Skip to content

[OpenAPI 3.1] Phase 0: test harness for paired 3.0/3.1 fixtures #948

Description

@DrSatyr

Part of the OpenAPI 3.1 plan: #942 (Stage 1, Phase 0)

Goal

The test setup that every later phase uses.

Scope

  • Fixture layout: core/src/test/resources/oas31/<rule>/ with old-3.0.yaml, new-3.0.yaml, old-3.1.yaml, new-3.1.yaml
  • A parameterized test that checks 3.0→3.0, 3.1→3.1 and 3.0→3.1 give the same verdict
  • An equivalence test: the same spec written in 3.0 and in 3.1 gives no changes
  • First fixtures: type change, null allowed, exclusive bounds, array items, allOf, oneOf
  • One Swagger 2.0 fixture
  • Today's 3.1 failures are marked as known (disabled with a reference to the phase that fixes them), so the build stays green

Done when

  • Harness and first fixtures merged, build green
  • Known failures are listed in this issue, each linked to the phase that fixes it

Activity

  1. added this to the 3.0.0 milestone on Nov 15, 2026
  2. added theissue type on Nov 15, 2026
  3. Inf1mum commented on Nov 16, 2026

    @Inf1mum

    Hi @DrSatyr — happy to take Phase 0. Plan: paired fixtures under core/src/test/resources/oas31/<rule>/ (old/new × 3.0/3.1) for type change, nullable, exclusive bounds, array items, allOf and oneOf, plus one Swagger 2.0 pair; a parameterized test asserting 3.0→3.0, 3.1→3.1 and 3.0→3.1 produce the same verdict, and the equivalence check that the same spec written as 3.0 and as 3.1 diffs to "no changes". Today's 3.1 gaps get marked disabled, each with a reference to the phase that fixes it — I hit several of these in the fixture work on #938, so I know where they live. Planning to start in the next few days unless you're already on it.

  4. Inf1mum commented on Nov 16, 2026

    @Inf1mum

    Known-failures inventory for the harness (PR #961). Everything failing on today's engine is kept as a disabled test in Oas31PairedFixturesTest, each naming the phase that un-disables it, so the build stays green.

    Same-version verdicts (3.1->3.1 must agree with 3.0->3.0) - fixed by Phase 2 (#950), engine reads 3.1 fields:

    Rule Today's 3.1->3.1 Should be Root cause
    type-change no_changes incompatible engine compares getType(), null for 3.1; type lives in the types set (#938, #910)
    nullable no_changes compatible null rule for "null" in types not implemented
    exclusive-bounds compatible incompatible 3.1 numeric exclusiveMinimum is not read
    array-items no_changes incompatible items compared only on ArraySchema, 3.1 arrays are JsonSchema
    all-of no_changes incompatible allOf flattened only on ComposedSchema (#938)
    one-of no_changes incompatible oneOf members compared only on ComposedSchema

    Cross-version verdicts (3.0->3.1) and the written equivalence - fixed by Phase 3 (#951), converter wired in:

    Passing today and enabled: the 3.0->3.0 baseline verdicts for all six rules, and the Swagger 2.0 pair matching the 3.0 verdict. Each phase flips its own tests back on by removing the @Disabled annotations.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions