Dec. 22, 2019

Django tips for real life applications

Summary

Real life django applications often have requirements that are not well advertised through mostly django tutorials around the internet. Here I share a few tips that may help you build a better django app.

Using UUIDs instead of IDs to reference objects.

Let’s say for example that you are building an app that are going to disrupt the finance market, maybe it is bitcoins, or you are going to fix the banking system (please do it).

Your user logIn, do a transaction and see something like this:

http://financeapp.com/transaction/37/

The user will realize that so far only 37 transactions where made on your application, and to be honest, he will probably not care about that, but depending of your audience, hiding this information could be useful. It’s also one information that you may want hide from current competitors or even hide from the press if they have an eye on you.

The standard way to hide those sequential ID’s is replacing them by an UUID, something like this:

http://financeapp.com/transaction/e12616c3-d81d-47a4-b01a-30d785be5251/

The code necessary to achieve that in django is really simple and almost self explanatory:

# urls.py
urlpatterns = [
    path('admin/', admin.site.urls),
    path('transaction/<uuid:slug>/', TransactionDetailView.as_view()),
    path('transaction/<int:pk>/', TransactionDetailView.as_view()),
]

# views.py
class TransactionDetailView(DetailView):
    model = Transaction
    
# models.py
class Transaction(models.Model):
    slug = models.UUIDField(default=uuid.uuid4, db_index=True)
    amount = models.IntegerField(default=1000)

Generally speaking, class based views that accept a pk as parameter can accept instead a parameter named slug, and this slug parameter can be of any arbitrary type that you want (usually the UUID type).[1]1

As you probably have noted I left the access though regular ID too, but just to empathize that django accept both ways, you definitely don’t want to keep both ways of access in your app.

Limiting the edit/update/delete to the object’s owner.

One basic security question that you should address in your new application in the object level access permission, on a basic scale this usually revolves around only allowing the user to see, edit and delete objects that he is the rightful owner. Along the years I have seem many suggestions on how to do that on CBV, they always felt strange and the reason is, while technically correct they are missing the fact that django have a simpler way of achieving that behavior.

Let’s first say that we want to only display the Transactions that belong to the user:

class TransactionListView(ListView):
    models = Transaction
    
    def get_queryset():
        return Transaction.objects.filter(user=self.request.user)

Straight forward, simple, elegant and well know… Now let’s do the same for delete, hold your jaw:

class TransactionDeleteView(DeleteView):
    models = Transaction
    
    def get_queryset():
        return Transaction.objects.filter(user=self.request.user)

And again for update:

class TransactionUpdateView(UpdateView):
    models = Transaction
    
    def get_queryset():
        return Transaction.objects.filter(user=self.request.user)

And again for Detail:

class TransactionDetailView(DetailView):
    models = Transaction
    
    def get_queryset():
        return Transaction.objects.filter(user=self.request.user)

That is it! You don’t have to weirdly override the dispatch, get_object or any other strange way you have found in the internet…

It took some time for me to realize that… I guess that whenever I saw the get_queryset in a view for object level access I thought: “That’s weird, why there is a queryset method in an view that works on a single object”, but the fact is, this abstraction is converted to pure SQL and in SQL you don’t have a get_object or get_queryset it will always be a SELECT statement.

The get_queryset here is working as a pre-filter for the get_object and is ensuring that your actions only happens if the query is matching [2]2. If this is still looking weird, I promess you that soon this feeling will go away. And if it doesn’t, at least the weirdness will be living in an abstraction of the framework instead of in weird overrides that no one can agree about.

Rendering variables from CBV methods.

This one I learned late, less than two year ago… I don’t know if it’s actually a relatively new feature from django or if it was just hidding in the documentation, but it was a game changer for me.

Whenever you need to expose some variables to templates the most common way is simply adding something to the get_context_data:

class TransactionDeleteView(DeleteView):
    models = Transaction
    
    def get_queryset():
        return Transaction.objects.filter(user=self.request.user)

    def get_context_data(self, **kwargs):
        data = super().get_context_data(**kwargs)
        data['something_you_want_to_display_on_delete_template'] = '42'
        return data

There is a simpler way of doing that, a way that can help you have more organized code. Instead of packing a lot of variables and calls to the get_context_data you can create methods and access them in templates, like this:

class TransactionDeleteView(DeleteView):
    models = Transaction
    
    def get_queryset():
        return Transaction.objects.filter(user=self.request.user)

    def something_you_want_to_display_on_delete_template(self)
        return "42"

In your template:

<h1> Are you sure about deleting {{ object }} </h1>
<form>
    <buttom></button>
</form> 
<h2> Here I will display other relevant information to let you make a concious choice</h2>
<p> {{ view.something_you_want_to_display_on_delete_template }} </p>

Cache expensive computations that you need to access more than once per request.

Django have a builtin decorator @cached_property for caching short lived properties, the cache will live the same amount of time of the instance holding the property. As soon as the instance dies the cache dies. You often need this when accessing the same resource twice in your template.

You can put the decorator at any level, it can be on a view property:

# views.py
class TransactionDetailView(DetailView):
    models = Transaction
    
    @cached_property
    def some_expensive_computation_from_view()
        return magic_that_will_return_42()

Or even better(?), it can save you some queries to database:

# models.py
class Transaction(models.Model):
    slug = models.UUIDField(default=uuid.uuid4, db_index=True)
    amount = models.IntegerField(default=1000)
    
    @cached_property    
    def some_expensive_queryset_from_model()
        return ExpensiveQueryset.filter().all()

To be honest, using it at Model level could get you in situations where you are going to reuse the method in other parts of your code and get cached when it shouldn’t. Whenever you can, keep the @cached_property at View level.

For completeness let’s see a fake example of @cached_property in action, the first example will take 4 seconds to render:

# views.py
class TransactionDetailView(DetailView):
    models = Transaction
    
    def correct_answer()
        time.sleep(2)
        return 42
        
# transaction_detail.html
{% if view.correct_answer == 42 %}
    {{ view.correct_answer }}
{% endif %}

Now with the decorator it will take 2 seconds to render:

# views.py
class TransactionDetailView(DetailView):
    models = Transaction
    
    @cached_property
    def correct_answer()
        time.sleep(2)
        return 42
        
# transaction_detail.html
{% if view.correct_answer == 42 %}
    {{ view.correct_answer }}
{% endif %}

Even with us calling the view.correct_answer twice in our template, it will actually be executed only once during the conditional evaluation. The second call where we are printing the result will not happen, only the cached result will be returned.

Final words

Each one of the tips presented here can carry some trade offs and particularities. While in some cases not using those features can be a conscious choice, trying to replicate some of them from scratch will probably lead you to a worse or at least more laborious code base.