Testing

docs/TECHNICAL_SPEC.md

@@ -414,7 +414,10 @@ filter-escaped against injection. The challenge front-end is shared: both PAM an
 the same `basicAuth` middleware, and the page load (`/`) is challenged too so the browser's
 native login dialog actually appears (a background `fetch('/me')` 401 does not prompt).
 Because Basic credentials are sent on every request, uopi can terminate **TLS** itself
-(`[server.tls]` → `ListenAndServeTLS`) without a reverse proxy. The four identity sources
+(`[server.tls]` → `ListenAndServeTLS`) without a reverse proxy; an optional
+`redirect_from` plain-HTTP listener 301-redirects `http://` visitors to the HTTPS service
+so they are upgraded instead of hitting the TLS port with cleartext ("client sent an HTTP
+request to an HTTPS server"). The four identity sources
 (proxy header, Kerberos, PAM-Basic, LDAP-Basic) all resolve to the same `userHeader` and are
 mutually exclusive where they overlap. Access is **role-based** through group
 memberships: each `[[groups]]` block lists members by role along the cumulative ladder

docs/TEST_COVERAGE_REPORT.md

@@ -0,0 +1,87 @@
+# Test Coverage Report — uopi
+
+**Date:** 2026-06-24
+**Branch:** develop
+**Task:** #167 — Raise backend coverage toward 90%
+**Status:** All suites green — `go test ./... -race`, `go vet ./...`, and `gofmt -l` clean.
+
+## Overall
+
+- **Main module total coverage: 67.7%** of statements.
+- **38 test files, 218 test/bench/fuzz functions** in the main module.
+- **27 new test files** added across the coverage initiative.
+- 3-module `go.work` workspace — each module is tested independently (`go test ./...`
+  from the root only exercises the main module; `pkg/ca` and `pkg/pva` must be tested
+  by `cd`-ing into them).
+
+## Main module (`github.com/uopi/uopi`) — by package
+
+| Coverage | Package | Notes |
+|---|---|---|
+| 100.0% | `internal/config` | |
+| 100.0% | `internal/pamauth` | stub (non-PAM build) |
+| 97.1% | `internal/metrics` | |
+| 96.3% | `internal/broker` | signal fan-out core |
+| 89.2% | `internal/panelacl` | |
+| 89.0% | `internal/audit` | |
+| 85.5% | `internal/access` | |
+| 82.6% | `internal/storage` | |
+| 82.4% | `internal/datasource/stub` | |
+| 81.7% | `internal/confmgr` | |
+| 81.0% | `internal/dsp` | |
+| 79.1% | `internal/datasource/servervar` | |
+| 72.9% | `internal/datasource/synthetic` | |
+| 67.2% | `internal/api` | large handler surface |
+| 66.6% | `internal/controllogic` | engine/Lua/cron uncovered |
+| 55.6% | `internal/datasource` | iface + ctx helpers |
+| 34.1% | `internal/server` | HTTP/WebSocket |
+| 33.9% | `internal/ldapauth` | network-bound |
+| 33.6% | `internal/datasource/epics` | CGo/libca-bound |
+| 0.0% | `internal/datasource/pva` | network-bound |
+| 0.0% | `cmd/uopi`, `cmd/catools`, `cmd/pvtools`, `tools/buildfrontend` | mains — no tests |
+
+## Workspace modules
+
+| Module | Coverage |
+|---|---|
+| `pkg/ca` (goca) | **83.9%** root · 90.0% `proto` · `testca` 0% (test harness) |
+| `pkg/pva` (gopva) | 85.5% `pvdata` · 12.1% root (network client) |
+
+## Remaining gaps toward 90%
+
+The lowest packages are all **I/O- or platform-bound**, needing integration harnesses
+rather than unit tests:
+
+- `server` (34%) — HTTP/WebSocket handlers.
+- `ldapauth` (34%) — needs a mock LDAP server.
+- `datasource/epics` (34%) — CGo `libca` linkage.
+- `datasource/pva` (0%) — needs a PVA test server (analogous to `testca` for CA).
+- `cmd/*` mains — typically excluded from coverage targets.
+
+Pure-logic packages are now in the 80–100% range. The biggest realistic remaining
+wins are `api` (67%) and `controllogic` (67%), where the uncovered code is the
+control-logic **engine** (Lua runtime, cron scheduling, dialog emission) and the
+network-dependent API handlers (channelFinder, archiverSearch).
+
+## Coverage gains (this initiative)
+
+| Package | Before | After |
+|---|---|---|
+| `internal/api` | 53.8% | 67.2% |
+| `internal/audit` | 75.3% | 89.0% |
+| `internal/broker` | 78.9% | 96.3% |
+| `internal/panelacl` | 65.9% | 89.2% |
+| `internal/confmgr` | 73.4% | 81.7% |
+| `internal/dsp` | 66.3% | 81.0% |
+| `internal/datasource` | 0% | 55.6% |
+| `internal/pamauth` | 0% | 100% |
+| `pkg/ca` | 82.2% | 83.9% |
+
+## Notes
+
+- Two inherently racy assertions were deliberately dropped (broker cancelled-context
+  `ReadNow`; audit closed-channel synchronous fallback): both depended on
+  nondeterministic `select` ordering and flaked under `-race`. The surrounding code
+  paths remain covered by other tests.
+- All new test files are `gofmt`-clean, matching the CI gate
+  (`gofmt -l $(git ls-files '*.go')`).