9072f5eec6
27-assertion TAP-style suite for lib-management.dep-fetcher v0.1.0:
- 4 internal-helper tests (id_to_path, gitea_slug_for incl. .git
exception)
- 6 JSON decoder tests (objects/arrays/nesting)
- 3 conflict-detection tests (synthetic two-parent fixture)
- 3 closure-walk tests (clean fixture, check_only mode)
- 3 case-B (clean) tests
- 4 case-C (dirty) tests
- 4 case-D (wrong-tag → silent checkout, post-state verification)
Cases A (missing+clone) and E (broken describe) need network or
contrived corruption; out of scope for slice 5 per architecture spec.
Fixture matrix:
conflict_a/ — static, no .git needed
clean_b/ — .git at v0.1.0 (clean)
dirty_c/ — .git at v0.1.0-1-g…-dirty
wrong_tag_d/ — .git with tags v0.1.0 + v0.2.0, HEAD at v0.1.0
fixtures/_setup.sh reconstructs the .git states (idempotent — restores
manifest.lib content after the case-D silent-checkout side-effect).
Required pre-step before each test session; the test-runner blocks
with a clear error if fixtures/_setup.stamp is missing.
.gitignore covers the .git subdirs and the stamp file — they are
local-machine state, rebuilt by _setup.sh from the source-tracked
manifest.lib seed files.
2.8 KiB
lib-management.dep-fetcher-test
TAP-style assertion suite for lib-management.dep-fetcher (slice 5 of
the dep-fetcher implementation plan).
Coverage
| Case | Spec | What it verifies |
|---|---|---|
| Internal | — | _id_to_path, _gitea_slug_for (incl. .git-suffix exception), _json_decode |
| Conflict | §5 conflict-detection | Two parents pin same lib to different versions → ok=false, one conflict entry |
| Walk | §5 closure walk | Manifest tree resolves transitively; deduplication; closure non-empty |
| B | §5 case B (clean) | describe == v<X.Y.Z> → no warning, no error |
| C | §5 case C (dirty) | describe matches v<X.Y.Z>-N-g…[-dirty] → one kind="dirty" warning |
| D | §5 case D (wrong-tag) | describe != v<X.Y.Z> → silent checkout; post-state is case B |
Cases A (missing → clone) and E (broken describe) require live network or contrived corruption; they are out of scope for slice 5 (public-network only; private repos = paid content in later slice).
Running
# 1. Stage source to install (see install-reference-workflow.md)
bash ~/Projects/Sporel/install-reference.sh --no-verify
# 2. Reconstruct .git fixtures (case B / C / D)
bash /c/Games/Sporel/libs/lib-management/dep-fetcher-test/fixtures/_setup.sh
# 3. Run the test-runner module headless
cd /c/Games/Sporel
SPOREL_CI=1 ./bin/Sporel.exe --module=dep-fetcher-test-runner
# expected: 27/27 assertions pass; exit 0
The runner lives at sporel-modules/dep-fetcher-test-runner/. It is a
kind=test-module that re-exports the assertions from this lib so
they participate in test-all-modules.sh aggregation (and therefore
milestone-check).
Fixture layout
fixtures/
├── _setup.sh # rebuild .git state for cases B/C/D
├── _setup.stamp # touched by _setup.sh; presence-check in init.lua
│
├── conflict_a/ # static (no .git needed): two parents pin
│ ├── manifest.module lib-core.maps to different versions
│ └── install/libs/lib-x/consumer/manifest.lib
│
├── clean_b/ # _setup.sh creates .git at v0.1.0
├── dirty_c/ # _setup.sh creates .git at v0.1.0-1-g…-dirty
└── wrong_tag_d/ # _setup.sh creates .git with tags v0.1.0 + v0.2.0,
# HEAD at v0.1.0 (module pins v0.2.0)
_setup.sh is idempotent — re-runs after the case-D silent-checkout
side-effect restore the wrong-tag fixture to its initial v0.1.0 HEAD.
The script sets per-repo user.name / user.email so it does not
touch the user's global git config.
Why a separate runner module?
Test libs (role: "test") cannot be consumed by production modules
(kind: "module"). The runner is therefore kind: "test-module",
which can depend on test libs (per engine_lib_resolver role-check).