feat(core): add ordered flag to attribute schema to ignore list eleme…

[ajtmccarty]

Why

When two branches both edit the same list-valued attribute (a List/JSON attribute, or a
Dropdown's choices, an attribute's enum, etc.) so that they end up holding the same elements
in a different order, the diff reports a value conflict and merge/rebase is refused — even though
there is no semantic divergence. This is the converged-state dead-end described in #9764 for
Dropdown.choices: once both branches hold the same set of choices, the branch still cannot rebase.

Goal: let a schema author declare that element order is not significant for an attribute, so
reordering (or converging to the same set in a different order) is not treated as a conflict.

Non-goals: this does not change the rebase guard to honor ResolveDiffConflict selections
— that is a separate change. It also does not alter merge value resolution: when there is no
conflict, the merge continues to apply the branch's value as before.

Closes #9764 (resolves the converged-choices dead-end; the rebase-guard resolution handling is
out of scope here and is captured in #9654).

What changed

Behavioral changes

• New ordered boolean on attribute schemas (default True). When set to False on a List or
  JSON-array attribute, reordering its elements is no longer a conflict during branch merge or
  rebase. Element multiplicity still matters — adding or removing an element is still a conflict.
• The built-in schema attributes enum, dropdown choices, generic used_by, and
  restricted_namespaces are marked ordered=False, so two branches converging to the same set of
  values in a different order no longer conflict.

Implementation notes

• Conflict detection lives in one place: ConflictsEnricher._have_same_value. For an unordered
  attribute it now compares the two JSON values as multisets (canonicalizing each element), falling
  back to exact comparison when a value is not a JSON array.
• The enricher resolves the attribute schema from the diff (source) branch, so a branch that
  changes an attribute's ordered flag has that change honored within its own diff. Schema lookups
  degrade gracefully (fall back to order-sensitive) when a schema/branch cannot be resolved.
• ordered uses UpdateSupport.ALLOWED: it only affects how diffs are interpreted, never stored
  data, so no data migration is required (unlike unique/read_only).

What stayed the same

• Default behavior is unchanged (ordered=True): order-sensitive conflict detection for every
  attribute that does not opt out.
• No data migration; merge value resolution and the rebase guard are unchanged.
• Conflicts are still re-evaluated against live values on every recompute (both the
  update_branch_diff and recalculate paths) — a conflict recorded during a real divergence is
  cleared once the two sides converge.

Suggested review order

1. backend/infrahub/core/diff/conflicts_enricher.py — the core comparison and schema resolution.
2. backend/infrahub/core/schema/definitions/internal.py — the new field and the four ordered=False flips.
3. backend/infrahub/dependencies/builder/diff/conflicts_enricher.py — wiring db into the enricher.
4. Tests: backend/tests/component/core/diff/test_ordered_attribute_conflicts.py and additions to test_conflicts_enricher.py.
5. Generated/rendered files (mechanical): core/schema/generated/attribute_schema.py, schema/openapi.json, docs/docs/reference/schema/*.mdx.

Impact & rollout

• Backward compatibility: fully backward compatible — ordered defaults to True, preserving
  existing order-sensitive behavior for all attributes that do not opt out.
• Performance: negligible; the enricher does one extra JSON parse + sort only for attributes
  flagged ordered=False.
• Config/env changes: none. New per-attribute schema field, no feature flag.
• Deployment notes: safe to deploy; no data migration required (UpdateSupport.ALLOWED).

Checklist

• Tests added/updated
• Changelog entry added (changelog/9764.fixed.md)
• External docs updated (schema reference regenerated)
• Internal .md docs updated (internal knowledge and AI code tools knowledge)
• I have reviewed AI generated content

---

Summary by cubic

Adds an ordered flag to attribute schemas so reordering elements in List/JSON arrays no longer triggers merge/rebase conflicts. Actual adds/removes or different counts still conflict.

• New Features
  • Added ordered: boolean to attribute schemas (default true).
  • Diff conflict logic treats List and JSON-array attributes with ordered=false as order-insensitive (multiset). Schema is resolved from the diff branch so a branch’s own ordered change applies.
  • Core schema sets ordered=false for: enum, dropdown choices, generic used_by, and restricted_namespaces.
  • ConflictsEnricher now takes db; DI updated. Internals use ujson for canonicalization.
  • Field exposed in generated schema and openapi.json; regenerated frontend REST types and refreshed .betterer.results. Docs updated (reference and conflict guide). Tests added for reorder scenarios, including JSON arrays and both recompute paths.

^{Written for commit 8ea3f4a. Summary will update on new commits.}

[cubic-dev-ai]

1 issue found across 10 files

Confidence score: 4/5

• In changelog/9764.fixed.md, the documented default for ordered conflicts with the schema (false vs actual true), which can mislead users into relying on the wrong behavior and cause unexpected conflict handling after upgrade; update the changelog text (including the “default” parenthetical) to match the real default before merging.

<sub>Shadow auto-approve: would not auto-approve because issues were found.

Re-trigger cubic</sub>

[codspeed-hq]

Merging this PR will not alter performance

✅ 12 untouched benchmarks

---

_{Comparing ajtm-07012026-allow-order-ignore (8ea3f4a) with develop (e2e8744)}

[cubic-dev-ai]

0 issues found across 2 files (changes from recent commits).

<sub>Shadow auto-approve: would require human review. Adds new ordered flag to attribute schemas and modifies core conflict detection logic for diff/merge. This is a feature change in a critical path that requires human review to ensure correctness and no unintended side effects.

Re-trigger cubic</sub>

[gmazoyer]

The List path is well covered, but the "JSON" branch of _UNORDERED_CAPABLE_KINDS never fires in any test. Could we add one JSON attribute with an array value (proves the JSON branch works) and one where the JSON value is a top-level dict (locks in the exact-comparison fallback)?

[cubic-dev-ai]

1 issue found across 3 files (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="python_sdk">

<violation number="1" location="python_sdk:1">
P2: The SDK submodule bump (e0c27a14) adds MERGE_FAILED to BranchStatus but does not add the new `ordered` attribute schema field that this PR introduces. AttributeSchema in the SDK is a Pydantic model that deserializes server schema responses — without `ordered`, SDK-based clients that fetch attribute schemas (e.g., infrahubctl schema dump) will silently drop the field. Consider adding `ordered: bool = True` to AttributeSchema so the SDK round-trips the new field correctly.</violation>
</file>

<sub>Shadow auto-approve: would not auto-approve because issues were found.
Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: The SDK submodule bump (e0c27a14) adds MERGE_FAILED to BranchStatus but does not add the new ordered attribute schema field that this PR introduces. AttributeSchema in the SDK is a Pydantic model that deserializes server schema responses — without ordered, SDK-based clients that fetch attribute schemas (e.g., infrahubctl schema dump) will silently drop the field. Consider adding ordered: bool = True to AttributeSchema so the SDK round-trips the new field correctly.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At python_sdk, line 1:

<comment>The SDK submodule bump (e0c27a14) adds MERGE_FAILED to BranchStatus but does not add the new `ordered` attribute schema field that this PR introduces. AttributeSchema in the SDK is a Pydantic model that deserializes server schema responses — without `ordered`, SDK-based clients that fetch attribute schemas (e.g., infrahubctl schema dump) will silently drop the field. Consider adding `ordered: bool = True` to AttributeSchema so the SDK round-trips the new field correctly.</comment>

<file context>
@@ -1 +1 @@
-Subproject commit b76594a5ce659265832cad22f4406e46af21d3cb
+Subproject commit e0c27a149d8d5442805d5660c0d86869a30d3298
</file context>