it's nice not to have sets with lone articles in them.

The featured sets will be:

Set name Tag(s) Number of articles*
About Creative Commons About CC 13
Around the World Global affiliates, plus all the tags that mention a specific country, city, or language 11
Policy, Advocacy & Copyright Reform Policy / advocacy / copyright reform 9
Events Events 7
Blog Posts Education / OER, Open access, Weblog 4
Interviews tag:cctalkswith 3

* At least, in the hatch at the time this tutorial was written.

We'll also choose some of the free-form tags to be non-featured sets. For these it doesn't matter if they have only a lone post associated with them. They won't show up on the main menu, only as tags at the bottom of articles. For example, we'll create a "Copyright Reform" set covering the tags tag:copyright reform, tag:eu copyright reform, and tag:#fixcopyright.

Creating a permanent shard for the sets

We'll write a Basin manifest for the permanent shard and use Basin to create it. The manifest looks something like this:

{
    "sets": [
        {
            "title": "About Creative Commons",
            "featured": true,
            "tags": ["EknSetObject"],
            "childTags": [
                "About CC"
            ]
        },

        ...

        {
            "title": "Copyright Reform",
            "featured": false,
            "tags": ["EknSetObject"],
            "childTags": [
                "tag:#fixcopyright",
                "tag:copyright reform",
                "tag:eu copyright reform"
            ]
        },

        ...
    ]
}

(This shows one example each of a featured set and a non-featured set. EknSetObject is a "magic" tag that the frontend requires; for now it must be added manually.)

Using Basin to convert the manifest into a shard is simple:

basin sets.json sets.shard

Compare this to how we created a shard earlier. To create this shard, we'll use the Basin tool itself instead of basin-hatch. The basin-hatch tool uses Basin under the hood. The Basin manifest is the lowest-level description of what goes into a shard, and if you are writing more complicated apps you can write a tool similar to basin-hatch that converts from some higher-level format into a Basin manifest and uses Basin under the hood to create the shard.

Further remarks and ideas

You can create more than one shard for the content and more than one shard for the sets if you want to organize your app into multiple, separately-installable "content packs."

You can also associate images with your sets, but we'll skip that for the purposes of this section, since the UI we'll use in the next section doesn't have a space to display the sets' images.

Subpages

Full code

The full refined ingester code

The results of the search are