jenkins improvements

This commit is contained in:
2026-04-21 17:37:07 +01:00
parent 401a56a4c6
commit 15e1e2321e
8 changed files with 209 additions and 93 deletions
@@ -21,7 +21,7 @@ Its goal is to turn the decision into concrete review and CI mechanics without c
Define the repository-level mechanics for:
- global coverage gate continuity;
- domain coverage gate enforcement in CI;
- domain-scoped coverage evidence during review;
- baseline tracking that can start at `0` per domain and tighten later;
- reviewer-visible rules for when domain tests are mandatory.
@@ -29,7 +29,7 @@ Define the repository-level mechanics for:
## Escopo
- Add repository commands or scripts for domain-oriented coverage evidence collection.
- Add repository-facing documentation that explains the global gate plus domain evidence workflow.
- Add repository-facing documentation that explains the domain gate plus domain evidence workflow.
- Align CI/coverage outputs with the new governance model without replacing `llvm-cov`.
## Fora de Escopo
@@ -51,7 +51,8 @@ Add one repository-visible command path that produces:
- existing global HTML/XML/JSON coverage artifacts;
- a documented procedure for mapping changed files/modules to one of the canonical domains;
- an initial baseline format that records per-domain coverage starting from `0`.
- an initial baseline format that records per-domain coverage starting from `0`;
- a CI-friendly command that fails when a domain baseline is not met.
**File(s):**
- `Makefile`
@@ -68,7 +69,7 @@ Prefer simple wrappers over new tooling. For example:
- capture `llvm-cov` JSON/HTML consistently;
- filter or summarize target files/modules for a domain;
- emit review-friendly evidence artifacts without changing the authoritative global gate.
- emit review-friendly evidence artifacts while also enforcing the authoritative domain gate.
**File(s):**
- `Makefile`
@@ -86,6 +87,7 @@ Document that any PR touching a canonical domain must either:
- justify why the observable contract did not change.
Document that domain coverage evidence is mandatory review input even when domain thresholds are still at baseline `0`.
Document that Jenkins consumes the repository helper target directly instead of re-implementing the gate inline.
**File(s):**
- `README.md`
@@ -94,10 +96,10 @@ Document that domain coverage evidence is mandatory review input even when domai
### Step 4 - Align CI outputs with the governance model
**What:**
Make the existing coverage pipeline clearly compatible with domain review.
Make the existing coverage pipeline clearly compatible with domain review and domain-based CI failure.
**How:**
Keep the current global Jenkins gate unchanged while ensuring produced artifacts are sufficient for domain inspection.
Make Jenkins call a repository helper target that runs the coverage pipeline, produces the existing artifacts, and fails from domain baselines.
If needed, archive additional summaries or helper outputs generated by the new scripts.
**File(s):**
@@ -106,10 +108,10 @@ If needed, archive additional summaries or helper outputs generated by the new s
## Criterios de Aceite
- [ ] The repository exposes a documented way to collect domain-oriented coverage evidence without replacing the global `llvm-cov` gate.
- [ ] The repository exposes a documented way to collect domain-oriented coverage evidence and enforce domain baselines in CI.
- [ ] Review guidance explicitly states when domain tests are mandatory.
- [ ] Domain baselines are represented in a form that can start at `0` and increase later.
- [ ] Existing global coverage enforcement remains intact.
- [ ] Jenkins uses a repository-owned helper target for the coverage gate.
## Tests / Validacao
@@ -122,9 +124,10 @@ If needed, archive additional summaries or helper outputs generated by the new s
- Show that the global coverage artifacts still build.
- Show an example of domain evidence collection for at least one domain.
- Show that the domain gate fails when a baseline is missed.
## Riscos
- Overdesigning domain evidence collection before real reviewer usage patterns exist.
- Accidentally creating a second competing coverage authority instead of a reviewer aid.
- Letting Jenkins re-encode policy logic instead of consuming the repository helper target.
- Spreading policy text across too many documents and making the rule hard to find.