mirror of https://github.com/H-uru/korman.git
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
740 lines
31 KiB
740 lines
31 KiB
# This file is part of Korman. |
|
# |
|
# Korman is free software: you can redistribute it and/or modify |
|
# it under the terms of the GNU General Public License as published by |
|
# the Free Software Foundation, either version 3 of the License, or |
|
# (at your option) any later version. |
|
# |
|
# Korman is distributed in the hope that it will be useful, |
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
|
# GNU General Public License for more details. |
|
# |
|
# You should have received a copy of the GNU General Public License |
|
# along with Korman. If not, see <http://www.gnu.org/licenses/>. |
|
|
|
import bpy |
|
from contextlib import ExitStack |
|
import itertools |
|
from PyHSPlasma import * |
|
from math import fabs |
|
from typing import Iterable |
|
import weakref |
|
|
|
from ..exporter.logger import ExportProgressLogger |
|
from . import explosions |
|
from .. import helpers |
|
from . import material |
|
from . import utils |
|
|
|
_MAX_VERTS_PER_SPAN = 0xFFFF |
|
_WARN_VERTS_PER_SPAN = 0x8000 |
|
|
|
_VERTEX_COLOR_LAYERS = {"col", "color", "colour"} |
|
|
|
class _GeoSpan: |
|
def __init__(self, bo, bm, geospan, pass_index=None): |
|
self.geospan = geospan |
|
self.pass_index = pass_index if pass_index is not None else 0 |
|
self.mult_color = self._determine_mult_color(bo, bm) |
|
|
|
def _determine_mult_color(self, bo, bm): |
|
"""Determines the color all vertex colors should be multipled by in this span.""" |
|
if self.geospan.props & plGeometrySpan.kDiffuseFoldedIn: |
|
color = bm.diffuse_color |
|
base_layer = self.geospan.material.object.layers[0].object.bottomOfStack.object |
|
return (color.r, color.b, color.g, base_layer.opacity) |
|
if not bo.plasma_modifiers.lighting.preshade: |
|
return (0.0, 0.0, 0.0, 0.0) |
|
return (1.0, 1.0, 1.0, 1.0) |
|
|
|
|
|
class _RenderLevel: |
|
MAJOR_OPAQUE = 0 |
|
MAJOR_FRAMEBUF = 1 |
|
MAJOR_DEFAULT = 2 |
|
MAJOR_BLEND = 4 |
|
MAJOR_LATE = 8 |
|
|
|
_MAJOR_SHIFT = 28 |
|
_MINOR_MASK = ((1 << _MAJOR_SHIFT) - 1) |
|
|
|
def __init__(self, bo, pass_index, blend_span=False): |
|
if blend_span: |
|
self.level = self._determine_level(bo, blend_span) |
|
else: |
|
self.level = 0 |
|
# Gulp... Hope you know what you're doing... |
|
self.minor += pass_index * 4 |
|
|
|
def __eq__(self, other): |
|
return self.level == other.level |
|
|
|
def __hash__(self): |
|
return hash(self.level) |
|
|
|
def _get_major(self): |
|
return self.level >> self._MAJOR_SHIFT |
|
def _set_major(self, value): |
|
self.level = self._calc_level(value, self.minor) |
|
major = property(_get_major, _set_major) |
|
|
|
def _get_minor(self): |
|
return self.level & self._MINOR_MASK |
|
def _set_minor(self, value): |
|
self.level = self._calc_level(self.major, value) |
|
minor = property(_get_minor, _set_minor) |
|
|
|
def _calc_level(self, major : int, minor : int=0) -> int: |
|
return ((major << self._MAJOR_SHIFT) & 0xFFFFFFFF) | minor |
|
|
|
def _determine_level(self, bo : bpy.types.Object, blend_span : bool) -> int: |
|
mods = bo.plasma_modifiers |
|
if mods.test_property("draw_framebuf"): |
|
return self._calc_level(self.MAJOR_FRAMEBUF) |
|
elif mods.test_property("draw_opaque"): |
|
return self._calc_level(self.MAJOR_OPAQUE) |
|
elif mods.test_property("draw_late"): |
|
return self._calc_level(self.MAJOR_LATE) |
|
elif mods.test_property("draw_no_defer"): |
|
blend_span = False |
|
|
|
blend_mod = mods.blend |
|
if blend_mod.enabled and blend_mod.has_dependencies: |
|
level = self._calc_level(self.MAJOR_FRAMEBUF) |
|
for i in blend_mod.iter_dependencies(): |
|
level = max(level, self._determine_level(i, blend_span)) |
|
return level + 4 |
|
elif blend_span: |
|
return self._calc_level(self.MAJOR_BLEND) |
|
else: |
|
return self._calc_level(self.MAJOR_DEFAULT) |
|
|
|
|
|
class _DrawableCriteria: |
|
def __init__(self, bo, geospan, pass_index): |
|
self.blend_span = bool(geospan.props & plGeometrySpan.kRequiresBlending) |
|
self.criteria = 0 |
|
|
|
if self.blend_span: |
|
if self._face_sort_allowed(bo): |
|
self.criteria |= plDrawable.kCritSortFaces |
|
if self._span_sort_allowed(bo): |
|
self.criteria |= plDrawable.kCritSortSpans |
|
self.render_level = _RenderLevel(bo, pass_index, self.blend_span) |
|
|
|
def __eq__(self, other): |
|
if not isinstance(other, _DrawableCriteria): |
|
return False |
|
for i in ("blend_span", "render_level", "criteria"): |
|
if getattr(self, i) != getattr(other, i): |
|
return False |
|
return True |
|
|
|
def __hash__(self): |
|
return hash(self.render_level) ^ hash(self.blend_span) ^ hash(self.criteria) |
|
|
|
def _face_sort_allowed(self, bo): |
|
# For now, only test the modifiers |
|
# This will need to be tweaked further for GUIs... |
|
return not bo.plasma_modifiers.test_property("no_face_sort") |
|
|
|
def _span_sort_allowed(self, bo): |
|
# For now, only test the modifiers |
|
# This will need to be tweaked further for GUIs... |
|
return not bo.plasma_modifiers.test_property("no_face_sort") |
|
|
|
@property |
|
def span_type(self): |
|
if self.blend_span: |
|
return "BlendSpans" |
|
else: |
|
return "Spans" |
|
|
|
|
|
class _GeoData: |
|
def __init__(self, numVtxs): |
|
self.blender2gs = [{} for i in range(numVtxs)] |
|
self.triangles = [] |
|
self.vertices = [] |
|
|
|
|
|
class _MeshManager: |
|
def __init__(self, report=None): |
|
self.context_stack = ExitStack() |
|
if report is not None: |
|
self._report = report |
|
self._entered = False |
|
self._overrides = {} |
|
|
|
@staticmethod |
|
def add_progress_presteps(report): |
|
report.progress_add_step("Applying Blender Mods") |
|
|
|
def _build_prop_dict(self, bstruct): |
|
props = {} |
|
for i in bstruct.bl_rna.properties: |
|
ident = i.identifier |
|
if ident == "rna_type": |
|
continue |
|
props[ident] = getattr(bstruct, ident) if getattr(i, "array_length", 0) == 0 else tuple(getattr(bstruct, ident)) |
|
return props |
|
|
|
def __enter__(self): |
|
assert self._entered is False, "_MeshManager is not reentrant" |
|
self._entered = True |
|
|
|
self.context_stack.__enter__() |
|
|
|
scene = bpy.context.scene |
|
self._report.progress_advance() |
|
self._report.progress_range = len(scene.objects) |
|
|
|
# Some modifiers like "Array" will procedurally generate new geometry that will impact |
|
# lightmap generation. The Blender Internal renderer does not seem to be smart enough to |
|
# take this into account. Thus, we temporarily apply modifiers to ALL meshes (even ones that |
|
# are not exported) such that we can generate proper lighting. |
|
mesh_type = bpy.types.Mesh |
|
for i in scene.objects: |
|
if isinstance(i.data, mesh_type) and i.is_modified(scene, "RENDER"): |
|
# Remember, storing actual pointers to the Blender objects can cause bad things to |
|
# happen because Blender's memory management SUCKS! |
|
self._overrides[i.name] = { "mesh": i.data.name, "modifiers": [] } |
|
i.data = i.to_mesh(scene, True, "RENDER", calc_tessface=False) |
|
|
|
# If the modifiers are left on the object, the lightmap bake can break under some |
|
# situations. Therefore, we now cache the modifiers and clear them away... |
|
if i.plasma_object.enabled: |
|
cache_mods = self._overrides[i.name]["modifiers"] |
|
for mod in i.modifiers: |
|
cache_mods.append(self._build_prop_dict(mod)) |
|
i.modifiers.clear() |
|
self._report.progress_increment() |
|
return self |
|
|
|
def __exit__(self, *exc_info): |
|
try: |
|
self.context_stack.__exit__(*exc_info) |
|
finally: |
|
data_bos, data_meshes = bpy.data.objects, bpy.data.meshes |
|
for obj_name, override in self._overrides.items(): |
|
bo = data_bos.get(obj_name) |
|
|
|
# Reapply the old mesh |
|
trash_mesh, bo.data = bo.data, data_meshes.get(override["mesh"]) |
|
data_meshes.remove(trash_mesh) |
|
|
|
# If modifiers were removed, reapply them now unless they're read-only. |
|
readonly_attributes = {("DECIMATE", "face_count"),} |
|
for cached_mod in override["modifiers"]: |
|
mod = bo.modifiers.new(cached_mod["name"], cached_mod["type"]) |
|
for key, value in cached_mod.items(): |
|
if key in {"name", "type"} or (cached_mod["type"], key) in readonly_attributes: |
|
continue |
|
setattr(mod, key, value) |
|
self._entered = False |
|
|
|
def is_collapsed(self, bo) -> bool: |
|
return bo.name in self._overrides |
|
|
|
|
|
class MeshConverter(_MeshManager): |
|
def __init__(self, exporter): |
|
self._exporter = weakref.ref(exporter) |
|
self.material = material.MaterialConverter(exporter) |
|
|
|
self._dspans = {} |
|
self._mesh_geospans = {} |
|
self._non_preshaded = {} |
|
|
|
# _report is a property on this subclass |
|
super().__init__() |
|
|
|
def _calc_num_uvchans(self, bo, mesh): |
|
max_user_texs = plGeometrySpan.kUVCountMask |
|
num_user_texs = len(mesh.tessface_uv_textures) |
|
total_texs = num_user_texs |
|
|
|
# Bump Mapping requires 2 magic channels |
|
if self.material.get_bump_layer(bo) is not None: |
|
total_texs += 2 |
|
max_user_texs -= 2 |
|
|
|
# Lightmapping requires its own LIGHTMAPGEN channel |
|
# NOTE: the LIGHTMAPGEN texture has already been created, so it is in num_user_texs |
|
lm = bo.plasma_modifiers.lightmap |
|
if lm.enabled and lm.bake_type == "lightmap": |
|
num_user_texs -= 1 |
|
max_user_texs -= 1 |
|
|
|
return (num_user_texs, total_texs, max_user_texs) |
|
|
|
def _check_vtx_alpha(self, mesh, material_idx): |
|
if material_idx is not None: |
|
polygons = (i for i in mesh.polygons if i.material_index == material_idx) |
|
else: |
|
polygons = mesh.polygons |
|
alpha_layer = self._find_vtx_alpha_layer(mesh.vertex_colors) |
|
if alpha_layer is None: |
|
return False |
|
alpha_loops = (alpha_layer[i.loop_start:i.loop_start+i.loop_total] for i in polygons) |
|
opaque = (sum(i.color) == len(i.color) for i in itertools.chain.from_iterable(alpha_loops)) |
|
has_alpha = not all(opaque) |
|
return has_alpha |
|
|
|
def _check_vtx_nonpreshaded(self, bo, mesh, material_idx, bm): |
|
def check(): |
|
# TODO: if this is an avatar, we can't be non-preshaded. |
|
# kShadeWhite (used for shadeless) is not handled for kLiteVtxNonPreshaded |
|
if bm is not None: |
|
if bm.use_shadeless: |
|
return False |
|
if self.material.requires_material_shading(bm): |
|
return False |
|
|
|
mods = bo.plasma_modifiers |
|
if mods.lighting.rt_lights: |
|
return True |
|
if mods.lightmap.bake_lightmap: |
|
return True |
|
if self._check_vtx_alpha(mesh, material_idx): |
|
return True |
|
|
|
return False |
|
|
|
# Safe version for inside the mesh converter. |
|
result = self._non_preshaded.get((bo, bm)) |
|
if result is None: |
|
result = check() |
|
self._non_preshaded[(bo, bm)] = result |
|
return result |
|
|
|
def _create_geospan(self, bo, mesh, material_idx, bm, hsgmatKey): |
|
"""Initializes a plGeometrySpan from a Blender Object and an hsGMaterial""" |
|
geospan = plGeometrySpan() |
|
geospan.material = hsgmatKey |
|
|
|
# GeometrySpan format |
|
# For now, we really only care about the number of UVW Channels |
|
user_uvws, total_uvws, max_user_uvws = self._calc_num_uvchans(bo, mesh) |
|
if total_uvws > plGeometrySpan.kUVCountMask: |
|
raise explosions.TooManyUVChannelsError(bo, bm, user_uvws, max_user_uvws) |
|
geospan.format = total_uvws |
|
|
|
def is_alpha_blended(layer): |
|
if layer.state.blendFlags & hsGMatState.kBlendMask: |
|
return True |
|
if layer.underLay is not None: |
|
return is_alpha_blended(layer.underLay.object) |
|
return False |
|
|
|
base_layer = hsgmatKey.object.layers[0].object |
|
if is_alpha_blended(base_layer) or self._check_vtx_alpha(mesh, material_idx): |
|
geospan.props |= plGeometrySpan.kRequiresBlending |
|
if self._check_vtx_nonpreshaded(bo, mesh, material_idx, bm): |
|
geospan.props |= plGeometrySpan.kLiteVtxNonPreshaded |
|
if (geospan.props & plGeometrySpan.kLiteMask) != plGeometrySpan.kLiteMaterial: |
|
geospan.props |= plGeometrySpan.kDiffuseFoldedIn |
|
|
|
mods = bo.plasma_modifiers |
|
if mods.lighting.rt_lights: |
|
geospan.props |= plGeometrySpan.kPropRunTimeLight |
|
if not bm.use_shadows: |
|
geospan.props |= plGeometrySpan.kPropNoShadow |
|
|
|
# Harvest lights |
|
permaLights, permaProjs = self._exporter().light.find_material_light_keys(bo, bm) |
|
for i in permaLights: |
|
geospan.addPermaLight(i) |
|
for i in permaProjs: |
|
geospan.addPermaProj(i) |
|
|
|
return geospan |
|
|
|
def finalize(self): |
|
"""Prepares all baked Plasma geometry to be flushed to the disk""" |
|
self._report.progress_advance() |
|
self._report.progress_range = len(self._dspans) |
|
inc_progress = self._report.progress_increment |
|
log_msg = self._report.msg |
|
indent = self._report.indent |
|
|
|
log_msg("\nFinalizing Geometry") |
|
with indent(): |
|
for loc in self._dspans.values(): |
|
for dspan in loc.values(): |
|
log_msg("[DrawableSpans '{}']", dspan.key.name) |
|
|
|
# This mega-function does a lot: |
|
# 1. Converts SourceSpans (geospans) to Icicles and bakes geometry into plGBuffers |
|
# 2. Calculates the Icicle bounds |
|
# 3. Builds the plSpaceTree |
|
# 4. Clears the SourceSpans |
|
dspan.composeGeometry(True, True) |
|
inc_progress() |
|
|
|
def _export_geometry(self, bo, mesh, materials, geospans, mat2span_LUT): |
|
self._report.msg(f"Converting geometry from '{mesh.name}'...") |
|
|
|
# Recall that materials is a mapping of exported materials to blender material indices. |
|
# Therefore, geodata maps blender material indices to working geometry data. |
|
# Maybe the logic is a bit inverted, but it keeps the inner loop simple. |
|
geodata = { idx: _GeoData(len(mesh.vertices)) for idx, _ in materials } |
|
bumpmap = self.material.get_bump_layer(bo) |
|
|
|
# Locate relevant vertex color layers now... |
|
lm = bo.plasma_modifiers.lightmap |
|
color = self._find_vtx_color_layer(mesh.tessface_vertex_colors, autocolor=not lm.bake_lightmap, manual=True) |
|
alpha = self._find_vtx_alpha_layer(mesh.tessface_vertex_colors) |
|
|
|
# Convert Blender faces into things we can stuff into libHSPlasma |
|
for i, tessface in enumerate(mesh.tessfaces): |
|
data = geodata.get(tessface.material_index) |
|
if data is None: |
|
continue |
|
|
|
face_verts = [] |
|
dPosDu = hsVector3(0.0, 0.0, 0.0) |
|
dPosDv = hsVector3(0.0, 0.0, 0.0) |
|
|
|
# Unpack normals |
|
tessface_normals = tessface.split_normals |
|
|
|
# Unpack the UV coordinates from each UV Texture layer |
|
# NOTE: Blender has no third (W) coordinate |
|
tessface_uvws = [uvtex.data[i].uv for uvtex in mesh.tessface_uv_textures] |
|
|
|
# Unpack colors |
|
if color is None: |
|
tessface_colors = ((1.0, 1.0, 1.0), (1.0, 1.0, 1.0), (1.0, 1.0, 1.0), (1.0, 1.0, 1.0)) |
|
else: |
|
src = color[i] |
|
tessface_colors = (src.color1, src.color2, src.color3, src.color4) |
|
|
|
# Unpack alpha values |
|
if alpha is None: |
|
tessface_alphas = (1.0, 1.0, 1.0, 1.0) |
|
else: |
|
src = alpha[i] |
|
# Some time between 2.79b and 2.80, vertex alpha colors appeared in Blender. However, |
|
# there is no way to actually visually edit them. That means that we need to keep that |
|
# fact in mind because we're just averaging the color to make alpha. |
|
tessface_alphas = ((sum(src.color1[:3]) / 3), (sum(src.color2[:3]) / 3), |
|
(sum(src.color3[:3]) / 3), (sum(src.color4[:3]) / 3)) |
|
|
|
if bumpmap is not None: |
|
gradPass = [] |
|
gradUVWs = [] |
|
|
|
if len(tessface.vertices) != 3: |
|
gradPass.append([tessface.vertices[0], tessface.vertices[1], tessface.vertices[2]]) |
|
gradPass.append([tessface.vertices[0], tessface.vertices[2], tessface.vertices[3]]) |
|
gradUVWs.append((tuple((uvw[0] for uvw in tessface_uvws)), |
|
tuple((uvw[1] for uvw in tessface_uvws)), |
|
tuple((uvw[2] for uvw in tessface_uvws)))) |
|
gradUVWs.append((tuple((uvw[0] for uvw in tessface_uvws)), |
|
tuple((uvw[2] for uvw in tessface_uvws)), |
|
tuple((uvw[3] for uvw in tessface_uvws)))) |
|
else: |
|
gradPass.append(tessface.vertices) |
|
gradUVWs.append((tuple((uvw[0] for uvw in tessface_uvws)), |
|
tuple((uvw[1] for uvw in tessface_uvws)), |
|
tuple((uvw[2] for uvw in tessface_uvws)))) |
|
|
|
for p, vids in enumerate(gradPass): |
|
dPosDu += self._get_bump_gradient(bumpmap[1], gradUVWs[p], mesh, vids, bumpmap[0], 0) |
|
dPosDv += self._get_bump_gradient(bumpmap[1], gradUVWs[p], mesh, vids, bumpmap[0], 1) |
|
dPosDv = -dPosDv |
|
|
|
# Convert to per-material indices |
|
for j, vertex in enumerate(tessface.vertices): |
|
vertex_normal = tuple(tessface_normals[j]) |
|
uvws = tuple([tuple(uvw[j]) for uvw in tessface_uvws]) |
|
|
|
# Calculate vertex colors. |
|
if mat2span_LUT: |
|
mult_color = geospans[mat2span_LUT[tessface.material_index]].mult_color |
|
else: |
|
mult_color = (1.0, 1.0, 1.0, 1.0) |
|
tessface_color, tessface_alpha = tessface_colors[j], tessface_alphas[j] |
|
vertex_color = (int(tessface_color[0] * mult_color[0] * 255), |
|
int(tessface_color[1] * mult_color[1] * 255), |
|
int(tessface_color[2] * mult_color[2] * 255), |
|
int(tessface_alpha * mult_color[0] * 255)) |
|
|
|
# Now, we'll index into the vertex dict using the per-face elements :( |
|
# We're using tuples because lists are not hashable. The many mathutils and PyHSPlasma |
|
# types are not either, and it's entirely too much work to fool with all that. |
|
normcoluv = (vertex_normal, vertex_color, uvws) |
|
if normcoluv not in data.blender2gs[vertex]: |
|
source = mesh.vertices[vertex] |
|
geoVertex = plGeometrySpan.TempVertex() |
|
geoVertex.position = hsVector3(*source.co) |
|
|
|
# MOUL/DX9 craps its pants if any element of the normal is exactly 0.0 |
|
normal = map(lambda x: max(x, 0.01) if x >= 0.0 else min(x, -0.01), vertex_normal) |
|
normal = hsVector3(*normal) |
|
normal.normalize() |
|
geoVertex.normal = normal |
|
|
|
geoVertex.color = hsColor32(*vertex_color) |
|
uvs = [hsVector3(uv[0], 1.0 - uv[1], 0.0) for uv in uvws] |
|
if bumpmap is not None: |
|
uvs.append(dPosDu) |
|
uvs.append(dPosDv) |
|
geoVertex.uvs = uvs |
|
|
|
idx = len(data.vertices) |
|
data.blender2gs[vertex][normcoluv] = idx |
|
data.vertices.append(geoVertex) |
|
face_verts.append(idx) |
|
else: |
|
# If we have a bump mapping layer, then we need to add the bump gradients for |
|
# this face to the vertex's magic channels |
|
if bumpmap is not None: |
|
num_user_uvs = len(uvws) |
|
geoVertex = data.vertices[data.blender2gs[vertex][normcoluv]] |
|
|
|
# Unfortunately, PyHSPlasma returns a copy of everything. Previously, editing |
|
# in place would result in silent failures; however, as of python_refactor, |
|
# PyHSPlasma now returns tuples to indicate this. |
|
geoUVs = list(geoVertex.uvs) |
|
geoUVs[num_user_uvs] += dPosDu |
|
geoUVs[num_user_uvs+1] += dPosDv |
|
geoVertex.uvs = geoUVs |
|
face_verts.append(data.blender2gs[vertex][normcoluv]) |
|
|
|
# Convert to triangles, if need be... |
|
num_faces = len(face_verts) |
|
if num_faces == 3: |
|
data.triangles += face_verts |
|
elif num_faces == 4: |
|
data.triangles += (face_verts[0], face_verts[1], face_verts[2]) |
|
data.triangles += (face_verts[0], face_verts[2], face_verts[3]) |
|
|
|
# Time to finish it up... |
|
for i, data in enumerate(geodata.values()): |
|
geospan = geospans[i].geospan |
|
numVerts = len(data.vertices) |
|
numUVs = geospan.format & plGeometrySpan.kUVCountMask |
|
|
|
# There is a soft limit of 0x8000 vertices per span in Plasma, but the limit is |
|
# theoretically 0xFFFF because this field is a 16-bit integer. However, bad things |
|
# happen in MOUL when we have over 0x8000 vertices. I've also received tons of reports |
|
# of stack dumps in PotS when modifiers are applied, so we're going to limit to 0x8000. |
|
# TODO: consider busting up the mesh into multiple geospans? |
|
# or hack plDrawableSpans::composeGeometry to do it for us? |
|
if numVerts > _WARN_VERTS_PER_SPAN: |
|
raise explosions.TooManyVerticesError(bo.data.name, geospan.material.name, numVerts) |
|
|
|
# If we're bump mapping, we need to normalize our magic UVW channels |
|
if bumpmap is not None: |
|
for vtx in data.vertices: |
|
uvMap = vtx.uvs |
|
uvMap[numUVs - 2].normalize() |
|
uvMap[numUVs - 1].normalize() |
|
vtx.uvs = uvMap |
|
|
|
# If we're still here, let's add our data to the GeometrySpan |
|
geospan.indices = data.triangles |
|
geospan.vertices = data.vertices |
|
|
|
|
|
def _get_bump_gradient(self, xform, uvws, mesh, vIds, uvIdx, iUV): |
|
v0 = hsVector3(*mesh.vertices[vIds[0]].co) |
|
v1 = hsVector3(*mesh.vertices[vIds[1]].co) |
|
v2 = hsVector3(*mesh.vertices[vIds[2]].co) |
|
|
|
uv0 = (uvws[0][uvIdx][0], uvws[0][uvIdx][1], 0.0) |
|
uv1 = (uvws[1][uvIdx][0], uvws[1][uvIdx][1], 0.0) |
|
uv2 = (uvws[2][uvIdx][0], uvws[2][uvIdx][1], 0.0) |
|
|
|
notUV = int(not iUV) |
|
_REAL_SMALL = 0.000001 |
|
|
|
delta = uv0[notUV] - uv1[notUV] |
|
if fabs(delta) < _REAL_SMALL: |
|
return v1 - v0 if uv0[iUV] - uv1[iUV] < 0 else v0 - v1 |
|
|
|
delta = uv2[notUV] - uv1[notUV] |
|
if fabs(delta) < _REAL_SMALL: |
|
return v1 - v2 if uv2[iUV] - uv1[iUV] < 0 else v2 - v1 |
|
|
|
delta = uv2[notUV] - uv0[notUV] |
|
if fabs(delta) < _REAL_SMALL: |
|
return v0 - v2 if uv2[iUV] - uv0[iUV] < 0 else v2 - v0 |
|
|
|
# On to the real fun... |
|
delta = uv0[notUV] - uv1[notUV] |
|
delta = 1.0 / delta |
|
v0Mv1 = v0 - v1 |
|
v0Mv1 *= delta |
|
v0uv = (uv0[iUV] - uv1[iUV]) * delta |
|
|
|
delta = uv2[notUV] - uv1[notUV] |
|
delta = 1.0 / delta |
|
v2Mv1 = v2 - v1 |
|
v2Mv1 *= delta |
|
v2uv = (uv2[iUV] - uv1[iUV]) * delta |
|
|
|
return v0Mv1 - v2Mv1 if v0uv > v2uv else v2Mv1 - v0Mv1 |
|
|
|
def _enumerate_materials(self, bo, mesh): |
|
material_source = mesh.materials |
|
valid_materials = set((tf.material_index for tf in mesh.tessfaces if material_source[tf.material_index] is not None)) |
|
# Sequence of tuples (material_index, material) |
|
return sorted(((i, material_source[i]) for i in valid_materials), key=lambda x: x[0]) |
|
|
|
def export_object(self, bo, so : plSceneObject): |
|
# If this object has modifiers, then it's a unique mesh, and we don't need to try caching it |
|
# Otherwise, let's *try* to share meshes as best we can... |
|
if bo.modifiers: |
|
drawables = self._export_object(bo) |
|
else: |
|
drawables = self._mesh_geospans.get(bo.data, None) |
|
if drawables is None: |
|
drawables = self._export_object(bo) |
|
|
|
# Create the DrawInterface |
|
if drawables: |
|
diface = self._mgr.find_create_object(plDrawInterface, bl=bo, so=so) |
|
for dspan_key, idx in drawables: |
|
diface.addDrawable(dspan_key, idx) |
|
|
|
def _export_object(self, bo): |
|
# Apply all transforms if we don't have a CI. Empirical evidence suggests that simply |
|
# stashing the transform matrices into the spans can be wiped away by plEnableMsg (WTF) |
|
if self._exporter().has_coordiface(bo): |
|
return self._export_mesh(bo, bo.data) |
|
else: |
|
mesh = bo.to_mesh(bpy.context.scene, True, "RENDER", calc_tessface=False) |
|
with helpers.TemporaryObject(mesh, bpy.data.meshes.remove): |
|
utils.transform_mesh(mesh, bo.matrix_world) |
|
return self._export_mesh(bo, mesh) |
|
|
|
def _export_mesh(self, bo, mesh): |
|
mesh.calc_normals_split() |
|
mesh.calc_tessface() |
|
|
|
# Step 0.8: Determine materials needed for export... Three considerations here: |
|
# 1) Some materials can be None, so that's junk. |
|
# 2) Some materials are present but have no valid geometry (D'oh) |
|
# 3) TODO: Materials may be attached to the object, not the mesh. |
|
materials = self._enumerate_materials(bo, mesh) |
|
if not materials: |
|
return None |
|
|
|
# Step 1: Export all of the doggone materials. |
|
geospans, mat2span_LUT = self._export_material_spans(bo, mesh, materials) |
|
|
|
# Step 2: Export Blender mesh data to Plasma GeometrySpans |
|
self._export_geometry(bo, mesh, materials, geospans, mat2span_LUT) |
|
|
|
# Step 3: Add plGeometrySpans to the appropriate DSpan and create indices |
|
_diindices = {} |
|
for i in geospans: |
|
dspan = self._find_create_dspan(bo, i.geospan, i.pass_index) |
|
self._report.msg("Exported hsGMaterial '{}' geometry into '{}'", |
|
i.geospan.material.name, dspan.key.name) |
|
idx = dspan.addSourceSpan(i.geospan) |
|
diidx = _diindices.setdefault(dspan, []) |
|
diidx.append(idx) |
|
|
|
# Step 3.1: Harvest Span indices and create the DIIndices |
|
drawables = [] |
|
for dspan, indices in _diindices.items(): |
|
dii = plDISpanIndex() |
|
dii.indices = indices |
|
idx = dspan.addDIIndex(dii) |
|
drawables.append((dspan.key, idx)) |
|
return drawables |
|
|
|
def _export_material_spans(self, bo, mesh, materials): |
|
"""Exports all Materials and creates plGeometrySpans""" |
|
waveset_mod = bo.plasma_modifiers.water_basic |
|
if waveset_mod.enabled: |
|
if len(materials) > 1: |
|
self._report.warn(f"'{bo.name}' is a WaveSet -- only one material is supported") |
|
blmat = materials[0][1] |
|
self._check_vtx_nonpreshaded(bo, mesh, 0, blmat) |
|
matKey = self.material.export_waveset_material(bo, blmat) |
|
geospan = self._create_geospan(bo, mesh, None, blmat, matKey) |
|
|
|
# FIXME: Can some of this be generalized? |
|
geospan.props |= (plGeometrySpan.kWaterHeight | plGeometrySpan.kLiteVtxNonPreshaded | |
|
plGeometrySpan.kPropReverseSort | plGeometrySpan.kPropNoShadow) |
|
geospan.waterHeight = bo.matrix_world.translation[2] |
|
return [_GeoSpan(bo, blmat, geospan)], None |
|
else: |
|
geospans = [None] * len(materials) |
|
mat2span_LUT = {} |
|
for i, (blmat_idx, blmat) in enumerate(materials): |
|
self._check_vtx_nonpreshaded(bo, mesh, blmat_idx, blmat) |
|
matKey = self.material.export_material(bo, blmat) |
|
geospans[i] = _GeoSpan(bo, blmat, |
|
self._create_geospan(bo, mesh, blmat_idx, blmat, matKey), |
|
blmat.pass_index) |
|
mat2span_LUT[blmat_idx] = i |
|
return geospans, mat2span_LUT |
|
|
|
def _find_create_dspan(self, bo, geospan, pass_index): |
|
location = self._mgr.get_location(bo) |
|
if location not in self._dspans: |
|
self._dspans[location] = {} |
|
|
|
# This is where we figure out which DSpan this goes into. To vaguely summarize the rules... |
|
# BlendSpans: anything with an alpha blended layer |
|
# SortSpans: means we should sort the spans in this DSpan with all other span in this pass |
|
# SortFaces: means we should sort the faces in this span only |
|
# We're using pass index to do just what it was designed for. Cyan has a nicer "depends on" |
|
# draw component, but pass index is the Blender way, so that's what we're doing. |
|
crit = _DrawableCriteria(bo, geospan, pass_index) |
|
|
|
if crit not in self._dspans[location]: |
|
# AgeName_[District_]_Page_RenderLevel_Crit[Blend]Spans |
|
# Just because it's nice to be consistent |
|
node = self._mgr.get_scene_node(location=location) |
|
name = "{}_{:08X}_{:X}{}".format(node.name, crit.render_level.level, crit.criteria, crit.span_type) |
|
dspan = self._mgr.add_object(pl=plDrawableSpans, name=name, loc=location) |
|
|
|
criteria = crit.criteria |
|
dspan.criteria = criteria |
|
if criteria & plDrawable.kCritSortFaces: |
|
dspan.props |= plDrawable.kPropSortFaces |
|
if criteria & plDrawable.kCritSortSpans: |
|
dspan.props |= plDrawable.kPropSortSpans |
|
dspan.renderLevel = crit.render_level.level |
|
dspan.sceneNode = node # AddViaNotify |
|
|
|
self._dspans[location][crit] = dspan |
|
return dspan |
|
else: |
|
return self._dspans[location][crit] |
|
|
|
def _find_vtx_alpha_layer(self, color_collection): |
|
alpha_layer = next((i for i in color_collection if i.name.lower() == "alpha"), None) |
|
if alpha_layer is not None: |
|
return alpha_layer.data |
|
return None |
|
|
|
def _find_vtx_color_layer(self, color_collection, autocolor: bool = True, manual: bool = True): |
|
if manual: |
|
manual_layer = next((i for i in color_collection if i.name.lower() in _VERTEX_COLOR_LAYERS), None) |
|
if manual_layer is not None: |
|
return manual_layer.data |
|
if autocolor: |
|
baked_layer = color_collection.get("autocolor") |
|
if baked_layer is not None: |
|
return baked_layer.data |
|
return None |
|
|
|
def is_nonpreshaded(self, bo: bpy.types.Object, bm: bpy.types.Material) -> bool: |
|
return self._non_preshaded[(bo, bm)] |
|
|
|
@property |
|
def _mgr(self): |
|
return self._exporter().mgr |
|
|
|
@property |
|
def _report(self): |
|
return self._exporter().report
|
|
|