# 09 — Testing

## Required before merge

```bash
go test ./...
```

All tests must pass before any change is merged.

## Where to add tests

| Change area | Test location |
|-------------|---------------|
| Collectors | `internal/collector/*_test.go` |
| HTTP handlers | `internal/server/*_test.go` |
| Exporters | `internal/exporter/*_test.go` |
| Parsers | `internal/parser/vendors/<vendor>/*_test.go` |

## Exporter tests

The Reanimator exporter has comprehensive coverage:

| Test file | Coverage |
|-----------|----------|
| `reanimator_converter_test.go` | Unit tests per conversion function |
| `reanimator_integration_test.go` | Full export with realistic `AnalysisResult` |

Run exporter tests only:
```bash
go test ./internal/exporter/...
go test ./internal/exporter/... -v -run Reanimator
go test ./internal/exporter/... -cover
```

## Guidelines

- Prefer table-driven tests for parsing logic (multiple input variants).
- Do not rely on network access in unit tests.
- Test both the happy path and edge cases (missing fields, empty collections).
- When adding a new vendor parser, include at minimum:
  - `Detect()` test with a positive and a negative sample file list.
  - `Parse()` test with a minimal but representative archive.

## Dedup and filtering functions — mandatory coverage

Any function that deduplicates, filters, or classifies hardware inventory items
**must** have tests covering all three axes before the code is considered done:

| Axis | What to test | Why |
|------|-------------|-----|
| **True positive** | Items that ARE duplicates are collapsed to one | Proves the function works |
| **True negative** | Items that are NOT duplicates are kept separate | Proves the function doesn't over-collapse |
| **Counter-case** | The scenario that motivated the original code still works after changes | Prevents regression from future fixes |

### Worked example — GPU dedup regression (2026-03-11)

`collectGPUsFromProcessors` was added for MSI (chassis Id matches processor Id).
No tests → when Supermicro HGX arrived (chassis Id = "HGX_GPU_SXM_1", processor Id = "GPU_SXM_1"),
the chassis lookup silently returned nothing, serial stayed empty, UUID was new → 8 duplicate GPUs.

Simultaneously, fixing `gpuDocDedupKey` to use `slot|model` before path collapsed two distinct
GraphicsControllers GPUs with the same model into one — breaking an existing test that had no
counter-case for the path-fallback scenario.

**Required test matrix for any dedup function:**

```
TestXxx_CollapsesDuplicates   — same item via two sources → 1 result
TestXxx_KeepsDistinct          — two different items with same model → 2 results
TestXxx_<VendorThatMotivated>  — the specific vendor/setup that triggered the code
```

### Practical rule

When you write a new filter/dedup/classify function, ask:
1. Does my test cover the vendor that motivated this code?
2. Does my test cover a *different* vendor or naming convention where the function must NOT fire?
3. If I change the dedup key logic, do existing tests still exercise the old correct behavior?

If any answer is "no" — add the missing test before committing.