Merge pull request #1885 from dandi/1839-path-ordering

Use different collation for Asset `path` field

dandiapi/api/migrations/0006_asset_path_collation.py

@@ -0,0 +1,24 @@
+# Generated by Django 4.1.13 on 2024-03-06 17:04
+from __future__ import annotations
+
+from django.db import migrations, models
+
+import dandiapi.api.models.asset
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ('api', '0005_null_charfield'),
+    ]
+
+    operations = [
+        migrations.AlterField(
+            model_name='asset',
+            name='path',
+            field=models.CharField(
+                db_collation='C.utf8',
+                max_length=512,
+                validators=[dandiapi.api.models.asset.validate_asset_path],
+            ),
+        ),
+    ]

dandiapi/api/models/asset.py

@@ -134,7 +134,7 @@ class Status(models.TextChoices):
         INVALID = 'Invalid'
 
     asset_id = models.UUIDField(unique=True, default=uuid.uuid4)
-    path = models.CharField(max_length=512, validators=[validate_asset_path])
+    path = models.CharField(max_length=512, validators=[validate_asset_path], db_collation='C.utf8')
     blob = models.ForeignKey(
         AssetBlob, related_name='assets', on_delete=models.CASCADE, null=True, blank=True
     )

dandiapi/api/tests/test_asset.py

@@ -369,6 +369,25 @@ def test_asset_rest_list_ordering(api_client, version, asset_factory, order_para
     assert result_paths == ordering
 
 
+@pytest.mark.django_db()
+def test_asset_path_ordering(api_client, version, asset_factory):
+    # The default collation will ignore special characters, including slashes, on the first pass. If
+    # there are ties, it uses these characters to break ties. This means that in the below example,
+    # removing the slashes leads to a comparison of 'az' and 'aaz', which would obviously sort the
+    # latter before the former. However, with the slashes, it's clear that 'a/z' should come before
+    # 'aa/z'. This is fixed by changing the collation of the path field, and as such this test
+    # serves as a regression test.
+    a = asset_factory(path='a/z')
+    b = asset_factory(path='aa/z')
+    version.assets.add(a)
+    version.assets.add(b)
+
+    asset_listing = Asset.objects.filter(versions__in=[version]).order_by('path')
+    assert asset_listing.count() == 2
+    assert asset_listing[0].pk == a.pk
+    assert asset_listing[1].pk == b.pk
+
+
 @pytest.mark.django_db()
 def test_asset_rest_retrieve(api_client, version, asset, asset_factory):
     version.assets.add(asset)