plasma-prp 0.1.0

//! plClusterGroup parser — extracts instanced vegetation geometry from .prp data
//!
//! C++ ref: plClusterGroup.cpp, plCluster.cpp, plSpanTemplate.cpp, plSpanInstance.cpp
//! plClusterGroup contains a shared vertex template (plSpanTemplate) that is stamped
//! at many world positions via plSpanInstance transforms. At runtime, C++ unpacks these
//! into plDrawableSpans with kLiteVtxNonPreshaded | kPropRunTimeLight.

use anyhow::{Result, bail};
use std::io::{Read, Seek};
use crate::resource::prp::PlasmaRead;
use crate::core::uoid::{Uoid, read_key_uoid};

/// Vertex produced by cluster unpacking — position, normal, color, UVs, flags.
/// Mirrors the engine renderer's Vertex layout so callers can transmute or convert.
#[derive(Debug, Clone, Copy)]
#[repr(C)]
pub struct ClusterVertex {
    pub position: [f32; 4],
    pub normal: [f32; 4],
    pub color: [f32; 4],
    pub uv: [f32; 2],
    pub uv2: [f32; 2],
    pub uv3: [f32; 2],
    pub flags: f32,
    pub _pad: f32,
    pub uv4: [f32; 2],
    pub uv5: [f32; 2],
}

/// Parsed cluster group — template geometry + per-instance transforms
#[derive(Debug)]
pub struct ClusterGroup {
    pub template: SpanTemplate,
    pub material_uoid: Option<Uoid>,
    pub clusters: Vec<Cluster>,
}

/// Shared vertex/index template for all instances in a cluster group
#[derive(Debug)]
pub struct SpanTemplate {
    pub format: u16,
    pub num_verts: u16,
    pub num_tris: u16,
    pub stride: usize,
    pub vert_data: Vec<u8>,
    pub index_data: Vec<u16>,
}

/// A cluster contains multiple span instances sharing the same template
#[derive(Debug)]
pub struct Cluster {
    pub encoding: SpanEncoding,
    pub instances: Vec<SpanInstance>,
}

/// Encoding descriptor for per-instance position deltas and colors
#[derive(Debug, Clone, Copy)]
pub struct SpanEncoding {
    pub code: u8,
    pub pos_scale: f32,
}

/// A single instance: 3x4 transform matrix + optional position deltas + optional colors
#[derive(Debug)]
pub struct SpanInstance {
    pub l2w: [[f32; 4]; 3],
    pub pos_delta: Option<Vec<u8>>,
    pub col: Option<Vec<u8>>,
}

// SpanEncoding position format flags
const POS_NONE: u8 = 0x0;
const POS_888: u8 = 0x1;
const POS_161616: u8 = 0x2;
const POS_101010: u8 = 0x4;
const POS_008: u8 = 0x8;
const POS_MASK: u8 = POS_888 | POS_161616 | POS_101010 | POS_008;

// SpanEncoding color format flags
const COL_NONE: u16 = 0x0;
const COL_A8: u16 = 0x10;
const COL_I8: u16 = 0x20;
const COL_AI88: u16 = 0x40;
const COL_RGB888: u16 = 0x80;
const COL_ARGB8888: u16 = 0x100;
const COL_MASK: u16 = COL_A8 | COL_I8 | COL_AI88 | COL_RGB888 | COL_ARGB8888;

// SpanTemplate format flags
const TMPL_POS: u16 = 0x1;
const TMPL_NORM: u16 = 0x2;
const TMPL_COLOR: u16 = 0x4;
const TMPL_WGTIDX: u16 = 0x8;
const TMPL_UVW_MASK: u16 = 0xF0;
const TMPL_WEIGHT_MASK: u16 = 0x300;
const TMPL_COLOR2: u16 = 0x400;

impl SpanEncoding {
    fn pos_stride(&self) -> usize {
        match self.code & POS_MASK {
            POS_888 => 3,
            POS_161616 => 6,
            POS_101010 => 4,
            POS_008 => 1,
            _ => 0,
        }
    }

    fn col_stride(&self) -> usize {
        let code = self.code as u16;
        match code & COL_MASK {
            COL_A8 | COL_I8 => 1,
            COL_AI88 => 2,
            COL_RGB888 => 3,
            COL_ARGB8888 => 4,
            _ => 0,
        }
    }
}

impl SpanTemplate {
    pub fn num_uvws(&self) -> usize { ((self.format & TMPL_UVW_MASK) >> 4) as usize }
    fn num_weights(&self) -> usize { ((self.format & TMPL_WEIGHT_MASK) >> 8) as usize }
    fn has_pos(&self) -> bool { self.format & TMPL_POS != 0 }
    fn has_norm(&self) -> bool { self.format & TMPL_NORM != 0 }
    fn has_color(&self) -> bool { self.format & TMPL_COLOR != 0 }
    fn has_color2(&self) -> bool { self.format & TMPL_COLOR2 != 0 }
    fn has_wgt_idx(&self) -> bool { self.format & TMPL_WGTIDX != 0 }

    fn calc_stride(format: u16) -> usize {
        let mut s = 0usize;
        if format & TMPL_POS != 0 { s += 12; } // hsPoint3
        let num_weights = ((format & TMPL_WEIGHT_MASK) >> 8) as usize;
        s += num_weights * 4; // float per weight
        if format & TMPL_WGTIDX != 0 { s += 4; } // uint32
        if format & TMPL_NORM != 0 { s += 12; } // hsVector3
        if format & TMPL_COLOR != 0 { s += 4; } // uint32
        if format & TMPL_COLOR2 != 0 { s += 4; } // uint32
        let num_uvws = ((format & TMPL_UVW_MASK) >> 4) as usize;
        s += num_uvws * 12; // hsPoint3 per UVW
        s
    }

    // Offsets into per-vertex data
    fn pos_offset(&self) -> usize { 0 }
    fn norm_offset(&self) -> usize {
        let mut o = if self.has_pos() { 12 } else { 0 };
        o += self.num_weights() * 4;
        if self.has_wgt_idx() { o += 4; }
        o
    }
    fn color_offset(&self) -> usize {
        self.norm_offset() + if self.has_norm() { 12 } else { 0 }
    }
    fn color2_offset(&self) -> usize {
        self.color_offset() + if self.has_color() { 4 } else { 0 }
    }
    fn uvw_offset(&self) -> usize {
        self.color2_offset() + if self.has_color2() { 4 } else { 0 }
    }
}

impl ClusterGroup {
    /// Parse a plClusterGroup from object data.
    /// C++ ref: plClusterGroup::Read (plClusterGroup.cpp:103-136)
    pub fn read(reader: &mut (impl Read + Seek)) -> Result<Self> {
        // hsKeyedObject::Read — read self-key
        let _self_key = read_key_uoid(reader)?;

        // Read plSpanTemplate
        let template = SpanTemplate::read(reader)?;

        // Read material key ref (mgr->ReadKeyNotifyMe)
        let material_uoid = read_key_uoid(reader)?;

        // Read clusters
        let num_clusters = reader.read_u32()? as usize;
        let mut clusters = Vec::with_capacity(num_clusters);
        for _ in 0..num_clusters {
            clusters.push(Cluster::read(reader, template.num_verts)?);
        }

        // Skip vis regions (key refs)
        let num_regions = reader.read_u32()? as usize;
        for _ in 0..num_regions {
            let _ = read_key_uoid(reader)?;
        }

        // Skip light refs (key refs)
        let num_lights = reader.read_u32()? as usize;
        for _ in 0..num_lights {
            let _ = read_key_uoid(reader)?;
        }

        // LOD distances
        let _min_dist = reader.read_f32()?;
        let _max_dist = reader.read_f32()?;

        // Render level
        let _render_level = reader.read_u32()?;

        // Scene node key
        let _ = read_key_uoid(reader)?;

        let total_insts: usize = clusters.iter().map(|c| c.instances.len()).sum();
        log::debug!("ClusterGroup: {} clusters, {} total instances, template: {} verts, {} tris, format=0x{:X}",
            clusters.len(), total_insts, template.num_verts, template.num_tris, template.format);

        Ok(Self {
            template,
            material_uoid,
            clusters,
        })
    }

    /// Unpack all instances into individual meshes (vertices + indices).
    /// Each cluster becomes one mesh with all its instances merged.
    /// C++ ref: plCluster::UnPack (plCluster.cpp:112-207)
    pub fn unpack_meshes(&self) -> Vec<(Vec<ClusterVertex>, Vec<u16>)> {
        let templ = &self.template;
        let verts_per_inst = templ.num_verts as usize;
        let indices_per_inst = templ.num_tris as usize * 3;

        let mut result = Vec::new();

        for cluster in &self.clusters {
            let num_insts = cluster.instances.len();
            if num_insts == 0 { continue; }

            let mut all_verts = Vec::with_capacity(verts_per_inst * num_insts);
            let mut all_indices = Vec::with_capacity(indices_per_inst * num_insts);
            let mut idx_offset = 0u16;

            for inst in &cluster.instances {
                // Copy template indices with offset
                for &idx in &templ.index_data {
                    all_indices.push(idx + idx_offset);
                }
                idx_offset += verts_per_inst as u16;

                // Build local-to-world matrix (3x4 → 4x4 row-major flat)
                let l2w = &inst.l2w;
                let m: [f32; 16] = [
                    l2w[0][0], l2w[0][1], l2w[0][2], l2w[0][3],
                    l2w[1][0], l2w[1][1], l2w[1][2], l2w[1][3],
                    l2w[2][0], l2w[2][1], l2w[2][2], l2w[2][3],
                    0.0, 0.0, 0.0, 1.0,
                ];

                // Compute inverse-transpose for normal transform
                // For a 3x3 rotation+scale, transpose of inverse ≈ cofactor matrix
                let w2l_t = transpose_inverse_3x3(&m);

                // Create position delta iterator
                let enc = &cluster.encoding;
                let pos_stride = enc.pos_stride();
                let col_stride = enc.col_stride();

                for vi in 0..verts_per_inst {
                    let base = vi * templ.stride;

                    // Read position from template
                    let pos_off = templ.pos_offset();
                    let (mut px, mut py, mut pz) = if templ.has_pos() && base + pos_off + 12 <= templ.vert_data.len() {
                        let o = base + pos_off;
                        (
                            f32::from_le_bytes(templ.vert_data[o..o+4].try_into().unwrap()),
                            f32::from_le_bytes(templ.vert_data[o+4..o+8].try_into().unwrap()),
                            f32::from_le_bytes(templ.vert_data[o+8..o+12].try_into().unwrap()),
                        )
                    } else {
                        (0.0, 0.0, 0.0)
                    };

                    // Apply position delta if present
                    if let Some(ref pos_data) = inst.pos_delta {
                        let delta_off = vi * pos_stride;
                        let (dx, dy, dz) = decode_pos_delta(enc, pos_data, delta_off);
                        px += dx;
                        py += dy;
                        pz += dz;
                    }

                    // Transform position by L2W
                    let wx = px * m[0] + py * m[1] + pz * m[2] + m[3];
                    let wy = px * m[4] + py * m[5] + pz * m[6] + m[7];
                    let wz = px * m[8] + py * m[9] + pz * m[10] + m[11];

                    // Read normal from template
                    let norm_off = templ.norm_offset();
                    let (nx, ny, nz) = if templ.has_norm() && base + norm_off + 12 <= templ.vert_data.len() {
                        let o = base + norm_off;
                        (
                            f32::from_le_bytes(templ.vert_data[o..o+4].try_into().unwrap()),
                            f32::from_le_bytes(templ.vert_data[o+4..o+8].try_into().unwrap()),
                            f32::from_le_bytes(templ.vert_data[o+8..o+12].try_into().unwrap()),
                        )
                    } else {
                        (0.0, 1.0, 0.0)
                    };

                    // Transform normal by inverse-transpose
                    let wnx = nx * w2l_t[0] + ny * w2l_t[1] + nz * w2l_t[2];
                    let wny = nx * w2l_t[3] + ny * w2l_t[4] + nz * w2l_t[5];
                    let wnz = nx * w2l_t[6] + ny * w2l_t[7] + nz * w2l_t[8];
                    let nlen = (wnx*wnx + wny*wny + wnz*wnz).sqrt().max(1e-10);

                    // Read color from template, apply instance color override
                    let (r, g, b, a) = if templ.has_color() {
                        let col_off = templ.color_offset();
                        let o = base + col_off;
                        if o + 4 <= templ.vert_data.len() {
                            let mut c = u32::from_le_bytes(templ.vert_data[o..o+4].try_into().unwrap());
                            // Apply instance color override
                            if let Some(ref col_data) = inst.col {
                                c = decode_color(enc, col_data, vi * col_stride, c);
                            }
                            // D3DCOLOR format: 0xAARRGGBB (LE bytes: B,G,R,A)
                            let ca = ((c >> 24) & 0xFF) as f32 / 255.0;
                            let cr = ((c >> 16) & 0xFF) as f32 / 255.0;
                            let cg = ((c >> 8) & 0xFF) as f32 / 255.0;
                            let cb = (c & 0xFF) as f32 / 255.0;
                            (cr, cg, cb, ca)
                        } else {
                            (1.0, 1.0, 1.0, 1.0)
                        }
                    } else {
                        (1.0, 1.0, 1.0, 1.0)
                    };

                    // Read UV0 from template
                    let uvw_off = templ.uvw_offset();
                    let (u, v) = if templ.num_uvws() > 0 {
                        let o = base + uvw_off;
                        if o + 8 <= templ.vert_data.len() {
                            (
                                f32::from_le_bytes(templ.vert_data[o..o+4].try_into().unwrap()),
                                f32::from_le_bytes(templ.vert_data[o+4..o+8].try_into().unwrap()),
                            )
                        } else {
                            (0.0, 0.0)
                        }
                    } else {
                        (0.0, 0.0)
                    };

                    // Read UV1 from template
                    let (u2, v2) = if templ.num_uvws() > 1 {
                        let o = base + uvw_off + 12; // skip UV0 (hsPoint3 = 12 bytes)
                        if o + 8 <= templ.vert_data.len() {
                            (
                                f32::from_le_bytes(templ.vert_data[o..o+4].try_into().unwrap()),
                                f32::from_le_bytes(templ.vert_data[o+4..o+8].try_into().unwrap()),
                            )
                        } else {
                            (0.0, 0.0)
                        }
                    } else {
                        (0.0, 0.0)
                    };

                    all_verts.push(ClusterVertex {
                        position: [wx, wy, wz, 0.0],
                        normal: [wnx / nlen, wny / nlen, wnz / nlen, 0.0],
                        color: [r, g, b, a],
                        uv: [u, v],
                        uv2: [u2, v2],
                        uv3: [0.0, 0.0],
                        // kLiteVtxNonPreshaded flag = bit 1 (2.0)
                        flags: 2.0,
                        _pad: 0.0,
                        uv4: [0.0, 0.0],
                        uv5: [0.0, 0.0],
                    });
                }
            }

            if !all_verts.is_empty() && !all_indices.is_empty() {
                result.push((all_verts, all_indices));
            }
        }

        result
    }
}

impl SpanTemplate {
    fn read(reader: &mut (impl Read + Seek)) -> Result<Self> {
        let num_verts = reader.read_u16()?;
        let format = reader.read_u16()?;
        let num_tris = reader.read_u16()?;

        let stride = Self::calc_stride(format);
        let vert_size = num_verts as usize * stride;
        let idx_size = num_tris as usize * 3;

        let mut vert_data = vec![0u8; vert_size];
        reader.read_exact(&mut vert_data)?;

        let mut index_data = Vec::with_capacity(idx_size);
        for _ in 0..idx_size {
            index_data.push(reader.read_u16()?);
        }

        Ok(Self { format, num_verts, num_tris, stride, vert_data, index_data })
    }
}

impl Cluster {
    fn read(reader: &mut (impl Read + Seek), num_template_verts: u16) -> Result<Self> {
        // plSpanEncoding
        let code = reader.read_u8()?;
        let pos_scale = reader.read_f32()?;
        let encoding = SpanEncoding { code, pos_scale };

        let num_verts = num_template_verts as usize;
        let num_insts = reader.read_u32()? as usize;
        let mut instances = Vec::with_capacity(num_insts);

        let pos_stride = encoding.pos_stride();
        let col_stride = encoding.col_stride();

        for _ in 0..num_insts {
            // 3x4 transform matrix (12 floats)
            let mut l2w = [[0.0f32; 4]; 3];
            for row in &mut l2w {
                for col in row.iter_mut() {
                    *col = reader.read_f32()?;
                }
            }

            // Position deltas (optional, based on encoding)
            let pos_delta = if pos_stride > 0 {
                let size = num_verts * pos_stride;
                let mut data = vec![0u8; size];
                reader.read_exact(&mut data)?;
                Some(data)
            } else {
                None
            };

            // Color data (optional, based on encoding)
            let col = if col_stride > 0 {
                let size = num_verts * col_stride;
                let mut data = vec![0u8; size];
                reader.read_exact(&mut data)?;
                Some(data)
            } else {
                None
            };

            instances.push(SpanInstance { l2w, pos_delta, col });
        }

        Ok(Self { encoding, instances })
    }
}

/// Decode position delta from encoded data.
/// C++ ref: plSpanInstanceIter::DelPos (plSpanInstance.h:246-266)
fn decode_pos_delta(enc: &SpanEncoding, data: &[u8], offset: usize) -> (f32, f32, f32) {
    match enc.code & POS_MASK {
        POS_888 => {
            if offset + 3 <= data.len() {
                let dx = data[offset] as i8 as f32 * enc.pos_scale;
                let dy = data[offset + 1] as i8 as f32 * enc.pos_scale;
                let dz = data[offset + 2] as i8 as f32 * enc.pos_scale;
                (dx, dy, dz)
            } else { (0.0, 0.0, 0.0) }
        }
        POS_161616 => {
            if offset + 6 <= data.len() {
                let dx = i16::from_le_bytes(data[offset..offset+2].try_into().unwrap()) as f32 * enc.pos_scale;
                let dy = i16::from_le_bytes(data[offset+2..offset+4].try_into().unwrap()) as f32 * enc.pos_scale;
                let dz = i16::from_le_bytes(data[offset+4..offset+6].try_into().unwrap()) as f32 * enc.pos_scale;
                (dx, dy, dz)
            } else { (0.0, 0.0, 0.0) }
        }
        POS_101010 => {
            if offset + 4 <= data.len() {
                let packed = u32::from_le_bytes(data[offset..offset+4].try_into().unwrap());
                let dx = (packed & 0x3F) as f32 * enc.pos_scale;
                let dy = ((packed >> 10) & 0x3F) as f32 * enc.pos_scale;
                let dz = ((packed >> 20) & 0x3F) as f32 * enc.pos_scale;
                (dx, dy, dz)
            } else { (0.0, 0.0, 0.0) }
        }
        POS_008 => {
            if offset < data.len() {
                let dz = data[offset] as i8 as f32 * enc.pos_scale;
                (0.0, 0.0, dz)
            } else { (0.0, 0.0, 0.0) }
        }
        _ => (0.0, 0.0, 0.0),
    }
}

/// Decode color from encoded data, merging with template color.
/// C++ ref: plSpanInstanceIter::Color (plSpanInstance.h:271-299)
fn decode_color(enc: &SpanEncoding, data: &[u8], offset: usize, template_color: u32) -> u32 {
    let code = enc.code as u16;
    match code & COL_MASK {
        COL_A8 => {
            if offset < data.len() {
                (template_color & 0x00FFFFFF) | ((data[offset] as u32) << 24)
            } else { template_color }
        }
        COL_I8 => {
            if offset < data.len() {
                let v = data[offset] as u32;
                (template_color & 0xFF000000) | (v << 16) | (v << 8) | v
            } else { template_color }
        }
        COL_AI88 => {
            if offset + 2 <= data.len() {
                let val = u16::from_le_bytes(data[offset..offset+2].try_into().unwrap());
                let col = (val & 0xFF) as u32;
                let alpha = ((val & 0xFF00) >> 8) as u32;
                (alpha << 24) | (col << 16) | (col << 8) | col
            } else { template_color }
        }
        COL_RGB888 => {
            if offset + 3 <= data.len() {
                (template_color & 0xFF000000)
                    | ((data[offset] as u32) << 16)
                    | ((data[offset + 1] as u32) << 8)
                    | (data[offset + 2] as u32)
            } else { template_color }
        }
        COL_ARGB8888 => {
            if offset + 4 <= data.len() {
                u32::from_le_bytes(data[offset..offset+4].try_into().unwrap())
            } else { template_color }
        }
        _ => template_color,
    }
}

/// Compute 3x3 inverse-transpose for normal transformation.
/// Returns [row0_x, row0_y, row0_z, row1_x, row1_y, row1_z, row2_x, row2_y, row2_z]
fn transpose_inverse_3x3(m: &[f32; 16]) -> [f32; 9] {
    // Extract 3x3 rotation/scale from row-major 4x4
    let a = m[0]; let b = m[1]; let c = m[2];
    let d = m[4]; let e = m[5]; let f = m[6];
    let g = m[8]; let h = m[9]; let i = m[10];

    let det = a*(e*i - f*h) - b*(d*i - f*g) + c*(d*h - e*g);
    if det.abs() < 1e-10 {
        return [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0];
    }
    let inv_det = 1.0 / det;

    // Cofactor matrix (which is the transpose of the inverse × det)
    // Then transpose that to get inverse-transpose
    // inverse-transpose = cofactor / det ... but we want cofactor^T / det
    // Actually: (M^-1)^T = cofactor(M) / det
    // So inverse-transpose row i, col j = cofactor(i,j) / det
    [
        (e*i - f*h) * inv_det, (f*g - d*i) * inv_det, (d*h - e*g) * inv_det,
        (c*h - b*i) * inv_det, (a*i - c*g) * inv_det, (b*g - a*h) * inv_det,
        (b*f - c*e) * inv_det, (c*d - a*f) * inv_det, (a*e - b*d) * inv_det,
    ]
}