Django 1.6 release notes

Dedicated to Malcolm Tredinnick

On March 17, 2013, the Django project and the free software community lost a very dear friend and developer.

Malcolm was a long-time contributor to Django, a model community member, a brilliant mind, and a friend. His contributions to Django — and to many other open source projects — are nearly impossible to enumerate. Many on the core Django team had their first patches reviewed by him; his mentorship enriched us. His consideration, patience, and dedication will always be an inspiration to us.

This release of Django is for Malcolm.

-- The Django Developers

November 6, 2013

Welcome to Django 1.6!

These release notes cover the , as well as some backwards incompatible changes you’ll want to be aware of when upgrading from Django 1.5 or older versions. We’ve also dropped some features, which are detailed in , and we’ve begun the deprecation process for some features.

Django 1.6, like Django 1.5, requires Python 2.6.5 or above. Python 3 is also officially supported. We highly recommend the latest minor release for each supported Python series (2.6.X, 2.7.X, 3.2.X, and 3.3.X).

Django 1.6 will be the final release series to support Python 2.6; beginning with Django 1.7, the minimum supported Python version will be 2.7.

Python 3.4 is not supported, but support will be added in Django 1.7.

The default templates used by and startapp have been simplified and modernized. The is now enabled by default in new projects; the sites framework no longer is. is now on and the database defaults to SQLite.

If the default templates don’t suit your tastes, you can use custom project and app templates.

Improved transaction management

Django’s transaction management was overhauled. Database-level autocommit is now turned on by default. This makes transaction handling more explicit and should improve performance. The existing APIs were deprecated, and new APIs were introduced, as described in the .

Persistent database connections

Django now supports reusing the same database connection for several requests. This avoids the overhead of re-establishing a connection at the beginning of each request. For backwards compatibility, this feature is disabled by default. See 持久连接 for details.

Discovery of tests in any test module

Django 1.6 ships with a new test runner that allows more flexibility in the location of tests. The previous runner (django.test.simple.DjangoTestSuiteRunner) found tests only in the models.py and tests.py modules of a Python package in .

The new runner (django.test.runner.DiscoverRunner) uses the test discovery features built into unittest2 (the version of unittest in the Python 2.7+ standard library, and bundled with Django). With test discovery, tests can be located in any module whose name matches the pattern test*.py.

In addition, the test labels provided to ./manage.py test to nominate specific tests to run must now be full Python dotted paths (or directory paths), rather than applabel.TestCase.test_method_name pseudo-paths. This allows running tests located anywhere in your codebase, rather than only in INSTALLED_APPS. For more details, see .

This change is backwards-incompatible; see the backwards-incompatibility notes.

Time zone aware aggregation

The support for introduced in Django 1.4 didn’t work well with QuerySet.dates(): aggregation was always performed in UTC. This limitation was lifted in Django 1.6. Use to perform time zone aware aggregation on a DateTimeField.

Support for savepoints in SQLite

Django 1.6 adds support for savepoints in SQLite, with some .

BinaryField model field

A new django.db.models.BinaryField model field allows storage of raw binary data in the database.

GeoDjango form widgets

GeoDjango now provides for its geo-specialized fields. They are OpenLayers-based by default, but they can be customized to use any other JS framework.

check management command added for verifying compatibility

A check management command was added, enabling you to verify if your current configuration (currently oriented at settings) is compatible with the current version of Django.

algorithm changed

The Model.save() method now tries to directly UPDATE the database if the instance has a primary key value. Previously SELECT was performed to determine if UPDATE or INSERT were needed. The new algorithm needs only one query for updating an existing row while the old algorithm needed two. See for more details.

In some rare cases the database doesn’t report that a matching row was found when doing an UPDATE. An example is the PostgreSQL ON UPDATE trigger which returns NULL. In such cases it is possible to set django.db.models.Options.select_on_save flag to force saving to use the old algorithm.

次要特性

• Authentication backends can raise PermissionDenied to immediately fail the authentication chain.
• The HttpOnly flag can be set on the CSRF cookie with .
• The assertQuerysetEqual() now checks for undefined order and raises if undefined order is spotted. The order is seen as undefined if the given QuerySet isn’t ordered and there is more than one ordered value to compare against.
• Added earliest() for symmetry with .
• In addition to year, and day, the ORM now supports , minute and lookups.
• Django now wraps all PEP 249 exceptions.
• The default widgets for , URLField, , FloatField and use the new type attributes available in HTML5 (type='email', type='url', type='number'). Note that due to erratic support of the number input type with localized numbers in current browsers, Django only uses it when numeric fields are not localized.
• The number argument for lazy plural translations can be provided at translation time rather than at definition time.
• The of DeletionMixin is now interpolated with its object’s __dict__.
• and HttpResponsePermanentRedirect now provide an url attribute (equivalent to the URL the response will redirect to).
• The MemcachedCache cache backend now uses the latest protocol available.
• Added SuccessMessageMixin which provides a success_message attribute for based classes.
• Added the django.db.models.ForeignKey.db_constraint and options.
• The jQuery library embedded in the admin has been upgraded to version 1.9.1.
• Syndication feeds (django.contrib.syndication) can now pass extra context through to feed templates using a new callback.
• The admin list columns have a column-<field_name> class in the HTML so the columns header can be styled with CSS, e.g. to set a column width.
• The isolation level can be customized under PostgreSQL.
• The template tag now respects TEMPLATE_STRING_IF_INVALID for variables not present in the context, just like other template constructs.
• SimpleLazyObjects will now present more helpful representations in shell debugging situations.
• Generic GeometryField is now editable with the OpenLayers widget in the admin.
• The documentation contains a .
• The diffsettings command gained a --all option.
• django.forms.fields.Field.__init__ now calls super(), allowing field mixins to implement __init__() methods that will reliably be called.
• The validate_max parameter was added to BaseFormSet and , and ModelForm and inline versions of the same. The behavior of validation for formsets with max_num was clarified. The previously undocumented behavior that hardened formsets against memory exhaustion attacks was documented, and the undocumented limit of the higher of 1000 or max_num forms was changed so it is always 1000 more than max_num.
• Added BCryptSHA256PasswordHasher to resolve the password truncation issue with bcrypt.

• Pillow is now the preferred image manipulation library to use with Django. is pending deprecation (support to be removed in Django 1.8). To upgrade, you should first uninstall PIL, then install Pillow.

• ModelForm accepts several new Meta options.

  • Fields included in the localized_fields list will be localized (by setting localize on the form field).
  • The labels, help_texts and error_messages options may be used to customize the default fields, see for details.
• The choices argument to model fields now accepts an iterable of iterables instead of requiring an iterable of lists or tuples.
• The reason phrase can be customized in HTTP responses using reason_phrase.
• When giving the URL of the next page for django.contrib.auth.views.logout(), django.contrib.auth.views.password_reset(), django.contrib.auth.views.password_reset_confirm(), and django.contrib.auth.views.password_change(), you can now pass URL names and they will be resolved.
• The new option specifies the primary keys of objects to dump. This option can only be used with one model.
• Added QuerySet methods first() and which are convenience methods returning the first or last object matching the filters. Returns None if there are no objects matching.
• View and now support HTTP PATCH method.
• GenericForeignKey now takes an optional for_concrete_model argument, which when set to False allows the field to reference proxy models. The default is True to retain the old behavior.
• The LocaleMiddleware now stores the active language in session if it is not present there. This prevents loss of language settings after session flush, e.g. logout.
• has been differentiated into a number of subclasses, and each will log to a matching named logger under the django.security logging hierarchy. Along with this change, a handler400 mechanism and default view are used whenever a SuspiciousOperation reaches the WSGI handler to return an HttpResponseBadRequest.
• The DoesNotExist exception now includes a message indicating the name of the attribute used for the lookup.
• The method no longer requires at least one keyword argument.
• The SimpleTestCase class includes a new assertion helper for testing formset errors: .
• The list of related fields added to a QuerySet by can be cleared using select_related(None).
• The get_extra() and methods on InlineModelAdmin may be overridden to customize the extra and maximum number of inline forms.
• Formsets now have a method.
• ModelForm fields can now override error messages defined in model fields by using the argument of a Field’s constructor. To take advantage of this new feature with your custom fields, see the updated recommendation for raising a ValidationError.
• now preserves filters on the list view after creating, editing or deleting an object. It’s possible to restore the previous behavior of clearing filters by setting the preserve_filters attribute to False.
• Added (which returns FormMixin.prefix by default) to allow customizing the of the form.
• Raw queries (Manager.raw() or cursor.execute()) can now use the “pyformat” parameter style, where placeholders in the query are given as '%(name)s' and the parameters are passed as a dictionary rather than a list (except on SQLite). This has long been possible (but not officially supported) on MySQL and PostgreSQL, and is now also available on Oracle.
• The default iteration count for the PBKDF2 password hasher has been increased by 20%. This backwards compatible change will not affect existing passwords or users who have subclassed django.contrib.auth.hashers.PBKDF2PasswordHasher to change the default value. Passwords will be upgraded to use the new iteration count as necessary.

警告

In addition to the changes outlined in this section, be sure to review the 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.

New transaction management model

Behavior changes

Database-level autocommit is enabled by default in Django 1.6. While this doesn’t change the general spirit of Django’s transaction management, there are a few backwards-incompatibilities.

Savepoints and

The changes in transaction management may result in additional statements to create, release or rollback savepoints. This is more likely to happen with SQLite, since it didn’t support savepoints until this release.

If tests using assertNumQueries() fail because of a higher number of queries than expected, check that the extra queries are related to savepoints, and adjust the expected number of queries accordingly.

Autocommit option for PostgreSQL

In previous versions, database-level autocommit was only an option for PostgreSQL, and it was disabled by default. This option is now ignored and can be removed.

New test runner

In order to maintain greater consistency with Python’s unittest module, the new test runner (django.test.runner.DiscoverRunner) does not automatically support some types of tests that were supported by the previous runner:

• Tests in models.py and tests/__init__.py files will no longer be found and run. Move them to a file whose name begins with test.
• Doctests will no longer be automatically discovered. To integrate doctests in your test suite, follow the .

Django bundles a modified version of the doctest module from the Python standard library (in django.test._doctest) and includes some additional doctest utilities. These utilities are deprecated and will be removed in Django 1.8; doctest suites should be updated to work with the standard library’s doctest module (or converted to unittest-compatible tests).

If you wish to delay updates to your test suite, you can set your setting to django.test.simple.DjangoTestSuiteRunner to fully restore the old test behavior. DjangoTestSuiteRunner is deprecated but will not be removed from Django until version 1.8.

Removal of django.contrib.gis.tests.GeoDjangoTestSuiteRunner GeoDjango custom test runner

This is for developers working on the GeoDjango application itself and related to the item above about changes in the test runners:

The django.contrib.gis.tests.GeoDjangoTestSuiteRunner test runner has been removed and the standalone GeoDjango tests execution setup it implemented isn’t supported anymore. To run the GeoDjango tests simply use the new DiscoverRunner and specify the django.contrib.gis app.

Custom user models in tests

The introduction of the new test runner has also slightly changed the way that test models are imported. As a result, any test that overrides AUTH_USER_MODEL to test behavior with one of Django’s test user models ( django.contrib.auth.tests.custom_user.CustomUser and django.contrib.auth.tests.custom_user.ExtensionUser) must now explicitly import the User model in your test module:

This import forces the custom user model to be registered. Without this import, the test will be unable to swap in the custom user model, and you will get an error reporting:

ImproperlyConfigured: AUTH_USER_MODEL refers to model 'auth.CustomUser' that has not been installed

Django 1.6 introduces time zone support for day, , and week_day lookups when is True. These lookups were previously performed in UTC regardless of the current time zone.

This requires time zone definitions in the database. If you’re using SQLite, you must install . If you’re using MySQL, you must install pytz and load the time zone tables with .

Addition of QuerySet.datetimes()

When the time zone support added in Django 1.4 was active, lookups returned unexpected results, because the aggregation was performed in UTC. To fix this, Django 1.6 introduces a new API, QuerySet.datetimes(). This requires a few changes in your code.

QuerySet.dates() returns date objects

now returns a list of date. It used to return a list of .

QuerySet.datetimes() returns a list of .

QuerySet.dates() no longer usable on DateTimeField

QuerySet.dates() raises an error if it’s used on when time zone support is active. Use QuerySet.datetimes() instead.

date_hierarchy requires time zone definitions

The feature of the admin now relies on QuerySet.datetimes() when it’s used on a .

date_list in generic views requires time zone definitions

For the same reason, accessing date_list in the context of a date-based generic view requires time zone definitions in the database when the view is based on a DateTimeField and is True. Learn more.

New lookups may clash with model fields

Django 1.6 introduces hour, minute, and second lookups on . If you had model fields called hour, minute, or second, the new lookups will clash with you field names. Append an explicit exact lookup if this is an issue.

BooleanField no longer defaults to False

When a doesn’t have an explicit default, the implicit default value is None. In previous version of Django, it was False, but that didn’t represent accurately the lack of a value.

Code that relies on the default value being False may raise an exception when saving new model instances to the database, because None isn’t an acceptable value for a . You should either specify default=False in the field definition, or ensure the field is set to True or False before saving the object.

Translations and comments in templates

Extraction of translations after comments

Extraction of translatable literals from templates with the makemessages command now correctly detects i18n constructs when they are located after a {# / #}-type comment on the same line. E.g.:

{# A comment #}{% trans "This literal was incorrectly ignored. Not anymore" %}

Location of translator comments

specified using {# / #} need to be at the end of a line. If they are not, the comments are ignored and makemessages will generate a warning. For example:

{# Translators: This is ignored #}{% trans "Translate me" %}
{{ title }}{# Translators: Extracted and associated with 'Welcome' below #}
<h1>{% trans "Welcome" %}</h1>

Quoting in reverse()

When reversing URLs, Django didn’t apply django.utils.http.urlquote to arguments before interpolating them in URL patterns. This bug is fixed in Django 1.6. If you worked around this bug by applying URL quoting before passing arguments to reverse(), this may result in double-quoting. If this happens, simply remove the URL quoting from your code. You will also have to replace special characters in URLs used in with their encoded versions.

Storage of IP addresses in the comments app

The comments app now uses a GenericIPAddressField for storing commenters’ IP addresses, to support comments submitted from IPv6 addresses. Until now, it stored them in an IPAddressField, which is only meant to support IPv4. When saving a comment made from an IPv6 address, the address would be silently truncated on MySQL databases, and raise an exception on Oracle. You will need to change the column type in your database to benefit from this change.

For MySQL, execute this query on your project’s database:

ALTER TABLE django_comments MODIFY ip_address VARCHAR(39);

For Oracle, execute this query:

ALTER TABLE DJANGO_COMMENTS MODIFY (ip_address VARCHAR2(39));

If you do not apply this change, the behavior is unchanged: on MySQL, IPv6 addresses are silently truncated; on Oracle, an exception is generated. No database change is needed for SQLite or PostgreSQL databases.

Percent literals in cursor.execute queries

When you are running raw SQL queries through the cursor.execute method, the rule about doubling percent literals (%) inside the query has been unified. Past behavior depended on the database backend. Now, across all backends, you only need to double literal percent characters if you are also providing replacement parameters. For example:

# No parameters, no percent doubling
cursor.execute("SELECT foo FROM bar WHERE baz = '30%'")
# Parameters passed, non-placeholders have to be doubled
cursor.execute("SELECT foo FROM bar WHERE baz = '30%%' and id = %s", [self.id])

SQLite users need to check and update such queries.

Help text of model form fields for ManyToManyField fields

HTML rendering of model form fields corresponding to model fields used to get the hard-coded sentence:

(or its translation to the active locale) imposed as the help legend shown along them if neither model nor help_text attributes were specified by the user (or this string was appended to any help_text that was provided).

Since this happened at the model layer, there was no way to prevent the text from appearing in cases where it wasn’t applicable such as form fields that implement user interactions that don’t involve a keyboard and/or a mouse.

Starting with Django 1.6, as an ad-hoc temporary backward-compatibility provision, the logic to add the “Hold down…” sentence has been moved to the model form field layer and modified to add the text only when the associated widget is SelectMultiple or selected subclasses.

The change can affect you in a backward incompatible way if you employ custom model form fields and/or widgets for ManyToManyField model fields whose UIs do rely on the automatic provision of the mentioned hard-coded sentence. These form field implementations need to adapt to the new scenario by providing their own handling of the help_text attribute.

Applications that use Django facilities together with Django built-in form fields and aren’t affected but need to be aware of what’s described in Munging of help text of model form fields for ManyToManyField fields below.

QuerySet iteration

The QuerySet iteration was changed to immediately convert all fetched rows to Model objects. In Django 1.5 and earlier the fetched rows were converted to Model objects in chunks of 100.

Existing code will work, but the amount of rows converted to objects might change in certain use cases. Such usages include partially looping over a queryset or any usage which ends up doing __bool__ or __contains__.

Notably most database backends did fetch all the rows in one go already in 1.5.

It is still possible to convert the fetched rows to Model objects lazily by using the method.

BoundField.label_tag now includes the form’s

This is consistent with how methods like Form.as_p and render labels.

If you manually render label_tag in your templates:

you’ll want to remove the colon (or whatever other separator you may be using) to avoid duplicating it when upgrading to Django 1.6. The following template in Django 1.6 will render identically to the above template in Django 1.5, except that the colon will appear inside the <label> element.

{{ form.my_field.label_tag }} {{ form.my_field }}

will render something like:

<label for="id_my_field">My Field:</label> <input id="id_my_field" type="text" name="my_field" />

If you want to keep the current behavior of rendering label_tag without the label_suffix, instantiate the form label_suffix=''. You can also customize the label_suffix on a per-field basis using the new label_suffix parameter on label_tag().

Admin views _changelist_filters GET parameter

To achieve preserving and restoring list view filters, admin views now pass around the _changelist_filters GET parameter. It’s important that you account for that change if you have custom admin templates or if your tests rely on the previous URLs. If you want to revert to the original behavior you can set the attribute to False.

django.contrib.auth password reset uses base 64 encoding of User PK

Past versions of Django used base 36 encoding of the User primary key in the password reset views and URLs (django.contrib.auth.views.password_reset_confirm()). Base 36 encoding is sufficient if the user primary key is an integer, however, with the introduction of custom user models in Django 1.5, that assumption may no longer be true.

django.contrib.auth.views.password_reset_confirm() has been modified to take a uidb64 parameter instead of uidb36. If you are reversing this view, for example in a custom password_reset_email.html template, be sure to update your code.

A temporary shim for django.contrib.auth.views.password_reset_confirm() that will allow password reset links generated prior to Django 1.6 to continue to work has been added to provide backwards compatibility; this will be removed in Django 1.7. Thus, as long as your site has been running Django 1.6 for more than PASSWORD_RESET_TIMEOUT_DAYS, this change will have no effect. If not (for example, if you upgrade directly from Django 1.5 to Django 1.7), then any password reset links generated before you upgrade to Django 1.7 or later won’t work after the upgrade.

In addition, if you have any custom password reset URLs, you will need to update them by replacing uidb36 with uidb64 and the dash that follows that pattern with a slash. Also add _\- to the list of characters that may match the uidb64 pattern.

例子：

url(r'^reset/(?P<uidb36>[0-9A-Za-z]+)-(?P<token>.+)/$',
    'django.contrib.auth.views.password_reset_confirm',
    name='password_reset_confirm'),

becomes:

url(r'^reset/(?P<uidb64>[0-9A-Za-z_\-]+)/(?P<token>.+)/$',
    'django.contrib.auth.views.password_reset_confirm',
    name='password_reset_confirm'),

You may also want to add the shim to support the old style reset links. Using the example above, you would modify the existing url by replacing django.contrib.auth.views.password_reset_confirm with django.contrib.auth.views.password_reset_confirm_uidb36 and also remove the name argument so it doesn’t conflict with the new url:

url(r'^reset/(?P<uidb36>[0-9A-Za-z]+)-(?P<token>.+)/$',
    'django.contrib.auth.views.password_reset_confirm_uidb36'),

You can remove this URL pattern after your app has been deployed with Django 1.6 for .

Default session serialization switched to JSON

Historically, django.contrib.sessions used to serialize session data before storing it in the backend. If you’re using the signed cookie session backend and is known by an attacker (there isn’t an inherent vulnerability in Django that would cause it to leak), the attacker could insert a string into their session which, when unpickled, executes arbitrary code on the server. The technique for doing so is simple and easily available on the internet. Although the cookie session storage signs the cookie-stored data to prevent tampering, a SECRET_KEY leak immediately escalates to a remote code execution vulnerability.

This attack can be mitigated by serializing session data using JSON rather than . To facilitate this, Django 1.5.3 introduced a new setting, SESSION_SERIALIZER, to customize the session serialization format. For backwards compatibility, this setting defaulted to using in Django 1.5.3, but we’ve changed the default to JSON in 1.6. If you upgrade and switch from pickle to JSON, sessions created before the upgrade will be lost. While JSON serialization does not support all Python objects like pickle does, we highly recommend using JSON-serialized sessions. Be aware of the following when checking your code to determine if JSON serialization will work for your application:

• JSON requires string keys, so you will likely run into problems if you are using non-string keys in request.session.

See the documentation for more details.

Object Relational Mapper changes

Django 1.6 contains many changes to the ORM. These changes fall mostly in three categories:

1. Bug fixes (e.g. proper join clauses for generic relations, query combining, join promotion, and join trimming fixes)
2. Preparation for new features. For example the ORM is now internally ready for multicolumn foreign keys.
3. General cleanup.

These changes can result in some compatibility problems. For example, some queries will now generate different table aliases. This can affect QuerySet.extra(). In addition some queries will now produce different results. An example is where the condition is a complex one (referencing multijoins inside Q objects). In many cases the affected queries didn’t produce correct results in Django 1.5 but do now. Unfortunately there are also cases that produce different results, but neither Django 1.5 nor 1.6 produce correct results.

Finally, there have been many changes to the ORM internal APIs.

• The django.db.models.query.EmptyQuerySet can’t be instantiated any more - it is only usable as a marker class for checking if has been called: isinstance(qs.none(), EmptyQuerySet)

• If your CSS/JavaScript code used to access HTML input widgets by type, you should review it as type='text' widgets might be now output as type='email', type='url' or type='number' depending on their corresponding field type.

• Form field’s error_messages that contain a placeholder should now always use a named placeholder ("Value '%(value)s' is too big" instead of "Value '%s' is too big"). See the corresponding field documentation for details about the names of the placeholders. The changes in 1.6 particularly affect and .

• Some for IntegerField, , IPAddressField, GenericIPAddressField, and have been suppressed because they duplicated error messages already provided by validators tied to the fields.

• There have been changes in the way timeouts are handled in cache backends. Explicitly passing in timeout=None no longer results in using the default timeout. It will now set a non-expiring timeout. Passing 0 into the memcache backend no longer uses the default timeout, and now will set-and-expire-immediately the value.

• The django.contrib.flatpages app used to set custom HTTP headers for debugging purposes. This functionality was not documented and made caching ineffective so it has been removed, along with its generic implementation, previously available in django.core.xheaders.

• The XViewMiddleware has been moved from django.middleware.doc to django.contrib.admindocs.middleware because it is an implementation detail of admindocs, proven not to be reusable in general.

• GenericIPAddressField will now only allow blank values if null values are also allowed. Creating a GenericIPAddressField where blank is allowed but null is not will trigger a model validation error because blank values are always stored as null. Previously, storing a blank value in a field which did not allow null would cause a database exception at runtime.

• If a NoReverseMatch exception is raised from a method when rendering a template, it is not silenced. For example, {{ obj.view_href }} will cause template rendering to fail if view_href() raises NoReverseMatch. There is no change to the tag, it causes template rendering to fail like always when NoReverseMatch is raised.

• django.test.Client.logout() now calls which will send the user_logged_out() signal.

• are now reversed by name, not their locations in django.contrib.auth.views. If you are using the views without a name, you should update your urlpatterns to use django.conf.urls.url() with the name parameter. For example:

  (r'^reset/done/$', 'django.contrib.auth.views.password_reset_complete')

  becomes:

• RedirectView now has a pattern_name attribute which allows it to choose the target by reversing the URL.

• In Django 1.4 and 1.5, a blank string was unintentionally not considered to be a valid password. This meant would save a blank password as an unusable password like set_unusable_password() does, and thus always returned False for blank passwords. This has been corrected in this release: blank passwords are now valid.

• The admin changelist_view previously accepted a pop GET parameter to signify it was to be displayed in a popup. This parameter has been renamed to _popup to be consistent with the rest of the admin views. You should update your custom templates if they use the previous parameter name.

• now accepts email addresses with localhost as the domain.

• The new makemessages --keep-pot option prevents deleting the temporary .pot file generated before creating the .po file.

• The undocumented django.core.servers.basehttp.WSGIServerException has been removed. Use socket.error provided by the standard library instead. This change was also released in Django 1.5.5.

• The signature of has changed and now accepts positional arguments as well (*args, **kwargs). Any unnamed captured group will now be passed to get_redirect_url() which may result in a TypeError if you don’t update the signature of your custom method.

Transaction management APIs

Transaction management was completely overhauled in Django 1.6, and the current APIs are deprecated:

• django.middleware.transaction.TransactionMiddleware
• django.db.transaction.autocommit
• django.db.transaction.commit_on_success
• django.db.transaction.commit_manually
• the TRANSACTIONS_MANAGED setting

django.contrib.comments

Django’s comment framework has been deprecated and is no longer supported. It will be available in Django 1.6 and 1.7, and removed in Django 1.8. Most users will be better served with a custom solution, or a hosted product like Disqus.

The code formerly known as django.contrib.comments is .

Support for PostgreSQL versions older than 8.4

The end of upstream support periods was reached in December 2011 for PostgreSQL 8.2 and in February 2013 for 8.3. As a consequence, Django 1.6 sets 8.4 as the minimum PostgreSQL version it officially supports.

You’re strongly encouraged to use the most recent version of PostgreSQL available, because of performance improvements and to take advantage of the native streaming replication available in PostgreSQL 9.x.

Changes to cycle and

The template system generally escapes all variables to avoid XSS attacks. However, due to an accident of history, the cycle and tags render their arguments as-is.

Django 1.6 starts a process to correct this inconsistency. The future template library provides alternate implementations of cycle and that autoescape their inputs. If you’re using these tags, you’re encouraged to include the following line at the top of your templates to enable the new behavior:

{% load cycle from future %}

或者：

{% load firstof from future %}

The tags implementing the old behavior have been deprecated, and in Django 1.8, the old behavior will be replaced with the new behavior. To ensure compatibility with future versions of Django, existing templates should be modified to use the future versions.

If necessary, you can temporarily disable auto-escaping with mark_safe() or .

CACHE_MIDDLEWARE_ANONYMOUS_ONLY setting

CacheMiddleware and UpdateCacheMiddleware used to provide a way to cache requests only if they weren’t made by a logged-in user. This mechanism was largely ineffective because the middleware correctly takes into account the Vary: Cookie HTTP header, and this header is being set on a variety of occasions, such as:

• accessing the session, or
• using CSRF protection, which is turned on by default, or
• using a client-side library which sets cookies, like Google Analytics.

This makes the cache effectively work on a per-session basis regardless of the CACHE_MIDDLEWARE_ANONYMOUS_ONLY setting.

SEND_BROKEN_LINK_EMAILS setting

used to provide basic reporting of broken links by email when SEND_BROKEN_LINK_EMAILS is set to True.

Because of intractable ordering problems between CommonMiddleware and , this feature was split out into a new middleware: BrokenLinkEmailsMiddleware.

If you’re relying on this feature, you should add 'django.middleware.common.BrokenLinkEmailsMiddleware' to your MIDDLEWARE_CLASSES setting and remove SEND_BROKEN_LINK_EMAILS from your settings.

_has_changed method on widgets

If you defined your own form widgets and defined the _has_changed method on a widget, you should now define this method on the form field itself.

module_name model _meta attribute

Model._meta.module_name was renamed to model_name. Despite being a private API, it will go through a regular deprecation path.

get_(add|change|delete)_permission model _meta methods

Model._meta.get_(add|change|delete)_permission methods were deprecated. Even if they were not part of the public API they’ll also go through a regular deprecation path. You can replace them with django.contrib.auth.get_permission_codename('action', Model._meta) where 'action' is 'add', 'change', or 'delete'.

get_query_set and similar methods renamed to get_queryset

Methods that return a QuerySet such as Manager.get_query_set or ModelAdmin.queryset have been renamed to get_queryset.

If you are writing a library that implements, for example, a Manager.get_query_set method, and you need to support old Django versions, you should rename the method and conditionally add an alias with the old name:

class CustomManager(models.Manager):
    def get_queryset(self):
        pass # ...
    if django.VERSION < (1, 6):
        get_query_set = get_queryset
    # For Django >= 1.6, models.Manager provides a get_query_set fallback
    # that emits a warning when used.

If you are writing a library that needs to call the get_queryset method and must support old Django versions, you should write:

get_queryset = (some_manager.get_query_set
                if hasattr(some_manager, 'get_query_set')
                else some_manager.get_queryset)
return get_queryset() # etc

In the general case of a custom manager that both implements its own get_queryset method and calls that method, and needs to work with older Django versions, and libraries that have not been updated yet, it is useful to define a get_queryset_compat method as below and use it internally to your manager:

class YourCustomManager(models.Manager):
    def get_queryset(self):
        return YourCustomQuerySet() # for example
    if django.VERSION < (1, 6):
        get_query_set = get_queryset
    def active(self): # for example
        return self.get_queryset_compat().filter(active=True)
    def get_queryset_compat(self):
        get_queryset = (self.get_query_set
                        if hasattr(self, 'get_query_set')
                        else self.get_queryset)
        return get_queryset()

This helps to minimize the changes that are needed, but also works correctly in the case of subclasses (such as RelatedManagers from Django 1.5) which might override either get_query_set or get_queryset.

shortcut view and URLconf

The shortcut view was moved from django.views.defaults to django.contrib.contenttypes.views shortly after the 1.0 release, but the old location was never deprecated. This oversight was corrected in Django 1.6 and you should now use the new location.

The URLconf django.conf.urls.shortcut was also deprecated. If you’re including it in an URLconf, simply replace:

(r'^prefix/', include('django.conf.urls.shortcut')),

with:

ModelForm without fields or exclude

Previously, if you wanted a to use all fields on the model, you could simply omit the Meta.fields attribute, and all fields would be used.

This can lead to security problems where fields are added to the model and, unintentionally, automatically become editable by end users. In some cases, particular with boolean fields, it is possible for this problem to be completely invisible. This is a form of Mass assignment vulnerability.

For this reason, this behavior is deprecated, and using the Meta.exclude option is strongly discouraged. Instead, all fields that are intended for inclusion in the form should be listed explicitly in the fields attribute.

If this security concern really does not apply in your case, there is a shortcut to explicitly indicate that all fields should be used - use the special value "__all__" for the fields attribute:

class MyModelForm(ModelForm):
    class Meta:
        fields = "__all__"
        model = MyModel

If you have custom ModelForms that only need to be used in the admin, there is another option. The admin has its own methods for defining fields (fieldsets etc.), and so adding a list of fields to the ModelForm is redundant. Instead, simply omit the Meta inner class of the ModelForm, or omit the Meta.model attribute. Since the ModelAdmin subclass knows which model it is for, it can add the necessary attributes to derive a functioning ModelForm. This behavior also works for earlier Django versions.

UpdateView and CreateView without explicit fields

The generic views and UpdateView, and anything else derived from , are vulnerable to the security problem described in the section above, because they can automatically create a ModelForm that uses all fields for a model.

For this reason, if you use these views for editing models, you must also supply the fields attribute (new in Django 1.6), which is a list of model fields and works in the same way as the ModelForm Meta.fields attribute. Alternatively, you can set the form_class attribute to a ModelForm that explicitly defines the fields to be used. Defining an UpdateView or CreateView subclass to be used with a model but without an explicit list of fields is deprecated.

Munging of help text of model form fields for ManyToManyField fields

Help text of these fields will need to be handled either by applications, custom form fields or widgets, just like happens with the rest of the model field types.