Files
zeekayandHanzo Dev cec6d69791 test(object): the package ran ZERO tests and hid a live defect behind "ok"
TestMain called os.Exit(0) before m.Run() whenever no seeded database was
present. That bought an honest "these did not run" for the integration tests at
the price of the entire package: `go test ./object/` printed a cheerful "ok" in
0.1s having executed NOTHING — all 182 tests, including the ~150 that touch no
database at all.

That is false assurance, and it was not theoretical. It concealed a real defect
for as long as it existed:

    --- FAIL: TestCrawlStorageEndpoint
        host = "minio.hanzo.svc.cluster.local:9000", want "s3.hanzo.svc:9000"

The crawl-storage default still named a MinIO Service that no longer exists in
the cluster (verified: no minio Service in any namespace; s3 is the replacement).
crawlStorageDefaultEndpoint had already been corrected to s3.hanzo.svc:9000 with
a comment explaining exactly this, but conf/app.conf still SET the key to the
dead host, so the corrected constant could never apply and the test guarding it
could never report. conf/app.conf now leaves it empty and the constant is the
single source of the default. (Production was unaffected — its mounted app.conf
carries no such line — so this bit local dev and any deploy shipping the repo
default.)

Only TWO tests in this package acquire a store, so the gate belongs in THEM, not
over all 182. requireStore(t) skips on an absent database and the panics that
made a missing store abort the whole binary are now t.Skip/t.Fatalf. Both
packages' helpers defer to ONE probe (object.StoreProbeReason, the standard
export_test.go idiom) so there is no second copy to drift.

Measured, before -> after: 0 tests executed -> 153 PASS, 0 FAIL, 26 honest SKIP,
0 panics. Full suite: 22 packages ok, 0 FAIL.

Co-authored-by: Hanzo Dev <[email protected]>
2026-08-04 18:23:34 -07:00
..