test-coverage hardening: raise low-ratio subsystems + audit placeholder tests

[danielraffel]

From the 2026-04-16 alpha-hardening audit:

• Static test/code LOC ratio: 0.261 across 744 code files / 151,726 LOC.
• core/dsl: 0.00 — header-only subsystem with no tests. Confirmed: only include/ + CMakeLists.txt.
• Weak-surface subsystems (priority order): core/platform 0.28, core/host 0.32, core/osc 0.37, core/runtime 0.44, core/view 0.51, core/format 0.52.

Scope (prioritized, tackle toward end of production-readiness push)

1. Audit placeholder tests flagged as succeed-only scaffolding. Decide per file: make real or delete.
   • test_validation_harness.cpp, test_text_input.cpp, test_uia_mapping.cpp,
     test_negative_path.cpp, test_tempo_hook.cpp, test_showcase.cpp,
     test_js_engine.cpp, test_websocket_channel.cpp, test_atk_mapping.cpp,
     test_announce.cpp, test_memory_pressure.cpp, test_transport_hook.cpp,
     test_osc_channel.cpp, test_widget_bridge.cpp, test_audio_system_hotplug.cpp,
     test_network_stream.cpp, test_screenshot.cpp, test_token_diff.cpp.
2. Priority subsystem hardening — raise coverage on: platform, host, format, runtime, view, osc, dsl (user-specified priority). Each gets a focused sweep: identify the 3-5 highest-blast-radius entry points, write behaviour tests.
3. core/dsl: decide whether it's live. If yes, add parser + interpreter suite. If dead, delete the subsystem.
4. Regression-test policy: every new bug-fix PR carries a narrow regression test or an explicit "no test practical" note.
5. Runtime coverage job: add an LLVM/gcov CI lane at a later phase; don't block current bugs on coverage infra.

Schedule: tackle after production-readiness workstream PRs have drained (per user: "obv towards end").

[danielraffel]

Placeholder-test audit walked all 18 flagged files in test/:

Real placeholders (should be filled or removed):

• test_showcase.cpp (6 lines, just REQUIRE(true)) — comment says "will be populated when design tool merges". Still pending.
• test_token_diff.cpp (2 lines, just REQUIRE(true)) — no trailing context; looks like it was committed as a stub and forgotten.

False positives (real coverage, flagged on the heuristic because they contain one SUCCEED for a cross-platform no-op branch):

• test_validation_harness (312L / 42 asserts), test_text_input (122L / 11), test_uia_mapping (80L / 28),
  test_negative_path (166L / 9), test_tempo_hook (53L / 5), test_js_engine (408L / 73),
  test_websocket_channel (165L / 18), test_atk_mapping (72L / 24), test_announce (73L / 9),
  test_memory_pressure (55L / 6), test_transport_hook (61L / 8), test_osc_channel (70L / 6),
  test_widget_bridge (693L / 102), test_audio_system_hotplug (58L / 6), test_network_stream (115L / 12),
  test_screenshot (123L / 20).

Recommended action:

1. Delete test_token_diff.cpp outright (no context, no referenced subsystem).
2. Decide on test_showcase.cpp: either wire it to the design-tool example (Production readiness — Workstream 01: Format adapter completeness #214 scope) or delete.
3. Leave the other 16 alone — their SUCCEED markers gate platform-specific branches where "this platform has no implementation" is a correct passing outcome.

[danielraffel]

Correction to earlier placeholder audit: both test_showcase.cpp and test_token_diff.cpp are tied to the AI Designer workstream (CMake comments: "AI Designer: token diff tests", "design tool"). They're genuine scaffolds waiting for the design-tool merge, not abandoned stubs. Leave both alone for now; revisit when the design-tool slice (#214 scope) lands its own tests.