test(packer): qualify asset contracts and improve Jenkins reporting
This commit is contained in:
+85
@@ -0,0 +1,85 @@
|
||||
---
|
||||
id: LSN-0082
|
||||
ticket: packer-runtime-graphics-conformance
|
||||
title: Packer Contracts and CI Failure Evidence
|
||||
created: 2026-10-16
|
||||
tags: [packer, studio, testing, jenkins]
|
||||
---
|
||||
|
||||
# Packer Contracts and CI Failure Evidence
|
||||
|
||||
## Original problem and scope
|
||||
|
||||
A broad runtime-readiness audit suggested graphics integration qualification,
|
||||
but historical review did not establish a new asset format migration. Existing
|
||||
Studio work already covered RGBA8888, scene identity and variable palettes.
|
||||
The actual opportunity was to strengthen producer-boundary evidence and ensure
|
||||
CI preserved useful results when tests failed.
|
||||
|
||||
Owner: packer; transverse owner: Studio Gradle/Jenkins infrastructure. The user
|
||||
explicitly excluded runtime provisioning in Jenkins and all new automated
|
||||
runtime tests, including local ones. Existing games remain the manual integrated
|
||||
verification method. This lesson does not claim those games were run.
|
||||
|
||||
## Consolidated decision and implementation
|
||||
|
||||
The packer's ordinary Java/JUnit suite now checks complete serialized GLYPH
|
||||
payloads for tile sizes 8/16/32 and palette counts 1/3/64, using independently
|
||||
authored hex expectations. Alpha, ordinary index zero and high-first nibble
|
||||
order are compared as bytes. A complete four-layer SCENE fixture covers
|
||||
little-endian identity, parallax, flags/flips and decoded-size accounting.
|
||||
An out-of-range palette cannot publish a new artifact.
|
||||
|
||||
A mixed-asset test reads the embedded header of assets.pa, checks sidecar
|
||||
agreement and proves that all payload slices cover the file contiguously.
|
||||
Two GLYPH preloads use slots 0/1 while SCENE uses its own slot 0; a fourth
|
||||
asset is deliberately not preloaded. AssetId 16909060 remains the scene's
|
||||
cold dependency even though the glyph preload slot is 1. These identities
|
||||
must not be conflated.
|
||||
|
||||
Jenkins now uses the repository wrapper, headless JVMs and a separate clean
|
||||
stage. Verification uses --continue: independent tests and reports complete,
|
||||
but Gradle still fails on test failure. Unconditional post handling isolates
|
||||
JUnit, coverage, HTML and archive publishers. Missing evidence after a successful
|
||||
verification is an error; failed preparation does not publish old reports.
|
||||
Existing coverage thresholds and their UNSTABLE policy remain unchanged.
|
||||
No production writer change was necessary.
|
||||
|
||||
## Evidence and limits
|
||||
|
||||
Studio base: 13c0ababbb0ff8623eb4df671729a891788b1ab1. Contract reference:
|
||||
runtime f7d7c167531bb662a1af9d73fc326bb476bf8aa7. Validation on 2026-10-16:
|
||||
|
||||
- Packer suite: 119 tests passed, zero skips; five new contract cases included.
|
||||
- Ordinary headless repository run: 897 tests, two failures, 13 existing skips.
|
||||
Consolidated JaCoCo XML and HTML were generated despite test failure.
|
||||
- Failures in unchanged files: PbsGradleDependencyAllowlistTest rejects the
|
||||
existing build-pipeline testImplementation PBS dependency; McpContractTest
|
||||
rejects an existing tool exampleArguments contract. The complete repository
|
||||
test gate is not green; these failures were not suppressed by this change.
|
||||
- Groovy 4.0.29 parsed the Jenkinsfile. Seven local step-double scenarios passed:
|
||||
success, test failure, publisher failure, missing contract report, absent
|
||||
reports on success/failure and clean failure. This does not validate actual
|
||||
controller plugins or constitute a Jenkins job run.
|
||||
- git diff --check and discussion validate passed.
|
||||
|
||||
No runtime invocation, automated frame comparison or manual gameplay was
|
||||
performed. Producer bytes passing these assertions do not prove live loading
|
||||
or presentation. Report manual game/revision/actions/results only after play.
|
||||
|
||||
## Pitfalls
|
||||
|
||||
- Deriving expected bytes through the production writer makes an oracle circular.
|
||||
- Checking only exported JSON misses corruption in the embedded artifact header.
|
||||
- Publishing reports only after a successful shell step hides failure diagnostics.
|
||||
- Allowing empty reports on successful verification can turn absent execution into
|
||||
apparent success. Allow absence only when the build already failed.
|
||||
- A broad contract corpus is not a reason to duplicate existing tests or port a
|
||||
runtime decoder into Java. Keep checks at the producer's actual output boundary.
|
||||
|
||||
## References
|
||||
|
||||
- [Fixture interpretation and commands](../../../prometeu-packer/prometeu-packer-v1/src/test/resources/fixtures/contracts/README.md).
|
||||
- [CI commands and failure behavior](../../../files/config/README.md).
|
||||
- [Build artifact specification](../../../docs/specs/packer/4.%20Build%20Artifacts%20and%20Deterministic%20Packing%20Specification.md).
|
||||
- Completed workflow provenance: AGD-0079, DEC-0066, PLN-0145 (housekept after implementation).
|
||||
Reference in New Issue
Block a user