Django 1.5 release notes

February 26, 2013

Welcome to Django 1.5!

These release notes cover the new features, as well as some backwards incompatible changes you’ll want to be aware of when upgrading from Django 1.4 or older versions. We’ve also dropped some features, which are detailed in our deprecation plan, and we’ve begun the deprecation process for some features.

Overview

The biggest new feature in Django 1.5 is the configurable User model. Before Django 1.5, applications that wanted to use Django’s auth framework (django.contrib.auth) were forced to use Django’s definition of a “user”. In Django 1.5, you can now swap out the User model for one that you write yourself. This could be a simple extension to the existing User model – for example, you could add a Twitter or Facebook ID field – or you could completely replace the User with one totally customized for your site.

Django 1.5 is also the first release with Python 3 support! We’re labeling this support “experimental” because we don’t yet consider it production-ready, but everything’s in place for you to start porting your apps to Python 3. Our next release, Django 1.6, will support Python 3 without reservations.

Other notable new features in Django 1.5 include:

Wherever possible we try to introduce new features in a backwards-compatible manner per our API stability policy. However, as with previous releases, Django 1.5 ships with some minor backwards incompatible changes; people upgrading from previous versions of Django should read that list carefully.

One deprecated feature worth noting is the shift to “new-style” url tag. Prior to Django 1.3, syntax like {% url myview %} was interpreted incorrectly (Django considered "myview" to be a literal name of a view, not a template variable named myview). Django 1.3 and above introduced the {% load url from future %} syntax to bring in the corrected behavior where myview was seen as a variable.

The upshot of this is that if you are not using {% load url from future %} in your templates, you’ll need to change tags like {% url myview %} to {% url "myview" %}. If you were using {% load url from future %} you can simply remove that line under Django 1.5

Python compatibility

Django 1.5 requires Python 2.6.5 or above, though we highly recommend Python 2.7.3 or above. Support for Python 2.5 and below has been dropped.

This change should affect only a small number of Django users, as most operating-system vendors today are shipping Python 2.6 or newer as their default version. If you’re still using Python 2.5, however, you’ll need to stick to Django 1.4 until you can upgrade your Python version. Per our support policy, Django 1.4 will continue to receive security support until the release of Django 1.6.

Django 1.5 does not run on a Jython final release, because Jython’s latest release doesn’t currently support Python 2.6. However, Jython currently does offer an alpha release featuring 2.7 support, and Django 1.5 supports that alpha release.

Python 3 support

Django 1.5 introduces support for Python 3 - specifically, Python 3.2 and above. This comes in the form of a single codebase; you don’t need to install a different version of Django on Python 3. This means that you can write applications targeted for just Python 2, just Python 3, or single applications that support both platforms.

However, we’re labeling this support “experimental” for now: although it’s received extensive testing via our automated test suite, it’s received very little real-world testing. We’ve done our best to eliminate bugs, but we can’t be sure we covered all possible uses of Django.

Some features of Django aren’t available because they depend on third-party software that hasn’t been ported to Python 3 yet, including:

Further, Django’s more than a web framework; it’s an ecosystem of pluggable components. At this point, very few third-party applications have been ported to Python 3, so it’s unlikely that a real-world application will have all its dependencies satisfied under Python 3.

Thus, we’re recommending that Django 1.5 not be used in production under Python 3. Instead, use this opportunity to begin porting applications to Python 3. If you’re an author of a pluggable component, we encourage you to start porting now.

We plan to offer first-class, production-ready support for Python 3 in our next release, Django 1.6.

What’s new in Django 1.5

Configurable User model

In Django 1.5, you can now use your own model as the store for user-related data. If your project needs a username with more than 30 characters, or if you want to store user’s names in a format other than first name/last name, or you want to put custom profile information onto your User object, you can now do so.

If you have a third-party reusable application that references the User model, you may need to make some changes to the way you reference User instances. You should also document any specific features of the User model that your application relies upon.

See the documentation on custom user models for more details.

Support for saving a subset of model’s fields

The method Model.save() has a new keyword argument update_fields. By using this argument it is possible to save only a select list of model’s fields. This can be useful for performance reasons or when trying to avoid overwriting concurrent changes.

Deferred instances (those loaded by .only() or .defer()) will automatically save just the loaded fields. If any field is set manually after load, that field will also get updated on save.

See the Model.save() documentation for more details.

When traversing relations, the ORM will avoid re-fetching objects that were previously loaded. For example, with the tutorial’s models:

  1. >>> first_poll = Poll.objects.all()[0]
  2. >>> first_choice = first_poll.choice_set.all()[0]
  3. >>> first_choice.poll is first_poll
  4. True

In Django 1.5, the third line no longer triggers a new SQL query to fetch first_choice.poll; it was set by the second line.

For one-to-one relationships, both sides can be cached. For many-to-one relationships, only the single side of the relationship can be cached. This is particularly helpful in combination with prefetch_related.

Explicit support for streaming responses

Before Django 1.5, it was possible to create a streaming response by passing an iterator to HttpResponse. But this was unreliable: any middleware that accessed the content attribute would consume the iterator prematurely.

You can now explicitly generate a streaming response with the new StreamingHttpResponse class. This class exposes a streaming_content attribute which is an iterator.

Since StreamingHttpResponse does not have a content attribute, middleware that needs access to the response content must test for streaming responses and behave accordingly.

{% verbatim %} template tag

To make it easier to deal with JavaScript templates which collide with Django’s syntax, you can now use the verbatim block tag to avoid parsing the tag’s content.

Retrieval of ContentType instances associated with proxy models

The methods ContentTypeManager.get_for_model() and ContentTypeManager.get_for_models() have a new keyword argument – respectively for_concrete_model and for_concrete_models. By passing False using this argument it is now possible to retrieve the ContentType associated with proxy models.

New view variable in class-based views context

In all generic class-based views (or any class-based view inheriting from ContextMixin), the context dictionary contains a view variable that points to the View instance.

GeoDjango

New tutorials

Additions to the docs include a revamped Tutorial 3 and a new tutorial on testing. A new section, “Advanced Tutorials”, offers How to write reusable apps as well as a step-by-step guide for new contributors in Writing your first patch for Django.

Minor features

Django 1.5 also includes several smaller improvements worth noting:

  • The template engine now interprets True, False and None as the corresponding Python objects.

  • django.utils.timezone provides a helper for converting aware datetimes between time zones. See localtime().

  • The generic views support OPTIONS requests.

  • Management commands do not raise SystemExit any more when called by code from call_command(). Any exception raised by the command (mostly CommandError) is propagated.

    Moreover, when you output errors or messages in your custom commands, you should now use self.stdout.write('message') and self.stderr.write('error') (see the note on management commands output).

  • The dumpdata management command outputs one row at a time, preventing out-of-memory errors when dumping large datasets.

  • In the localflavor for Canada, pq was added to the acceptable codes for Quebec. It’s an old abbreviation.

  • The receiver decorator is now able to connect to more than one signal by supplying a list of signals.

  • In the admin, you can now filter users by groups which they are members of.

  • QuerySet.bulk_create() now has a batch_size argument. By default the batch_size is unlimited except for SQLite where single batch is limited so that 999 parameters per query isn’t exceeded.

  • The LOGIN_URL and LOGIN_REDIRECT_URL settings now also accept view function names and named URL patterns. This allows you to reduce configuration duplication. More information can be found in the login_required() documentation.

  • Django now provides a mod_wsgi auth handler.

  • The QuerySet.delete() and Model.delete() can now take fast-path in some cases. The fast-path allows for less queries and less objects fetched into memory. See QuerySet.delete() for details.

  • An instance of ResolverMatch is stored on the request as resolver_match.

  • By default, all logging messages reaching the django logger when DEBUG is True are sent to the console (unless you redefine the logger in your LOGGING setting).

  • When using RequestContext, it is now possible to look up permissions by using {% if 'someapp.someperm' in perms %} in templates.

  • It’s not required any more to have 404.html and 500.html templates in the root templates directory. Django will output some basic error messages for both situations when those templates are not found. It’s still recommended as good practice to provide those templates in order to present pretty error pages to the user.

  • django.contrib.auth provides a new signal that is emitted whenever a user fails to login successfully. See user_login_failed

  • The new loaddata —ignorenonexistent option ignore data for fields that no longer exist.

  • assertXMLEqual() and assertXMLNotEqual() new assertions allow you to test equality for XML content at a semantic level, without caring for syntax differences (spaces, attribute order, etc.).

  • RemoteUserMiddleware now forces logout when the REMOTE_USER header disappears during the same browser session.

  • The cache-based session backend can store session data in a non-default cache.

  • Multi-column indexes can now be created on models. Read the index_together documentation for more information.

  • During Django’s logging configuration verbose Deprecation warnings are enabled and warnings are captured into the logging system. Logged warnings are routed through the console logging handler, which by default requires DEBUG to be True for output to be generated. The result is that DeprecationWarnings should be printed to the console in development environments the way they have been in Python versions < 2.7.

  • The API for django.contrib.admin.ModelAdmin.message_user() method has been modified to accept additional arguments adding capabilities similar to django.contrib.messages.add_message(). This is useful for generating error messages from admin actions.

  • The admin’s list filters can now be customized per-request thanks to the new django.contrib.admin.ModelAdmin.get_list_filter() method.

Backwards incompatible changes in 1.5

Warning

In addition to the changes outlined in this section, be sure to review the deprecation plan for any features that have been removed. If you haven’t updated your code within the deprecation timeline for a given feature, its removal may appear as a backwards incompatible change.

ALLOWED_HOSTS required in production

The new ALLOWED_HOSTS setting validates the request’s Host header and protects against host-poisoning attacks. This setting is now required whenever DEBUG is False, or else django.http.HttpRequest.get_host() will raise SuspiciousOperation. For more details see the full documentation for the new setting.

Managers on abstract models

Abstract models are able to define a custom manager, and that manager will be inherited by any concrete models extending the abstract model. However, if you try to use the abstract model to call a method on the manager, an exception will now be raised. Previously, the call would have been permitted, but would have failed as soon as any database operation was attempted (usually with a “table does not exist” error from the database).

If you have functionality on a manager that you have been invoking using the abstract class, you should migrate that logic to a Python staticmethod or classmethod on the abstract class.

Context in year archive class-based views

For consistency with the other date-based generic views, YearArchiveView now passes year in the context as a datetime.date rather than a string. If you are using {{ year }} in your templates, you must replace it with {{ year|date:"Y" }}.

next_year and previous_year were also added in the context. They are calculated according to allow_empty and allow_future.

Context in year and month archive class-based views

YearArchiveView and MonthArchiveView were documented to provide a date_list sorted in ascending order in the context, like their function-based predecessors, but it actually was in descending order. In 1.5, the documented order was restored. You may want to add (or remove) the reversed keyword when you’re iterating on date_list in a template:

  1. {% for date in date_list reversed %}

ArchiveIndexView still provides a date_list in descending order.

Context in TemplateView

For consistency with the design of the other generic views, TemplateView no longer passes a params dictionary into the context, instead passing the variables from the URLconf directly into the context.

Non-form data in HTTP requests

request.POST will no longer include data posted via HTTP requests with non form-specific content-types in the header. In prior versions, data posted with content-types other than multipart/form-data or application/x-www-form-urlencoded would still end up represented in the request.POST attribute. Developers wishing to access the raw POST data for these cases, should use the request.body attribute instead.

request_finished signal

Django used to send the request_finished signal as soon as the view function returned a response. This interacted badly with streaming responses that delay content generation.

This signal is now sent after the content is fully consumed by the WSGI gateway. This might be backwards incompatible if you rely on the signal being fired before sending the response content to the client. If you do, you should consider using middleware instead.

Note

Some WSGI servers and middleware do not always call close on the response object after handling a request, most notably uWSGI prior to 1.2.6 and Sentry’s error reporting middleware up to 2.0.7. In those cases the request_finished signal isn’t sent at all. This can result in idle connections to database and memcache servers.

OPTIONS, PUT and DELETE requests in the test client

Unlike GET and POST, these HTTP methods aren’t implemented by web browsers. Rather, they’re used in APIs, which transfer data in various formats such as JSON or XML. Since such requests may contain arbitrary data, Django doesn’t attempt to decode their body.

However, the test client used to build a query string for OPTIONS and DELETE requests like for GET, and a request body for PUT requests like for POST. This encoding was arbitrary and inconsistent with Django’s behavior when it receives the requests, so it was removed in Django 1.5.

If you were using the data parameter in an OPTIONS or a DELETE request, you must convert it to a query string and append it to the path parameter.

If you were using the data parameter in a PUT request without a content_type, you must encode your data before passing it to the test client and set the content_type argument.

System version of simplejson no longer used

As explained below, Django 1.5 deprecates django.utils.simplejson in favor of Python 2.6’s built-in json module. In theory, this change is harmless. Unfortunately, because of incompatibilities between versions of simplejson, it may trigger errors in some circumstances.

JSON-related features in Django 1.4 always used django.utils.simplejson. This module was actually:

  • A system version of simplejson, if one was available (i.e. import simplejson works), if it was more recent than Django’s built-in copy or it had the C speedups, or
  • The json module from the standard library, if it was available (i.e. Python 2.6 or greater), or
  • A built-in copy of version 2.0.7 of simplejson.

In Django 1.5, those features use Python’s json module, which is based on version 2.0.9 of simplejson.

There are no known incompatibilities between Django’s copy of version 2.0.7 and Python’s copy of version 2.0.9. However, there are some incompatibilities between other versions of simplejson:

  • While the simplejson API is documented as always returning Unicode strings, the optional C implementation can return a bytestring. This was fixed in Python 2.7.
  • simplejson.JSONEncoder gained a namedtuple_as_object keyword argument in version 2.2.

More information on these incompatibilities is available in ticket #18023.

The net result is that, if you have installed simplejson and your code uses Django’s serialization internals directly – for instance django.core.serializers.json.DjangoJSONEncoder, the switch from simplejson to json could break your code. (In general, changes to internals aren’t documented; we’re making an exception here.)

At this point, the maintainers of Django believe that using json from the standard library offers the strongest guarantee of backwards-compatibility. They recommend to use it from now on.

String types of hasher method parameters

If you have written a custom password hasher, your encode(), verify() or safe_summary() methods should accept Unicode parameters (password, salt or encoded). If any of the hashing methods need bytestrings, you can use the force_bytes() utility to encode the strings.

Validation of previous_page_number and next_page_number

When using object pagination, the previous_page_number() and next_page_number() methods of the Page object did not check if the returned number was inside the existing page range. It does check it now and raises an InvalidPage exception when the number is either too low or too high.

Behavior of autocommit database option on PostgreSQL changed

PostgreSQL’s autocommit option didn’t work as advertised previously. It did work for single transaction block, but after the first block was left the autocommit behavior was never restored. This bug is now fixed in 1.5. While this is only a bug fix, it is worth checking your applications behavior if you are using PostgreSQL together with the autocommit option.

Session not saved on 500 responses

Django’s session middleware will skip saving the session data if the response’s status code is 500.

Email checks on failed admin login

Prior to Django 1.5, if you attempted to log into the admin interface and mistakenly used your email address instead of your username, the admin interface would provide a warning advising that your email address was not your username. In Django 1.5, the introduction of custom user models has required the removal of this warning. This doesn’t change the login behavior of the admin site; it only affects the warning message that is displayed under one particular mode of login failure.

Changes in tests execution

Some changes have been introduced in the execution of tests that might be backward-incompatible for some testing setups:

Database flushing in django.test.TransactionTestCase

Previously, the test database was truncated before each test run in a TransactionTestCase.

In order to be able to run unit tests in any order and to make sure they are always isolated from each other, TransactionTestCase will now reset the database after each test run instead.

No more implicit DB sequences reset

TransactionTestCase tests used to reset primary key sequences automatically together with the database flushing actions described above.

This has been changed so no sequences are implicitly reset. This can cause TransactionTestCase tests that depend on hard-coded primary key values to break.

The new reset_sequences attribute can be used to force the old behavior for TransactionTestCase that might need it.

Ordering of tests

In order to make sure all TestCase code starts with a clean database, tests are now executed in the following order:

  • First, all unit tests (including unittest.TestCase, SimpleTestCase, TestCase and TransactionTestCase) are run with no particular ordering guaranteed nor enforced among them.
  • Then any other tests (e.g. doctests) that may alter the database without restoring it to its original state are run.

This should not cause any problems unless you have existing doctests which assume a TransactionTestCase executed earlier left some database state behind or unit tests that rely on some form of state being preserved after the execution of other tests. Such tests are already very fragile, and must now be changed to be able to run independently.

cleaned_data dictionary kept for invalid forms

The cleaned_data dictionary is now always present after form validation. When the form doesn’t validate, it contains only the fields that passed validation. You should test the success of the validation with the is_valid() method and not with the presence or absence of the cleaned_data attribute on the form.

Behavior of syncdb with multiple databases

syncdb now queries the database routers to determine if content types (when contenttypes is enabled) and permissions (when auth is enabled) should be created in the target database. Previously, it created them in the default database, even when another database was specified with the --database option.

If you use syncdb on multiple databases, you should ensure that your routers allow synchronizing content types and permissions to only one of them. See the docs on the behavior of contrib apps with multiple databases for more information.

XML deserializer will not parse documents with a DTD

In order to prevent exposure to denial-of-service attacks related to external entity references and entity expansion, the XML model deserializer now refuses to parse XML documents containing a DTD (DOCTYPE definition). Since the XML serializer does not output a DTD, this will not impact typical usage, only cases where custom-created XML documents are passed to Django’s model deserializer.

Formsets default max_num

A (default) value of None for the max_num argument to a formset factory no longer defaults to allowing any number of forms in the formset. Instead, in order to prevent memory-exhaustion attacks, it now defaults to a limit of 1000 forms. This limit can be raised by explicitly setting a higher value for max_num.

Miscellaneous

  • django.forms.ModelMultipleChoiceField now returns an empty QuerySet as the empty value instead of an empty list.
  • int_to_base36() properly raises a TypeError instead of ValueError for non-integer inputs.
  • The slugify template filter is now available as a standard Python function at django.utils.text.slugify(). Similarly, remove_tags is available at django.utils.html.remove_tags().
  • Uploaded files are no longer created as executable by default. If you need them to be executable change FILE_UPLOAD_PERMISSIONS to your needs. The new default value is 0o666 (octal) and the current umask value is first masked out.
  • The F expressions supported bitwise operators by & and |. These operators are now available using .bitand() and .bitor() instead. The removal of & and | was done to be consistent with Q() expressions and QuerySet combining where the operators are used as boolean AND and OR operators.
  • In a filter() call, when F expressions contained lookups spanning multi-valued relations, they didn’t always reuse the same relations as other lookups along the same chain. This was changed, and now F() expressions will always use the same relations as other lookups within the same filter() call.
  • The csrf_token template tag is no longer enclosed in a div. If you need HTML validation against pre-HTML5 Strict DTDs, you should add a div around it in your pages.
  • The template tags library adminmedia, which only contained the deprecated template tag {% admin_media_prefix %}, was removed. Attempting to load it with {% load adminmedia %} will fail. If your templates still contain that line you must remove it.
  • Because of an implementation oversight, it was possible to use django.contrib.redirects without enabling django.contrib.sites. This isn’t allowed any longer. If you’re using django.contrib.redirects, make sure INSTALLED_APPS contains django.contrib.sites.
  • BoundField.label_tag now escapes its contents argument. To avoid the HTML escaping, use django.utils.safestring.mark_safe() on the argument before passing it.
  • Accessing reverse one-to-one relations fetched via select_related() now raises DoesNotExist instead of returning None.

Features deprecated in 1.5

django.contrib.localflavor

The localflavor contrib app has been split into separate packages. django.contrib.localflavor itself will be removed in Django 1.6, after an accelerated deprecation.

The new packages are available on GitHub. The core team cannot efficiently maintain these packages in the long term — it spans just a dozen countries at this time; similar to translations, maintenance will be handed over to interested members of the community.

django.contrib.markup

The markup contrib module has been deprecated and will follow an accelerated deprecation schedule. Direct use of Python markup libraries or 3rd party tag libraries is preferred to Django maintaining this functionality in the framework.

AUTH_PROFILE_MODULE

With the introduction of custom user models, there is no longer any need for a built-in mechanism to store user profile data.

You can still define user profiles models that have a one-to-one relation with the User model - in fact, for many applications needing to associate data with a User account, this will be an appropriate design pattern to follow. However, the AUTH_PROFILE_MODULE setting, and the django.contrib.auth.models.User.get_profile() method for accessing the user profile model, should not be used any longer.

Streaming behavior of HttpResponse

Django 1.5 deprecates the ability to stream a response by passing an iterator to HttpResponse. If you rely on this behavior, switch to StreamingHttpResponse. See Explicit support for streaming responses above.

In Django 1.7 and above, the iterator will be consumed immediately by HttpResponse.

django.utils.simplejson

Since Django 1.5 drops support for Python 2.5, we can now rely on the json module being available in Python’s standard library, so we’ve removed our own copy of simplejson. You should now import json instead of django.utils.simplejson.

Unfortunately, this change might have unwanted side-effects, because of incompatibilities between versions of simplejson – see the backwards-incompatible changes section. If you rely on features added to simplejson after it became Python’s json, you should import simplejson explicitly.

django.utils.encoding.StrAndUnicode

The django.utils.encoding.StrAndUnicode mix-in has been deprecated. Define a __str__ method and apply the django.utils.encoding.python_2_unicode_compatible decorator instead.

django.utils.itercompat.product

The django.utils.itercompat.product function has been deprecated. Use the built-in itertools.product() instead.

cleanup management command

The cleanup management command has been deprecated and replaced by clearsessions.

daily_cleanup.py script

The undocumented daily_cleanup.py script has been deprecated. Use the clearsessions management command instead.

The depth keyword argument in select_related() has been deprecated. You should use field names instead.