jakub
6a667d8deb
Include the current addon, terrain integrations, tools, regression tests and recorded development evidence.
3.8 KiB
Native inventory callback and BMP ammunition correction
The actual inventory callback check completes 96 assertions and one cycle, with no script/runtime failures. All seven original native injury containers receive the one-argument blood notification through the adapter exactly once with explicit attachment refresh, then detach and receive no late update after menu close. Blood health is unchanged. This is a deliberately emitted native event, not an actual injury, treatment, manual navigation or rendered-appearance test. Its snapshot precedes only the later mounted sampler change; the exact delta is recorded.
The initial BMP diagnostic preserves 48 assertions, zero completions and two fixture failures. It establishes that the exact WCS empty-armament control has no runtime inventory-item component, despite its correct native forward mounting. The separately audited resource declares that component disabled. The additional zero-muzzle diagnostic does not separately identify its entity; the fix therefore proves every required relationship itself.
The production sampler now permits this exact zero-muzzle WCS control to use a unique native structural binding when the reverse inventory binding is absent. Its immediate parent must be in the same sampled hierarchy, its exact physical weapon owner must match, and at most eight local mount components may be examined. Mount, slot, parent container and attached entity must agree. A conflicting nonnull inventory binding, unknown resource, actual muzzle, foreign parent or ambiguous attachment is still refused. The existing periscope rule and real-weapon custody checks remain unchanged. Controls remain in the bounded deduplicated collection before omission, preserving the 32-weapon cap.
The corrected original-BMP fixture completes 60 assertions and one cycle, with no script/runtime failures. The original slot-4 vehicle gives a complete three-weapon, three-muzzle, three-magazine sample: 288 magazine rounds, capacity 290 and two chambered rounds. Its actual queued status later reaches a separate HQ Watch copy through the real radio route, preserving original vehicle/crew identity, report metadata and detached payload values. Queued values are checked for equality, not separately printed as numeric fields. One original convoy is checked; the fixture stops unrelated opening hunter/town admissions and does not prove full fleet reliability or physical interception.
All 309 registered regression programs pass on unchanged inputs. The old sampler and an attached-identity mutation compile and fail their intended runtime assertions; native boundaries are doubled in those offline controls. The exact final-source normal escape route additionally completes 64 assertions and one round, with no script/runtime failures, for the upcoming development handoff. That scripted route protects/repositions its participant, shortens defence and moves extraction; it does not qualify ordinary manual escape, remote clients or flying.
Every completed native run retains 13 known engine/resource errors, so strict clean qualification remains failed. The initial BMP failure retains those plus its two fixture failures. All native runs used the hidden, muted private environment with unchanged source/external inputs and confirmed cleanup. Exact source/report identities, reviews and limits are recorded in native-summary.json and the linked archive directories. No error is suppressed or reclassified as a clean release.
Original missions, staged work and all five prior sealed packages remain preserved. The installed successor runtime has not been changed. The current handoff predates these corrections; a new development handoff is being prepared. Manual play, multiplayer, operational support, comparative FPS and clean release qualification remain open.