Update underlying template to v0.4.2 & linkml to v1.10.0

[dalito]

• pre-commit hooks were updated
• various end-of-line/file fixes by pre-commit

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://nfdi-de.github.io/dcat-ap-plus/pr-preview/pr-64/
Built to branch gh-pages at 2026-03-04 00:03 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[HendrikBorgelt]

Hi @dalito,

@StroemPhi and I ran into the same errors that you are currently running into. @StroemPhi has worked on a good bug fix doc and analysis, which we wanted to discuss with you tomorrow, I think. So maybe if Philip hasn't contacted you yet, wait until tomorrow, and we can share some insights.

[dalito]

I looks to me like a linkml-runtime bug. But I need to verify.

[HendrikBorgelt]

yes it is,

here are some PR's touching on this issue
github.com/linkml/linkml/pull/3165
linkml/linkml#3183
linkml/linkml#3182

@StroemPhi pinpointed it down slightly more. I will share more with you via RocketChat.

[StroemPhi]

Fallback key heuristic in gen_postinit breaks _normalize_inlined for inlined_as_list slots

Summary

PR #3165 removed the and not slot.inlined_as_list guard from the fallback identifier search in gen_postinit. This causes the generated __post_init__ to call _normalize_inlined_as_list with a non-identifier required slot as key_name, triggering latent bugs in yamlutils._normalize_inlined that produce ValueError and TypeError at runtime when loading data via linkml-convert.

Affected versions

• Works: linkml 1.9.6
• Broken: linkml 1.10.0

Reproducer

Schema (minimal extract -- full schema at dcat-ap-plus):

classes:
  Entity:
    slots:
      - other_identifier
    slot_usage:
      other_identifier:
        range: Identifier
        multivalued: true
        inlined_as_list: true

  Identifier:
    slots:
      - notation
    slot_usage:
      notation:
        range: string
        required: true

Identifier has no identifier: true or key: true slot -- notation is simply a required slot.

Data:

other_identifier:
  - notation: https://pubchem.ncbi.nlm.nih.gov/compound/26248854

Command:

linkml-convert -s schema.yaml -C Entity -t rdf data.yaml

Error (linkml 1.10.0):

File "linkml_runtime/utils/yamlutils.py", line 182, in _normalize_inlined
    order_up(list_entry[lek], slot_type(list_entry))
ValueError: notation must be supplied

The same data passes linkml validate without error, because validation uses JSON Schema and never instantiates the generated Python dataclasses.

Root cause

In gen_postinit (python_generator.py), the elif slot.inlined: block searches for a fallback identifier when no true identifier/key exists:

1.9.6 (working):

if not identifier and not slot.inlined_as_list:
    for range_slot_name in slot_range_cls.slots:
        range_slot = self.schema.slots[range_slot_name]
        if range_slot.required:
            identifier = range_slot.name
            break
    keyed = False

1.10.0 (broken, after PR #3165):

if not identifier:
    for range_slot_name in slot_range_cls.slots:
        range_slot = self.schema.slots[range_slot_name]
        if range_slot.required and range_slot.range not in self.schema.classes:
            identifier = range_slot.name
            break
    keyed = False

The removal of and not slot.inlined_as_list means the fallback now also fires for inlined_as_list slots. This feeds a non-identifier required slot (notation) as key_name into _normalize_inlined_as_list, which then enters a code path in yamlutils._normalize_inlined (line ~182) that calls slot_type(list_entry) -- passing a dict as a positional argument instead of unpacking it as kwargs. The generated dataclass constructor expects keyword arguments, so notation is never set, and __post_init__ raises MissingRequiredField.

Even after patching that line to slot_type(**as_dict(list_entry)), a second bug surfaces in form_1 (line ~160) where raw_obj[key_name] = key fails with TypeError: list indices must be integers or slices, not str when raw_obj happens to be a list.

Both are latent bugs in _normalize_inlined that were never triggered before because inlined_as_list slots without a true identifier never entered that code path.

Why the fallback is conceptually wrong for lists

The fallback key heuristic exists to organize entries in a dict -- you need some key to index by. Lists don't need keys. The simple list comprehension that handles the no-identifier case works correctly:

sn = [v if isinstance(v, Type) else Type(**as_dict(v)) for v in sn]

Proposed fix

Move the guard inside the if not identifier block so it only restricts the fallback, not true identifiers:

if not identifier:
    if not slot.inlined_as_list:
        for range_slot_name in slot_range_cls.slots:
            range_slot = self.schema.slots[range_slot_name]
            if range_slot.required and range_slot.range not in self.schema.classes:
                identifier = range_slot.name
                break
    keyed = False

This preserves PR #3165's range_slot.range not in self.schema.classes refinement for dict cases while preventing the fallback from firing for list cases, where it causes the crash.

Behavior matrix:

Slot has identifier/key?	inlined_as_list?	Old (1.9.6)	New (1.10.0)	Fix
Yes	Yes	_normalize_inlined_as_list	_normalize_inlined_as_list	_normalize_inlined_as_list (unchanged)
Yes	No	_normalize_inlined_as_dict	_normalize_inlined_as_dict	_normalize_inlined_as_dict (unchanged)
No	Yes	list comprehension	crash	list comprehension (restored)
No	No	fallback -> _normalize_inlined_as_dict	fallback -> _normalize_inlined_as_dict	fallback -> _normalize_inlined_as_dict (unchanged)

Secondary issue in yamlutils._normalize_inlined

Independent of the gen-python fix, there are two latent bugs in linkml_runtime.utils.yamlutils._normalize_inlined that should be tracked separately:

1. Line ~182: slot_type(list_entry) passes a dict as a positional arg; should be slot_type(**as_dict(list_entry))
2. Line ~160: form_1 doesn't handle the case where raw_obj is a list, causing TypeError

These are pre-existing bugs in linkml-runtime that are only exposed when _normalize_inlined_as_list is called with a non-identifier key. The gen-python fix avoids triggering them, but they should still be fixed defensively.

Environment

• Python 3.12
• linkml 1.10.0
• linkml-runtime (as shipped with linkml 1.10.0)
• OS: Windows/WSL2

Regression tests

The PR includes pytest tests covering all four cells of the behavior matrix. Each test generates Python from a minimal schema, compiles it, and loads data via yaml_loader. The critical regression test is test_no_identifier_inlined_as_list (cell 3) which reproduces the crash.

Related

• PR #3165 (introduced the change)

[dalito]

With the fixes proposed for linkml-runtime in linkml/linkml#3247 the tests here will pass.