Fixes #406: Resolve correct to_field_name for CSV/YAML/JSON bulk import

[bctiemann]

Fixes: #406

Summary

Bulk import of Object and MultiObject fields referencing models without a name field (e.g. dcim.ModuleType, which uses model as its identifier) failed with "invalid accessor field name" because to_field_name was hardcoded to fall back to 'name'.

Root Cause

In ObjectFieldType.get_form_field() and MultiObjectFieldType.get_form_field(), the CSV import path used:

to_field_name = getattr(field, 'to_field_name', None) or 'name'

ModuleType has no name field — it uses model — so CSVModelChoiceField received an invalid accessor and raised FieldError, surfaced to the user as "invalid accessor field name".

NetBox core's own bulk import form for ModuleType explicitly sets to_field_name='model', confirming 'name' is not a universal default.

Fix

Add _csv_import_to_field_name(model_class, explicit=None) which probes the target model's fields in priority order ('name', 'slug', 'model', 'identifier') and returns the first match, falling back to 'pk'. Both get_form_field() call sites now use this helper.

With this fix, importing a CO instance that references a ModuleType works as expected:

---
testname: my-component
module_type: CPU567

Tests

• test_csv_import_field_uses_name_for_models_with_name — Device (has name) → 'name'
• test_csv_import_field_uses_model_for_module_type — ModuleType (no name) → 'model'
• test_csv_import_form_field_for_module_type_uses_model_field — end-to-end: get_form_field(for_csv_import=True) on a ModuleType COTF produces to_field_name='model'

🤖 Generated with Claude Code

[bctiemann]

@claude /review and limit response to areas of potential concern.

[github-actions]

Claude finished @bctiemann's task in 1m 45s —— View job

---

Review: Areas of Potential Concern

• Read changed files
• Analyze implementation
• Post review

---

1. Field existence vs. uniqueness (field_types.py:64–68)

_csv_import_to_field_name probes for field presence only — it does not verify the selected field is unique. CSVModelChoiceField requires the to_field_name to identify exactly one record; if it matches a field that is only contextually unique (scoped to a parent, for example), imports will raise MultipleObjectsReturned at runtime with no indication that the wrong to_field_name was chosen. The current candidates (name, slug, model, identifier) are globally unique on all common NetBox models, so the practical risk is low, but a future addition to _CSV_IDENTIFIER_FIELD_PRECEDENCE could silently introduce this bug. Consider adding a uniqueness check:

field_obj = model_class._meta.get_field(candidate)
if getattr(field_obj, 'unique', False):
    return candidate

Fix this →

---

2. explicit override is not validated (field_types.py:62–63)

When explicit is non-None it is returned immediately without checking that the field actually exists on model_class. If a stored to_field_name on a CustomObjectTypeField is stale (e.g. the referenced model was updated and the field was renamed), this silently passes an invalid accessor to CSVModelChoiceField, which will then raise a FieldError at import time — the same user-visible error the PR is trying to fix, just from a different path. A one-line guard would close this:

if explicit is not None:
    try:
        model_class._meta.get_field(explicit)
        return explicit
    except FieldDoesNotExist:
        logger.warning('Stored to_field_name %r not found on %s; probing.', explicit, model_class.__name__)

Fix this →

---

3. Test gaps (tests/test_field_types.py)

Two paths in _csv_import_to_field_name are not covered by the new tests:

• explicit= path: There is no test asserting that a non-None explicit value is returned directly (bypassing the probe loop). This is the primary path for any CustomObjectTypeField that has a stored to_field_name.
• pk fallback path: No test for the warning + 'pk' return when no candidate field matches. This is reachable with a mock model.

---

4. PR title says "CSV/YAML/JSON" — only CSV form field is affected

The fix is in get_form_field(for_csv_import=True), which covers NetBox's unified bulk import form (used for CSV, YAML, and JSON). That's correct — the title is accurate. Just confirming this isn't three separate code paths being claimed as fixed.

---

Overall: The core fix is correct and clearly targeted. Concerns 1 and 2 are worth addressing before merge; concern 3 is a test quality improvement.