July 29, 2009
Welcome to Django 1.1!
Django 1.1 includes a number of nifty new features, lots of bug fixes, and an easy upgrade path from Django 1.0.
Django has a policy of API stability. This means that, in general, code you develop against Django 1.0 should continue to work against 1.1 unchanged. However, we do sometimes make backwards-incompatible changes if they’re necessary to resolve bugs, and there are a handful of such (minor) changes between Django 1.0 and Django 1.1.
Before upgrading to Django 1.1 you should double-check that the following changes don’t impact you, and upgrade your code if they do.
Django 1.1 modifies the method used to generate database constraint names so that names are consistent regardless of machine word size. This change is backwards incompatible for some users.
If you are using a 32-bit platform, you’re off the hook; you’ll observe no differences as a result of this change.
However, users on 64-bit platforms may experience some problems using the
reset
management command. Prior to this change, 64-bit platforms
would generate a 64-bit, 16 character digest in the constraint name; for
example:
ALTER TABLE myapp_sometable ADD CONSTRAINT object_id_refs_id_5e8f10c132091d1e FOREIGN KEY ...
Following this change, all platforms, regardless of word size, will generate a 32-bit, 8 character digest in the constraint name; for example:
ALTER TABLE myapp_sometable ADD CONSTRAINT object_id_refs_id_32091d1e FOREIGN KEY ...
As a result of this change, you will not be able to use the reset
management command on any table made by a 64-bit machine. This is because the
new generated name will not match the historically generated name; as a
result, the SQL constructed by the reset command will be invalid.
If you need to reset an application that was created with 64-bit constraints,
you will need to manually drop the old constraint prior to invoking
reset
.
Django 1.1 runs tests inside a transaction, allowing better test performance (see test performance improvements for details).
This change is slightly backwards incompatible if existing tests need to test transactional behavior, if they rely on invalid assumptions about the test environment, or if they require a specific test case ordering.
For these cases, TransactionTestCase
can be used instead.
This is a just a quick fix to get around test case errors revealed by the new
rollback approach; in the long-term tests should be rewritten to correct the
test case.
SetRemoteAddrFromForwardedFor
middleware¶For convenience, Django 1.0 included an optional middleware class –
django.middleware.http.SetRemoteAddrFromForwardedFor
– which updated the
value of REMOTE_ADDR
based on the HTTP X-Forwarded-For
header commonly
set by some proxy configurations.
It has been demonstrated that this mechanism cannot be made reliable enough for
general-purpose use, and that (despite documentation to the contrary) its
inclusion in Django may lead application developers to assume that the value of
REMOTE_ADDR
is “safe” or in some way reliable as a source of authentication.
While not directly a security issue, we’ve decided to remove this middleware
with the Django 1.1 release. It has been replaced with a class that does nothing
other than raise a DeprecationWarning
.
If you’ve been relying on this middleware, the easiest upgrade path is:
Verify that it works correctly with your upstream proxy, modifying it to support your particular proxy (if necessary).
Introduce your modified version of SetRemoteAddrFromForwardedFor
as a
piece of middleware in your own project.
In Django 1.0, files uploaded and stored in a model’s FileField
were
saved to disk before the model was saved to the database. This meant that the
actual file name assigned to the file was available before saving. For example,
it was available in a model’s pre-save signal handler.
In Django 1.1 the file is saved as part of saving the model in the database, so the actual file name used on disk cannot be relied on until after the model has been saved.
In Django 1.1, BaseModelFormSet
now calls
ModelForm.save()
.
This is backwards-incompatible if you were modifying self.initial
in a model
formset’s __init__
, or if you relied on the internal _total_form_count
or _initial_form_count
attributes of BaseFormSet. Those attributes are now
public methods.
join
filter’s escaping behavior¶The join
filter no longer escapes the literal value that is
passed in for the connector.
This is backwards incompatible for the special situation of the literal string
containing one of the five special HTML characters. Thus, if you were writing
{{ foo|join:"&" }}
, you now have to write {{ foo|join:"&" }}
.
The previous behavior was a bug and contrary to what was documented and expected.
redirect_to()
generic view¶Django 1.1 adds a permanent
argument to the
django.views.generic.simple.redirect_to()
view. This is technically
backwards-incompatible if you were using the redirect_to
view with a
format-string key called ‘permanent’, which is highly unlikely.
One feature has been marked as deprecated in Django 1.1:
You should no longer use AdminSite.root()
to register that admin
views. That is, if your URLconf contains the line:
(r"^admin/(.*)", admin.site.root),
You should change it to read:
(r"^admin/", include(admin.site.urls)),
You should begin to remove use of this feature from your code immediately.
AdminSite.root
will raise a PendingDeprecationWarning
if used in
Django 1.1. This warning is hidden by default. In Django 1.2, this warning will
be upgraded to a DeprecationWarning
, which will be displayed loudly. Django
1.3 will remove AdminSite.root()
entirely.
For more details on our deprecation policies and strategy, see Django’s release process.
Quite a bit: since Django 1.0, we’ve made 1,290 code commits, fixed 1,206 bugs, and added roughly 10,000 lines of documentation.
The major new features in Django 1.1 are:
Two major enhancements have been added to Django’s object-relational mapper (ORM): aggregate support, and query expressions.
It’s now possible to run SQL aggregate queries (i.e. COUNT()
, MAX()
,
MIN()
, etc.) from within Django’s ORM. You can choose to either return the
results of the aggregate directly, or else annotate the objects in a
QuerySet
with the results of the aggregate
query.
This feature is available as new
aggregate()
and
annotate()
methods, and is covered in
detail in the ORM aggregation documentation.
Queries can now refer to another field on the query and can traverse
relationships to refer to fields on related models. This is implemented in the
new F
object; for full details, including examples,
consult the F expressions documentation
.
A number of features have been added to Django’s model layer:
You can now control whether or not Django manages the life-cycle of the database
tables for a model using the managed
model option. This
defaults to True
, meaning that Django will create the appropriate database
tables in syncdb
and remove them as part of the reset
command. That is, Django manages the database table’s lifecycle.
If you set this to False
, however, no database table creating or deletion
will be automatically performed for this model. This is useful if the model
represents an existing table or a database view that has been created by some
other means.
For more details, see the documentation for the managed
option.
You can now create proxy models: subclasses of existing models that only add Python-level (rather than database-level) behavior and aren’t represented by a new table. That is, the new model is a proxy for some underlying model, which stores all the real data.
All the details can be found in the proxy models documentation. This feature is similar on the surface to unmanaged models, so the documentation has an explanation of how proxy models differ from unmanaged models.
In some complex situations, your models might contain fields which could contain a lot of data (for example, large text fields), or require expensive processing to convert them to Python objects. If you know you don’t need those particular fields, you can now tell Django not to retrieve them from the database.
You’ll do this with the new queryset methods
defer()
and
only()
.
A few notable improvements have been made to the testing framework.
Tests written using Django’s testing framework now run dramatically faster (as much as 10 times faster in many cases).
This was accomplished through the introduction of transaction-based tests: when
using django.test.TestCase
, your tests will now be run in a transaction
which is rolled back when finished, instead of by flushing and re-populating the
database. This results in an immense speedup for most types of unit tests. See
the documentation for TestCase
and TransactionTestCase
for a
full description, and some important notes on database support.
A couple of small – but highly useful – improvements have been made to the test client:
The test Client
now can automatically follow redirects with the
follow
argument to Client.get()
and Client.post()
. This
makes testing views that issue redirects simpler.
It’s now easier to get at the template context in the response returned
the test client: you’ll simply access the context as
request.context[key]
. The old way, which treats request.context
as
a list of contexts, one for each rendered template in the inheritance
chain, is still available if you need it.
Django 1.1 adds a couple of nifty new features to Django’s admin interface:
You can now make fields editable on the admin list views via the new list_editable admin option. These fields will show up as form widgets on the list pages, and can be edited and saved in bulk.
You can now define admin actions that can perform some action to a group of models in bulk. Users will be able to select objects on the change list page and then apply these bulk actions to all selected objects.
Django ships with one pre-defined admin action to delete a group of objects in one fell swoop.
Django now has much better support for conditional view processing using the standard ETag
and
Last-Modified
HTTP headers. This means you can now easily short-circuit
view processing by testing less-expensive conditions. For many views this can
lead to a serious improvement in speed and reduction in bandwidth.
Django 1.1 improves named URL patterns with the introduction of URL “namespaces.”
In short, this feature allows the same group of URLs, from the same application, to be included in a Django URLConf multiple times, with varying (and potentially nested) named prefixes which will be used when performing reverse resolution. In other words, reusable applications like Django’s admin interface may be registered multiple times without URL conflicts.
For full details, see the documentation on defining URL namespaces.
In Django 1.1, GeoDjango (i.e.
django.contrib.gis
) has several new features:
Support for SpatiaLite – a spatial database for SQLite – as a spatial backend.
Geographic aggregates (Collect
, Extent
, MakeLine
, Union
)
and F
expressions.
New GeoQuerySet
methods: collect
, geojson
, and
snap_to_grid
.
A new list interface methods for GEOSGeometry
objects.
For more details, see the GeoDjango documentation.
Other new features and changes introduced since Django 1.0 include:
The CSRF protection middleware has been split into
two classes – CsrfViewMiddleware
checks incoming requests, and
CsrfResponseMiddleware
processes outgoing responses. The combined
CsrfMiddleware
class (which does both) remains for
backwards-compatibility, but using the split classes is now recommended in
order to allow fine-grained control of when and where the CSRF processing
takes place.
reverse()
and code which uses it (e.g., the {% url %}
template tag)
now works with URLs in Django’s administrative site, provided that the admin
URLs are set up via include(admin.site.urls)
(sending admin requests to
the admin.site.root
view still works, but URLs in the admin will not be
“reversible” when configured this way).
The include()
function in Django URLconf modules can now accept sequences
of URL patterns (generated by patterns()
) in addition to module names.
Instances of Django forms (see the forms overview)
now have two additional methods, hidden_fields()
and visible_fields()
,
which return the list of hidden – i.e., <input type="hidden">
– and
visible fields on the form, respectively.
The redirect_to
generic view
now accepts an additional keyword argument
permanent
. If permanent
is True
, the view will emit an HTTP
permanent redirect (status code 301). If False
, the view will emit an HTTP
temporary redirect (status code 302).
A new database lookup type – week_day
– has been added for DateField
and DateTimeField
. This type of lookup accepts a number between 1 (Sunday)
and 7 (Saturday), and returns objects where the field value matches that day
of the week. See the full list of lookup types for
details.
The {% for %}
tag in Django’s template language now accepts an optional
{% empty %}
clause, to be displayed when {% for %}
is asked to loop
over an empty sequence. See the list of built-in template tags for examples of this.
The dumpdata
management command now accepts individual
model names as arguments, allowing you to export the data just from
particular models.
There’s a new safeseq
template filter which works just like
safe
for lists, marking each item in the list as safe.
Cache backends now support incr()
and
decr()
commands to increment and decrement the value of a cache key.
On cache backends that support atomic increment/decrement – most
notably, the memcached backend – these operations will be atomic, and
quite fast.
Django now can easily delegate authentication to the web server via a new authentication backend that supports
the standard REMOTE_USER
environment variable used for this purpose.
There’s a new django.shortcuts.redirect()
function that makes it
easier to issue redirects given an object, a view name, or a URL.
The postgresql_psycopg2
backend now supports native PostgreSQL
autocommit. This is an advanced, PostgreSQL-specific
feature, that can make certain read-heavy applications a good deal
faster.
We’ll take a short break, and then work on Django 1.2 will begin – no rest for
the weary! If you’d like to help, discussion of Django development, including
progress toward the 1.2 release, takes place daily on the django-developers
mailing list and in the #django-dev
IRC channel on irc.libera.chat
.
Feel free to join the discussions!
Django’s online documentation also includes pointers on how to contribute to Django:
Contributions on any level – developing code, writing documentation or simply triaging tickets and helping to test proposed bugfixes – are always welcome and appreciated.
And that’s the way it is.
Nov 27, 2024