Leveraging Holistic Synergies

Mulitlingual Django Addendum

Django Addendum is a little library I wrote a couple years back to solve a recurring problem: making copy snippets on a web page editable. More specifically, making arbitrary blocks of text in Django templates admin editable regardless of the template's application source.

It's pretty simple: you tag a block of text using the snippet template tag - this is the default text - and then by using the snippet name you can add updated copy via the admin. The template tag does the lookup. It's designed to work with caching enabled so that after the first cache miss - if there is one

A few people requested multilingual support and it became apparent that this would be really valuable to quite a few people. I put it off long enough and after some investigation into the alternatives decided to draft up a new release.

Model translations

The obvious strategy was to integrate with an existing Django model translation app. I had previous and fairly happy experience with the fittingly named django-modeltranslation package so looked there first. However based on my memory and a fresh review it seems a better fit for direct project integration. It works by adding new attributes (i.e. table columns) for each translated language of each targeted content attribute (table column). As you might guess this requires migrations, project specific migrations. This is a non-starter for a reuasable app.

I also took at a look at django-hvad which uses a different strategy to accomplish a similar goal. Regardless of schema migration issues this package would require subclassing a custom meta model class for the core models, which means requiring the hvad package with Django Addendum. Using either one of these of course means making a decision about what model translation library Django Addendum users should use on the rest of their sites, or at least add an additional requirements.

At this point is just seemed more straightforward to add in some translations directly.

Model and table strategy

There are three ways to go about adding the translations:

  1. Allow the addition of language specific attributes/columns (e.g. text_en, text_es)
  2. Create language specific rows/instances using a language attribute/column
  3. Add a distinct model/table for translations

The first would be my choice for a totally project-managed app, but for a reusable app this goes out the window for reasons stated above.

In comparison to the third strategy the second has a lot going for it: it's simple and especially in the context of the Django ORM would require fewer queries (pretty hard to run a UNION query in the ORM). However I really wanted to be able to use a default snippet and to be able to edit all the content for a given snippet on one page.

The Django admin's inline admin classes make editing related content on one page a cinch, so it made sense to add the new model. There's a side benefit for non-i18n users as well - it makes it easy to hide this functionality if you don't need it.

Modifying the cache strategy

The next component was the cache. Caching is very important for how Django Addendum works. Performance in general is a good reason, but also it just feels dirty initiating database calls from a template.

The previous release used a manager method cache lookup that returned a Snippet instance from cache, a snippet instance from the database (after updating the cache), or None. The new version needs to account for multiple languages in a different model as well as a fallback snippet (the default).

The first strategy I tried was adding each snippet-language combination to the cache using its own key. The downside to this is the cache miss scenario. Worst case: language specific cache miss, language specific database miss, default snippet cache miss, default snippet database miss. In the scenario of a missing translation for a present snippet that still means an aniticpated cache miss every single time to check for the translation.

I'm quite opposed to inserting unnessary backend service calls in other people's websites.

The updated content caching strategy consists of caching all of a snippet's content - translations included - in a single language-keyed dictionary for a given snippet. For a given snippet a missed translation is simply handled by a default dictionary lookup for the given snippet.

Moving away from caching Snippet objects means deprecating the manager class, which was used only for getting cached snippets. Cache management is handled by two functions - a getter and setter, respectively - including an update function called on save by both the Snippet model and the SnippetTranslation model, ensuring the cached content is always up to date.

Creating an upgrade path

Adding translation support is now as simple as enabling internationalization in your Django project - and writing snippet translations, of course.

To ensure the new cache format plays nice with existing installations a new management command is available to update all snippets. And as mentioned earlier, if i18n is disabled in your project, you won't see any of this. The SnippetTranslation class is not exposed in the Django admin except through an inline, and that is condtionally hidden based on the i18n setting status in the project. Monolingual sites won't change, cache excepted.

And if anybody has any advice on fixing PyPI README formatting I'd love to hear it!

Originally published 2014-11-15