feat(all): qualified name resolution mode

[2ynn]

Please refer to sdk-gen-config PR#134 for the description of the new nameResolution flag strategy.

This PR introduces a new "qualified" resolution mode to preemptively prevent name conflicts and ensure names stay stable as the spec grows

Unlike shortest (formerly known as fixes.nameResolutionFeb2025), any inline schema (no $ref, title, or x-speakeasy-name-override) is systematically named after its property path from the nearest enclosing named schema , even in the absence of conflict:

Order:
  type: object
  properties:
    status:            # inline enum
      type: string
      enum: [pending, shipped]
    settings:
      type: object
      properties:
        level:         # nested inline enum
          type: string
          enum: [low, high]

schema	shortest	qualified
Order.status	Status	OrderStatus
Order.settings.level	Level	OrderSettingsLevel
Order.tags[] (array items)	Tag	OrderTag
Order.file ($ref to FileContent)	FileContent	FileContent (refs never prefixed)

Under shortest, adding a second status property anywhere later renames the existing Status type — a breaking change surfaced by unrelated spec edits. Under qualified the name is already qualified, so it never moves.

Not prefixed:

• $ref properties
• title/override-named schemas
• schemas with no named ancestor (operation request/response bodies, parameters)
• properties whose name already starts with the parent name (Redundant.redundantStatus stays RedundantStatus).

A conflict that still occurs (e.g. an actual component named like a qualified result) resolves reactively exactly as under shortest.

Structurally identical inline schemas shared by several parents split into per-parent types instead of merging into one bare-named, rename-prone type (stability over dedup); registration IDs already include the parent frame, so no dedup suppression was needed.

Implementation

qualifyInlineName in internal/namer/namer.go, gated on NameResolutionAtLeast(qualified) in the GetTypeName context-stack fallback:

• finds the last refName frame; requires every frame between it and the leaf property frame to be part of a plain property chain (property, component, modelNamespace) — any operation/request/response/oneOf frame falls through to the existing behaviour;
• joins the display names of the chain (Order_settings_level);
• returns a cloned stack with the consumed frames marked used and the leaf popped, so conflict resolution never re-applies the same labels (and sibling walks sharing a backing array are never mutated).

Explicit names ($ref/title/override) return earlier in GetTypeName, so they are unaffected by construction. Field names are untouched (GetFieldName passes a nil stack).

Test-config matrix

tests/config now pins nameResolution explicitly wherever the effective mode was previously implicit or expressed via the deprecated booleans:

variant	before	after	effective mode
primary	fixes.nameResolutionDec2023: false	nameResolution: legacy	legacy (unchanged)
secondary	fixes.nameResolutionDec2023: false	nameResolution: qualified	legacy → qualified
review	booleans true/true	nameResolution: shortest + booleans kept	shortest (unchanged)
tertiary	booleans only	unchanged	legacy via booleans (backward-compat coverage)
quaternary, no-zod	Dec2023: false	legacy mode added, booleans kept	legacy (unchanged)
security-options	booleans true/true	shortest mode added, booleans kept	shortest (unchanged)
client-credentials(-basic), oauth2-password	fixes block without naming booleans	nameResolution: legacy	legacy (unchanged)
variants with no fixes block	unchanged	unchanged	shortest via new-SDK defaults

Note on the last two rows: Fixes uses a custom YAML unmarshaller that decodes
into a zero-value struct, so the presence of any fixes: block wipes the
new-SDK defaults for every boolean not listed — which is why a block containing
only securityFeb2025: true was effectively running legacy naming. The
explicit pins freeze the current behaviour instead of leaving it dependent on
new-SDK detection.

---

Summary by cubic

Adds a new qualified name resolution mode that names inline schemas after their property path (e.g., Order.settings.level → OrderSettingsLevel) to keep names stable as the spec grows. The deprecated fixes.nameResolution* booleans are replaced by the nameResolution config option.

New Features

• Inline schemas without $ref, title, or x-speakeasy-name-override are now prefixed with the nearest enclosing named schema.
• Explicitly named schemas, $ref properties, and whole-document external refs that resolve to inline objects are not prefixed; an explicitly named inline schema becomes the naming root for its own children.
• A parent prefix is stripped only at word boundaries, so Redundant.redundantStatus stays RedundantStatus but Card.cardinality is not affected.
• Structurally identical inline schemas shared by multiple parents now get per-parent types.

Migration

• Replace fixes.nameResolutionFeb2025 and fixes.nameResolutionDec2023 with generation.nameResolution: legacy, shortest, or qualified.
• Test configs now pin the mode explicitly because any fixes: block wipes unlisted new-SDK defaults.
• Secondary test suites run under qualified, renaming inline SDK types such as BasicAuth → AuthServiceRequestBodyBasicAuth.
• sdk-gen-config is bumped to v1.59.0 to support the new mode.

^{Written for commit e9d0b74. Summary will update on new commits.}

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}

[cubic-dev-ai]

Running ultrareview automatically — This PR rewrites core naming resolution (new qualified mode, context-stack qualification, config migration) across 151 files — a subtle bug would silently break generated SDK type names and public APIs, so a deep review is warranted.. I'll post findings when complete.

[cubic-dev-ai]

Ultrareview completed in 16m 35s

All reported issues were addressed across 151 files

<sub>Tip: instead of fixing issues one by one fix them all with cubic

Re-trigger cubic</sub>

[AshGodfrey]

Not sure if you were ready for review but.. Qualified names can collide with each other and degrade to numeric suffixes. Foo.barBaz and FooBar.baz both qualify to FooBarBaz. Since the property and refName frames are consumed, the resolver has nothing left and emits FooBarBaz1/FooBarBaz2. Spec-growth test: with only Foo.barBaz you get FooBarBaz, then adding FooBar.baz renames it to FooBarBaz1. That is exactly the instability this mode exists to remove, and under shortest these never conflict (BarBaz/Baz). Rare, but it should be documented as a known limitation or handled, e.g. fall back to shortest for both sides when two qualified names collide, rather than numbering.

If this isn't an issue let me know.