ListSerializer (many=True): error on first item (index 0) loses its `0.` prefix in `attr` with DRF 3.18

[sshishov]

Summary

With DRF 3.18, a validation error raised on the first item (index 0) of a many=True / ListSerializer payload loses its top-level 0. index prefix in the formatted attr. Errors on items at index 1 and above are unaffected, and nested list indices (e.g. line_items.0.account) are also unaffected — only the top-level index 0 is dropped.

This makes it impossible for API clients to know which row of a bulk request a validation error belongs to when the error is on the first row.

Environment

• drf-standardized-errors==0.16.0
• djangorestframework==3.18.0

Reproduction

import django
from django.conf import settings

settings.configure(
    INSTALLED_APPS=["django.contrib.contenttypes", "django.contrib.auth", "rest_framework"],
    DATABASES={},
    USE_TZ=True,
)
django.setup()

from rest_framework import serializers
from rest_framework.exceptions import ValidationError
from drf_standardized_errors.formatter import ExceptionFormatter


class ItemSerializer(serializers.Serializer):
    email = serializers.EmailField()


def formatted_attrs(payload):
    serializer = ItemSerializer(data=payload, many=True)
    try:
        serializer.is_valid(raise_exception=True)
    except ValidationError as exc:
        errors = ExceptionFormatter(exc, None, None).get_errors()
        return [e.attr for e in errors]


# Error on the FIRST item (index 0)
print("error on index 0:", formatted_attrs([{"email": "not-an-email"}, {"email": "a@b.com"}]))
# Error on the SECOND item (index 1)
print("error on index 1:", formatted_attrs([{"email": "a@b.com"}, {"email": "not-an-email"}]))

Actual output (DRF 3.18)

error on index 0: ['email']
error on index 1: ['1.email']

Expected output

error on index 0: ['0.email']
error on index 1: ['1.email']

Root cause

DRF 3.18 changed ListSerializer.to_internal_value to collect child errors in an index-keyed dict instead of a positional list (serializers.py):

# DRF 3.17.x — errors is a positional list
errors = []
for item in data:
    try:
        ...
    except ValidationError as exc:
        errors.append(exc.detail)   # -> [{'email': [...]}, {}]
    else:
        errors.append({})
if any(errors):
    raise ValidationError(errors)

# DRF 3.18.0 — errors is an index-keyed dict
errors = {}
for index, item in enumerate(data):
    try:
        ...
    except ValidationError as exc:
        errors[index] = exc.detail   # -> {0: {'email': [...]}}
if errors:
    raise ValidationError(errors)

So the exception detail is now {0: {'email': [ErrorDetail(...)]}} instead of [{'email': [...]}, {}].

In flatten_errors (formatter.py), the dict branch only prefixes the child key when the parent attr is truthy:

elif isinstance(detail, dict):
    for key, value in detail.items():
        if attr:                       # <-- formatter.py:138
            key = f"{attr}{package_settings.NESTED_FIELD_SEPARATOR}{key}"
        fifo.append((value, key, None))

When the top-level dict key is the integer 0, it is passed down as attr, and if attr: evaluates if 0: → False, so the child (email) is never prefixed with 0.. Integer keys 1, 2, … are truthy, so those rows are prefixed correctly — which is why only index 0 is affected.

Suggested fix

Guard against None explicitly rather than relying on truthiness, so the integer 0 is treated as a valid attr segment:

elif isinstance(detail, dict):
    for key, value in detail.items():
        if attr is not None:
            key = f"{attr}{package_settings.NESTED_FIELD_SEPARATOR}{key}"
        fifo.append((value, key, None))

(The same if attr: truthiness check appears at formatter.py:126 in the list branch; worth reviewing for consistency, though that path builds attr from str(index) so it is not hit by an integer 0.)

[sshishov]

This is the PR in DRF which changed the format: encode/django-rest-framework#9837