(remote viewing) updates to the mc package

[verbiiii]

1. add inventory and items API with parsed meta json

[verbiiii]

this video was published at approximately this point in the code https://youtu.be/3G06jvit6Po

[verbiiii]

Minecraft Rendering Notes

This document tracks the current split vision stack and the historical commit lineage.

Current branch snapshot:

• Current commit: 5020512c394d8c4f4b27b93d6f6659cfc16606a0
• Both renderers now exist in the same branch/commit:
  • EntityVision.java
  • WorldVision.java
• Runtime selection is currently in XosViewportRuntime.renderAgentViewRgbaHex(...) and flushDeferredVisionCaptures(...) with an easy comment swap.

---

1) Entity Rendering

What EntityVision does now

EntityVision is the camera-driven render-pass path intended for accurate POV rendering:

• Per-camera context (CameraContext) keyed by agent UUID.
• Offscreen TextureTarget lifecycle management.
• Render-thread scheduling (queueCapture + flushDeferredCaptures) so world pass work stays on the correct thread.
• Capture using actual game render invocation (renderLevel) and readback.

Tiny internals example:

camera.target = new TextureTarget(width, height, true, Minecraft.ON_OSX);
mc.setCameraEntity(cameraEntity);
gameRendererRenderLevelMethod.invoke(mc.gameRenderer, args);
GL11C.glReadPixels(..., GL11C.GL_RGBA, GL11C.GL_UNSIGNED_BYTE, buf);

Why EntityVision is the "accurate entity" direction:

• It is camera-entity based (agent POV), not just projected world math.
• It uses renderer passes rather than hand-authored entity shapes.
• It keeps per-camera state and deferred capture semantics needed for correctness.

---

2) World Rendering

What WorldVision does now (ported from 4c790a70... lineage)

WorldVision keeps the block-state ray/clip approach from the b30d00d9... and 4c790a70... line:

• Per-pixel world ray casting using ClipContext.
• Block color from MapColor with registry-hash fallback.
• Distance and face-based shading for depth/readability.
• Entity overlay rendered as projected disks (orbs) with depth checks.

Tiny world hit example:

var hit = mc.level.clip(new ClipContext(origin, rayEnd, OUTLINE, NONE, agent));
BlockPos bp = hit.getBlockPos();
int rgb = rgbFromBlockState(mc, bp, state);

Tiny orb overlay example:

int color = rgbFromEntity(e);
if ((dx * dx + dy * dy) <= rr && camZ < depth[p]) {
  // paint projected entity disk
}

What WorldVision gives:

• World state structure that is close to live block layout.
• Stable, readable rendering and predictable performance shape.
• Explicit fallback behavior that is easy to reason about.

What WorldVision does not provide:

• Mesh/textured entity rendering.
• Full native-frame parity with Minecraft's standard on-screen camera pass.

---

EntityVision vs WorldVision

Both renderers now exist side by side in one branch so each can be stabilized independently.

• EntityVision

  • Camera/render-pass architecture (renderLevel + offscreen target + readback).
  • Best path toward true POV fidelity (including proper entity rendering).
  • More sensitive to render-thread/target-state correctness.

• WorldVision

  • Deterministic ray/clip world sampler with shaded block colors.
  • Entity visibility via synthetic projected orbs.
  • Easier to reason about for world-state output and debugging.

This split is intentional: verify each path independently, then merge strengths into one robust final renderer.