#version 430 core
#extension GL_ARB_shader_draw_parameters : require

layout(location = 0) in vec3 aPosition;
layout(location = 1) in vec3 aNormal;
layout(location = 2) in vec2 aTexCoord;

struct InstanceData {
    mat4 transform;
    // Reserved for Phase B.4 follow-up (selection-blink retail-faithful
    // highlight): vec4 highlightColor; — extend stride here, increase the
    // _instanceSsbo upload size in WbDrawDispatcher, add a flat varying out,
    // and consume in mesh_modern.frag.
};

struct BatchData {
    uvec2 textureHandle;   // bindless handle for sampler2DArray
    uint  textureLayer;    // layer index (always 0 for per-instance composites)
    uint  flags;           // reserved — N.5 dispatcher owns all blend state
                           // (glBlendFunc per pass). If a future phase wants
                           // shader-side per-batch additive flag (Decision 2
                           // fallback), encode it here as bit 0.
};

layout(std430, binding = 0) readonly buffer InstanceBuffer {
    InstanceData Instances[];
};

// binding=1 here is the SSBO namespace — distinct from the UBO namespace.
// SceneLighting UBO also uses binding=1 in the fragment shader; GL keeps
// GL_SHADER_STORAGE_BUFFER and GL_UNIFORM_BUFFER binding tables separate.
// Task 10 dispatcher binds:
//   glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, instanceSsbo)
//   glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, batchSsbo)
// Existing SceneLightingUboBinding handles the UBO side.
layout(std430, binding = 1) readonly buffer BatchBuffer {
    BatchData Batches[];
};

// === Phase U.3: per-cell screen-space clip gate (gl_ClipDistance) =============
// Two SSBOs add the clip mechanism without disturbing binding=0/1 above.
//
// binding=2 — SHARED per-frame clip regions, one CellClip per "slot". Uploaded
// ONCE per frame by ClipFrame.UploadShared (shared across WbDrawDispatcher +
// EnvCellRenderer). Slot 0 is RESERVED = no-clip (count 0 ⇒ every plane passes).
//
// binding=3 — PER-RENDERER per-instance slot index, parallel to the binding=0
// instance buffer and indexed by the IDENTICAL per-instance index
// (gl_BaseInstanceARB + gl_InstanceID). instanceClipSlot[i] selects which
// CellClip region instance i is clipped against. Default all-zeros in U.3 ⇒
// every instance maps to slot 0 ⇒ no clipping ⇒ identical render to pre-U.3.
//
// CellClip std430 layout (144 bytes/slot): a uint count + 3 pad uints (16 bytes)
// then vec4 planes[8] (8 × 16 = 128 bytes). vec4 array stride is 16 under std430.
// ClipFrame on the CPU side lays out the bytes to match exactly (verified by
// ClipFrameLayoutTests). A clip-space vertex is INSIDE iff dot(plane, gl_Position)
// >= 0 for every active plane (see ClipPlaneSet for the plane convention).
struct CellClip {
    uint count;
    uint _p0;
    uint _p1;
    uint _p2;
    vec4 planes[8];
};
layout(std430, binding = 2) readonly buffer ClipRegionBuf {
    CellClip clipRegions[];
};
layout(std430, binding = 3) readonly buffer ClipSlotBuf {
    uint instanceClipSlot[];
};

// === Fix B (A7 #3): per-OBJECT light selection — minimize_object_lighting =====
// retail picks up-to-8 point/spot lights PER OBJECT by the object's own position
// (minimize_object_lighting 0x0054d480), so a torch always lights the wall it
// sits on, camera-INDEPENDENTLY. The previous single global nearest-8-to-CAMERA
// UBO set (LightManager.Tick) made a wall brighten as the camera approached
// (its torches swapping into the global top-8). Two SSBOs replace that for
// point/spot lights (the SUN + ambient still come from the SceneLighting UBO):
//
// binding=4 — GLOBAL point/spot light array, uploaded once per frame from
//   LightManager.PointSnapshot. The index of a light here is stable for the frame.
// binding=5 — per-instance light SET: MaxLightsPerObject(8) int indices per
//   instance INTO gLights[] (-1 = unused slot), parallel to the binding=0
//   instance buffer and indexed by the SAME instanceIndex. WbDrawDispatcher fills
//   it once per entity (the set is constant across the entity's parts/tuples).
struct GlobalLight {
    vec4 posAndKind;
    vec4 dirAndRange;
    vec4 colorAndIntensity;
    vec4 coneAngleEtc;
};
layout(std430, binding = 4) readonly buffer GlobalLightBuf {
    GlobalLight gLights[];
};
layout(std430, binding = 5) readonly buffer InstanceLightSetBuf {
    int instanceLightIdx[];   // 8 per instance; -1 = unused
};

// Core profile: redeclare gl_PerVertex so writing gl_ClipDistance[] is legal
// alongside gl_Position. The array is sized 8 to match the CellClip plane budget
// and the GL guarantee (GL_MAX_CLIP_DISTANCES >= 8). The host enables
// GL_CLIP_DISTANCE0..7 once at startup; unused planes are set to +1.0 below so
// they pass everything (no clipping) when the slot's count < 8.
out gl_PerVertex {
    vec4  gl_Position;
    float gl_ClipDistance[8];
};

uniform mat4 uViewProjection;

// Phase Post-A.5 (ISSUE #52, 2026-05-10): per-pass offset into Batches[].
// gl_DrawIDARB resets to 0 at the start of each glMultiDrawElementsIndirect
// call, so the transparent pass — which begins later in the indirect buffer
// — was fetching Batches[0..transparentCount) instead of its actual section
// at Batches[opaqueCount..end). The lifestone crystal (a transparent draw)
// ended up reading the FIRST OPAQUE batch's TextureHandle every frame. As
// the camera moved and the opaque front-to-back sort reordered which group
// landed at BatchData[0], the lifestone's apparent texture flickered to
// whatever was first — frequently the player character's body parts.
//
// WbDrawDispatcher.Draw sets this to 0 before the opaque MDI call and to
// _opaqueDrawCount before the transparent MDI call, matching WorldBuilder's
// uDrawIDOffset pattern in BaseObjectRenderManager.cs line 845.
uniform int uDrawIDOffset;
uniform int uLightingMode;   // A7 Fix D: 0 = OBJECT (plain Lambert + sun), 1 = ENVCELL (half-Lambert wrap, no sun)

// SceneLighting UBO — binding=1 in the UBO namespace (GL keeps the SSBO and UBO
// binding tables separate, so this coexists with the binding=1 BatchBuffer SSBO
// above). IDENTICAL std140 layout to mesh_modern.frag.
//
// A7 (2026-06-15): lighting moved from the FRAGMENT shader to HERE (per-VERTEX) so
// torch/point lights Gouraud-interpolate across each triangle the way retail's
// fixed-function T&L does (D3D DrawEnvCell vertex bake + minimize_object_lighting for
// objects). A per-PIXEL evaluation made a tight bright "spotlight" pool on flat walls;
// per-vertex spreads it into a soft, broad gradient with no hard edge.
struct Light {
    vec4 posAndKind;
    vec4 dirAndRange;
    vec4 colorAndIntensity;
    vec4 coneAngleEtc;
};
layout(std140, binding = 1) uniform SceneLighting {
    Light uLights[8];
    vec4  uCellAmbient;
    vec4  uFogParams;
    vec4  uFogColor;
    vec4  uCameraAndTime;
};

// Faithful calc_point_light (0x0059c8b0) contribution from ONE point/spot light —
// the wrap + norm shape, factored out so the per-object SSBO loop shares it. D =
// light − vertex, used UN-normalised (length = dist); N is the unit vertex normal.
// Returns the RGB to ADD, already per-channel capped to the light's own colour.
vec3 pointContribution(vec3 N, vec3 worldPos, GlobalLight L) {
    int kind     = int(L.posAndKind.w);
    vec3 toL     = L.posAndKind.xyz - worldPos;   // D (un-normalised)
    float distsq = dot(toL, toL);
    float d      = sqrt(distsq);
    float range  = L.dirAndRange.w;               // falloff_eff = Falloff × 1.3
    if (d >= range || range <= 1e-4) return vec3(0.0);
    // A7 Fix D D-3: angular term by lighting path. ENVCELL bake (mode 1) keeps the
    // half-Lambert wrap (lights surfaces angled away, retail calc_point_light); OBJECT
    // mode (0) uses plain Lambert max(0,N·L) so a torch BEHIND a character contributes
    // nothing (retail's hardware path). toL is un-normalised (length d).
    float angular = (uLightingMode == 1)
        ? (1.0 / 1.5) * (dot(N, toL) + 0.5 * d)   // half-Lambert wrap (EnvCell bake)
        : max(0.0, dot(N, toL));                  // plain Lambert (object/hardware)
    if (angular <= 0.0) return vec3(0.0);
    // NORM branch (distance-cube): >1 m → distsq·d ≈ inverse-square soft far halo;
    // <1 m → just d (dodge the near singularity). "Punchy near, soft far."
    float norm      = (distsq > 1.0) ? (distsq * d) : d;
    float intensity = L.colorAndIntensity.w;
    float scale     = (1.0 - d / range) * intensity * (angular / norm);
    if (kind == 2) {
        // Spotlight: hard-edged cos-cone gate layered on the point ramp.
        vec3 Ldir = toL / max(d, 1e-4);
        float cos_edge = cos(L.coneAngleEtc.x * 0.5);
        float cos_l    = dot(-Ldir, L.dirAndRange.xyz);
        if (cos_l <= cos_edge) scale = 0.0;
    }
    // Per-channel no-blowout cap to the light's OWN colour (un-intensity-scaled):
    // a single light can't push a channel past its colour. Summed lit clamped in frag.
    vec3 baseCol = L.colorAndIntensity.xyz;
    return min(scale * baseCol, baseCol);
}

vec3 accumulateLights(vec3 N, vec3 worldPos, int instanceIndex) {
    vec3 lit = uCellAmbient.xyz;

    // SUN / directional — OBJECT path only (mode 0). retail's EnvCell path
    // (minimize_envcell_lighting) enables only dynamic lights, NEVER the sun, so
    // EnvCell walls (mode 1) get no directional sun wash (A7 Fix D D-4).
    if (uLightingMode == 0) {
        int activeLights = int(uCellAmbient.w);
        for (int i = 0; i < 8; ++i) {
            if (i >= activeLights) break;
            if (int(uLights[i].posAndKind.w) != 0) continue;   // directional only
            vec3 Ldir = -uLights[i].dirAndRange.xyz;
            float ndl = max(0.0, dot(N, Ldir));
            lit += uLights[i].colorAndIntensity.xyz * uLights[i].colorAndIntensity.w * ndl;
        }
    }

    // POINT / SPOT torches: their OWN accumulator (A7 Fix D, D-1). Retail's
    // SetStaticLightingVertexColors sums the static point lights from BLACK and
    // clamps the SUM to [0,1] before anything else (a baked emissive term), so a
    // few warm intensity-100 torches can't push the whole pixel to white the way
    // folding them into ambient+sun did. Mirrors LightBake.ComputeVertexColor
    // (LightBakeConformanceTests). Per-light cap inside pointContribution is unchanged.
    vec3 pointAcc = vec3(0.0);
    int base = instanceIndex * 8;
    for (int k = 0; k < 8; ++k) {
        int gi = instanceLightIdx[base + k];
        if (gi < 0) continue;
        pointAcc += pointContribution(N, worldPos, gLights[gi]);
    }
    lit += min(pointAcc, vec3(1.0));   // clamp the torch sum on its own (retail baked emissive)

    return lit;                        // frag still does the final min(lit, 1.0)
}

out vec3 vNormal;
out vec2 vTexCoord;
out vec3 vWorldPos;
out vec3 vLit;            // A7: per-vertex Gouraud lighting (ambient + capped lights)
out flat uvec2 vTextureHandle;
out flat uint  vTextureLayer;

void main() {
    int instanceIndex = gl_BaseInstanceARB + gl_InstanceID;
    mat4 model = Instances[instanceIndex].transform;

    vec4 worldPos = model * vec4(aPosition, 1.0);
    gl_Position = uViewProjection * worldPos;

    // Phase U.3: per-instance clip gate. instanceClipSlot is indexed by the
    // SAME instanceIndex used for the binding=0 transform above, so the slot
    // travels with the instance through the MDI BaseInstance offsets. Slot 0
    // (the U.3 default) has count 0 ⇒ the second loop sets all 8 distances to
    // +1.0 ⇒ nothing is clipped.
    uint _slot = instanceClipSlot[instanceIndex];
    CellClip _c = clipRegions[_slot];
    for (uint i = 0u; i < _c.count; ++i)
        gl_ClipDistance[i] = dot(_c.planes[i], gl_Position);
    for (uint i = _c.count; i < 8u; ++i)
        gl_ClipDistance[i] = 1.0;

    vWorldPos = worldPos.xyz;
    vNormal = normalize(mat3(model) * aNormal);
    vLit = accumulateLights(vNormal, vWorldPos, instanceIndex);   // A7: per-vertex Gouraud (per-object lights)
    vTexCoord = aTexCoord;

    BatchData b = Batches[uDrawIDOffset + gl_DrawIDARB];
    vTextureHandle = b.textureHandle;
    vTextureLayer = b.textureLayer;
}