Optimize `db_default` creation

[Adam Johnson]

Currently, Field._get_default() returns a lambda that instantiates a new DatabaseDefault each time it is called. This expression is identical each time, so this work is wasted. We can optimize it by creating a single expression that is returned each time.

Thanks to Adam Sołtysik for the hint in ​https://forum.djangoproject.com/t/faster-bulk-create-using-dictionaries/43891

Benchmarked with a quick modification to the test suite:

./tests/bulk_create/tests.py

@@ -877 +877 @@
             2 if connection.features.can_return_rows_from_bulk_insert else 1,
         )
 
+
+    def test_benchmark(self):
+        import time
+        start = time.perf_counter()
+        DbDefaultModel.objects.bulk_create(
+            [DbDefaultModel(name=f"obj {i}") for i in range(100_000)]
+        )
+        end = time.perf_counter()
+        print(f"{end - start:.3f} seconds")
+
     @skipUnlessDBFeature(
         "can_return_rows_from_bulk_insert", "supports_expression_defaults"
     )

Run with:

./runtests.py --parallel 1 bulk_create.tests.BulkCreateTests.test_benchmark -v 0
System check identified no issues (0 silenced).
0.617 seconds
----------------------------------------------------------------------
Ran 1 test in 0.618s

OK

Results, best of 3:

Before: 0.691 seconds
After: 0.610 seconds

A ~12% speedup on this bulk_create() operation, on a model with a single db_default field.

[Simon Charette]

Thanks for that Adam! When I read the thread I figured something was definitely wrong but I didn't have the time to look at it. Could you also submit a minimal benchmark againt django-asv that makes use of db_default?

[Jacob Walls <jacobtylerwalls@…>]

In 2b192bff:

Fixed #36858 -- Optimized Field._get_default() for db_default case.

Create and share a single instance of DatabaseDefault instead of making a new
one each time the lambda is called. The quick benchmark on the ticket shows a
~12% speedup for a large bulk_create() operation.

[Adam Johnson]

Replying to Simon Charette:

Thanks for that Adam! When I read the thread I figured something was definitely wrong but I didn't have the time to look at it. Could you also submit a minimal benchmark againt django-asv that makes use of db_default?

​https://github.com/django/django-asv/pull/95