summaryrefslogtreecommitdiff
path: root/indra/llprimitive/llprimitive.cpp
diff options
context:
space:
mode:
authorAndrey Lihatskiy <alihatskiy@productengine.com>2024-05-01 08:16:58 +0300
committerAndrey Lihatskiy <alihatskiy@productengine.com>2024-05-01 08:16:58 +0300
commit38c2a5bde985a6a8a96d912d432f8bdf7e5b60be (patch)
treeb3469444ea8dabe4e76a8a265ac086a9db78891c /indra/llprimitive/llprimitive.cpp
parent9bf2dfbb39032d7407295089cf181de0987083e5 (diff)
parente7eced3c87310b15ac20cc3cd470d67686104a14 (diff)
Merge branch 'marchcat/w-whitespace' into marchcat/x-ws-merge
Diffstat (limited to 'indra/llprimitive/llprimitive.cpp')
-rw-r--r--indra/llprimitive/llprimitive.cpp2892
1 files changed, 1446 insertions, 1446 deletions
diff --git a/indra/llprimitive/llprimitive.cpp b/indra/llprimitive/llprimitive.cpp
index 91b9e4f84e..6258a4dc1e 100644
--- a/indra/llprimitive/llprimitive.cpp
+++ b/indra/llprimitive/llprimitive.cpp
@@ -1,25 +1,25 @@
-/**
+/**
* @file llprimitive.cpp
* @brief LLPrimitive base class
*
* $LicenseInfo:firstyear=2001&license=viewerlgpl$
* Second Life Viewer Source Code
* Copyright (C) 2010, Linden Research, Inc.
- *
+ *
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation;
* version 2.1 of the License only.
- *
+ *
* This library 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
* Lesser General Public License for more details.
- *
+ *
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- *
+ *
* Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA
* $/LicenseInfo$
*/
@@ -57,9 +57,9 @@ const F32 OBJECT_TWIST_INC = 18.f;
// This is used for linear paths,
// since twist is used in a slightly different manner.
-const F32 OBJECT_TWIST_LINEAR_MIN = -180.f;
-const F32 OBJECT_TWIST_LINEAR_MAX = 180.f;
-const F32 OBJECT_TWIST_LINEAR_INC = 9.f;
+const F32 OBJECT_TWIST_LINEAR_MIN = -180.f;
+const F32 OBJECT_TWIST_LINEAR_MAX = 180.f;
+const F32 OBJECT_TWIST_LINEAR_INC = 9.f;
const F32 OBJECT_MIN_HOLE_SIZE = 0.05f;
const F32 OBJECT_MAX_HOLE_SIZE_X = 1.0f;
@@ -94,7 +94,7 @@ const F32 REFLECTION_PROBE_DEFAULT_CLIP_DISTANCE = 0.f;
// "Tension" => [0,10], increments of 0.1
const F32 FLEXIBLE_OBJECT_MIN_TENSION = 0.0f;
const F32 FLEXIBLE_OBJECT_DEFAULT_TENSION = 1.0f;
-const F32 FLEXIBLE_OBJECT_MAX_TENSION = 10.0f;
+const F32 FLEXIBLE_OBJECT_MAX_TENSION = 10.0f;
// "Drag" => [0,10], increments of 0.1
const F32 FLEXIBLE_OBJECT_MIN_AIR_FRICTION = 0.0f;
@@ -112,7 +112,7 @@ const F32 FLEXIBLE_OBJECT_DEFAULT_WIND_SENSITIVITY = 0.0f;
const F32 FLEXIBLE_OBJECT_MAX_WIND_SENSITIVITY = 10.0f;
// I'll explain later...
-const F32 FLEXIBLE_OBJECT_MAX_INTERNAL_TENSION_FORCE = 0.99f;
+const F32 FLEXIBLE_OBJECT_MAX_INTERNAL_TENSION_FORCE = 0.99f;
const F32 FLEXIBLE_OBJECT_DEFAULT_LENGTH = 1.0f;
const BOOL FLEXIBLE_OBJECT_DEFAULT_USING_COLLISION_SPHERE = FALSE;
@@ -121,9 +121,9 @@ const BOOL FLEXIBLE_OBJECT_DEFAULT_RENDERING_COLLISION_SPHERE = FALSE;
const LLUUID SCULPT_DEFAULT_TEXTURE("be293869-d0d9-0a69-5989-ad27f1946fd4"); // old inverted texture: "7595d345-a24c-e7ef-f0bd-78793792133e";
// Texture rotations are sent over the wire as a S16. This is used to scale the actual float
-// value to a S16. Don't use 7FFF as it introduces some odd rounding with 180 since it
+// value to a S16. Don't use 7FFF as it introduces some odd rounding with 180 since it
// can't be divided by 2. See DEV-19108
-const F32 TEXTURE_ROTATION_PACK_FACTOR = ((F32) 0x08000);
+const F32 TEXTURE_ROTATION_PACK_FACTOR = ((F32) 0x08000);
struct material_id_type // originally from llrendermaterialtable
{
@@ -154,7 +154,7 @@ struct material_id_type // originally from llrendermaterialtable
const U8 material_id_type::s_null_id[] = { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 };
-//static
+//static
// LEGACY: by default we use the LLVolumeMgr::gVolumeMgr global
// TODO -- eliminate this global from the codebase!
LLVolumeMgr* LLPrimitive::sVolumeManager = NULL;
@@ -162,61 +162,61 @@ LLVolumeMgr* LLPrimitive::sVolumeManager = NULL;
// static
void LLPrimitive::setVolumeManager( LLVolumeMgr* volume_manager )
{
- if ( !volume_manager || sVolumeManager )
- {
- LL_ERRS() << "LLPrimitive::sVolumeManager attempting to be set to NULL or it already has been set." << LL_ENDL;
- }
- sVolumeManager = volume_manager;
+ if ( !volume_manager || sVolumeManager )
+ {
+ LL_ERRS() << "LLPrimitive::sVolumeManager attempting to be set to NULL or it already has been set." << LL_ENDL;
+ }
+ sVolumeManager = volume_manager;
}
// static
bool LLPrimitive::cleanupVolumeManager()
{
- BOOL res = FALSE;
- if (sVolumeManager)
- {
- res = sVolumeManager->cleanup();
- delete sVolumeManager;
- sVolumeManager = NULL;
- }
- return res;
+ BOOL res = FALSE;
+ if (sVolumeManager)
+ {
+ res = sVolumeManager->cleanup();
+ delete sVolumeManager;
+ sVolumeManager = NULL;
+ }
+ return res;
}
//===============================================================
LLPrimitive::LLPrimitive()
-: mTextureList(),
- mNumTEs(0),
- mMiscFlags(0),
- mNumBumpmapTEs(0)
+: mTextureList(),
+ mNumTEs(0),
+ mMiscFlags(0),
+ mNumBumpmapTEs(0)
{
- mPrimitiveCode = 0;
+ mPrimitiveCode = 0;
- mMaterial = LL_MCODE_STONE;
- mVolumep = NULL;
+ mMaterial = LL_MCODE_STONE;
+ mVolumep = NULL;
- mChanged = UNCHANGED;
+ mChanged = UNCHANGED;
- mPosition.setVec(0.f,0.f,0.f);
- mVelocity.setVec(0.f,0.f,0.f);
- mAcceleration.setVec(0.f,0.f,0.f);
+ mPosition.setVec(0.f,0.f,0.f);
+ mVelocity.setVec(0.f,0.f,0.f);
+ mAcceleration.setVec(0.f,0.f,0.f);
- mRotation.loadIdentity();
- mAngularVelocity.setVec(0.f,0.f,0.f);
-
- mScale.setVec(1.f,1.f,1.f);
+ mRotation.loadIdentity();
+ mAngularVelocity.setVec(0.f,0.f,0.f);
+
+ mScale.setVec(1.f,1.f,1.f);
}
//===============================================================
LLPrimitive::~LLPrimitive()
{
- clearTextureList();
- // Cleanup handled by volume manager
- if (mVolumep && sVolumeManager)
- {
- sVolumeManager->unrefVolume(mVolumep);
- }
- mVolumep = NULL;
+ clearTextureList();
+ // Cleanup handled by volume manager
+ if (mVolumep && sVolumeManager)
+ {
+ sVolumeManager->unrefVolume(mVolumep);
+ }
+ mVolumep = NULL;
}
void LLPrimitive::clearTextureList()
@@ -227,83 +227,83 @@ void LLPrimitive::clearTextureList()
// static
LLPrimitive *LLPrimitive::createPrimitive(LLPCode p_code)
{
- LLPrimitive *retval = new LLPrimitive();
-
- if (retval)
- {
- retval->init_primitive(p_code);
- }
- else
- {
- LL_ERRS() << "primitive allocation failed" << LL_ENDL;
- }
+ LLPrimitive *retval = new LLPrimitive();
- return retval;
+ if (retval)
+ {
+ retval->init_primitive(p_code);
+ }
+ else
+ {
+ LL_ERRS() << "primitive allocation failed" << LL_ENDL;
+ }
+
+ return retval;
}
//===============================================================
void LLPrimitive::init_primitive(LLPCode p_code)
{
- clearTextureList();
- mPrimitiveCode = p_code;
+ clearTextureList();
+ mPrimitiveCode = p_code;
}
void LLPrimitive::setPCode(const U8 p_code)
{
- mPrimitiveCode = p_code;
+ mPrimitiveCode = p_code;
}
//===============================================================
LLTextureEntry* LLPrimitive::getTE(const U8 index) const
{
- return mTextureList.getTexture(index);
+ return mTextureList.getTexture(index);
}
//===============================================================
void LLPrimitive::setNumTEs(const U8 num_tes)
{
- mTextureList.setSize(num_tes);
+ mTextureList.setSize(num_tes);
}
//===============================================================
void LLPrimitive::setAllTETextures(const LLUUID &tex_id)
{
- mTextureList.setAllIDs(tex_id);
+ mTextureList.setAllIDs(tex_id);
}
//===============================================================
void LLPrimitive::setTE(const U8 index, const LLTextureEntry& te)
{
- if(mTextureList.copyTexture(index, te) != TEM_CHANGE_NONE && te.getBumpmap() > 0)
- {
- mNumBumpmapTEs++;
- }
+ if(mTextureList.copyTexture(index, te) != TEM_CHANGE_NONE && te.getBumpmap() > 0)
+ {
+ mNumBumpmapTEs++;
+ }
}
S32 LLPrimitive::setTETexture(const U8 index, const LLUUID &id)
{
- return mTextureList.setID(index, id);
+ return mTextureList.setID(index, id);
}
S32 LLPrimitive::setTEColor(const U8 index, const LLColor4 &color)
{
- return mTextureList.setColor(index, color);
+ return mTextureList.setColor(index, color);
}
S32 LLPrimitive::setTEColor(const U8 index, const LLColor3 &color)
{
- return mTextureList.setColor(index, color);
+ return mTextureList.setColor(index, color);
}
S32 LLPrimitive::setTEAlpha(const U8 index, const F32 alpha)
{
- return mTextureList.setAlpha(index, alpha);
+ return mTextureList.setAlpha(index, alpha);
}
//===============================================================
S32 LLPrimitive::setTEScale(const U8 index, const F32 s, const F32 t)
{
- return mTextureList.setScale(index, s, t);
+ return mTextureList.setScale(index, s, t);
}
@@ -311,7 +311,7 @@ S32 LLPrimitive::setTEScale(const U8 index, const F32 s, const F32 t)
// voodoo related to texture coords
S32 LLPrimitive::setTEScaleS(const U8 index, const F32 s)
{
- return mTextureList.setScaleS(index, s);
+ return mTextureList.setScaleS(index, s);
}
@@ -319,14 +319,14 @@ S32 LLPrimitive::setTEScaleS(const U8 index, const F32 s)
// voodoo related to texture coords
S32 LLPrimitive::setTEScaleT(const U8 index, const F32 t)
{
- return mTextureList.setScaleT(index, t);
+ return mTextureList.setScaleT(index, t);
}
//===============================================================
S32 LLPrimitive::setTEOffset(const U8 index, const F32 s, const F32 t)
{
- return mTextureList.setOffset(index, s, t);
+ return mTextureList.setOffset(index, s, t);
}
@@ -334,7 +334,7 @@ S32 LLPrimitive::setTEOffset(const U8 index, const F32 s, const F32 t)
// voodoo related to texture coords
S32 LLPrimitive::setTEOffsetS(const U8 index, const F32 s)
{
- return mTextureList.setOffsetS(index, s);
+ return mTextureList.setOffsetS(index, s);
}
@@ -342,289 +342,289 @@ S32 LLPrimitive::setTEOffsetS(const U8 index, const F32 s)
// voodoo related to texture coords
S32 LLPrimitive::setTEOffsetT(const U8 index, const F32 t)
{
- return mTextureList.setOffsetT(index, t);
+ return mTextureList.setOffsetT(index, t);
}
//===============================================================
S32 LLPrimitive::setTERotation(const U8 index, const F32 r)
{
- return mTextureList.setRotation(index, r);
+ return mTextureList.setRotation(index, r);
}
S32 LLPrimitive::setTEMaterialID(const U8 index, const LLMaterialID& pMaterialID)
{
- return mTextureList.setMaterialID(index, pMaterialID);
+ return mTextureList.setMaterialID(index, pMaterialID);
}
S32 LLPrimitive::setTEMaterialParams(const U8 index, const LLMaterialPtr pMaterialParams)
{
- return mTextureList.setMaterialParams(index, pMaterialParams);
+ return mTextureList.setMaterialParams(index, pMaterialParams);
}
LLMaterialPtr LLPrimitive::getTEMaterialParams(const U8 index)
{
- return mTextureList.getMaterialParams(index);
+ return mTextureList.getMaterialParams(index);
}
//===============================================================
S32 LLPrimitive::setTEBumpShinyFullbright(const U8 index, const U8 bump)
{
- updateNumBumpmap(index, bump);
- return mTextureList.setBumpShinyFullbright(index, bump);
+ updateNumBumpmap(index, bump);
+ return mTextureList.setBumpShinyFullbright(index, bump);
}
S32 LLPrimitive::setTEMediaTexGen(const U8 index, const U8 media)
{
- return mTextureList.setMediaTexGen(index, media);
+ return mTextureList.setMediaTexGen(index, media);
}
S32 LLPrimitive::setTEBumpmap(const U8 index, const U8 bump)
{
- updateNumBumpmap(index, bump);
- return mTextureList.setBumpMap(index, bump);
+ updateNumBumpmap(index, bump);
+ return mTextureList.setBumpMap(index, bump);
}
S32 LLPrimitive::setTEBumpShiny(const U8 index, const U8 bump_shiny)
{
- updateNumBumpmap(index, bump_shiny);
- return mTextureList.setBumpShiny(index, bump_shiny);
+ updateNumBumpmap(index, bump_shiny);
+ return mTextureList.setBumpShiny(index, bump_shiny);
}
S32 LLPrimitive::setTETexGen(const U8 index, const U8 texgen)
{
- return mTextureList.setTexGen(index, texgen);
+ return mTextureList.setTexGen(index, texgen);
}
S32 LLPrimitive::setTEShiny(const U8 index, const U8 shiny)
{
- return mTextureList.setShiny(index, shiny);
+ return mTextureList.setShiny(index, shiny);
}
S32 LLPrimitive::setTEFullbright(const U8 index, const U8 fullbright)
{
- return mTextureList.setFullbright(index, fullbright);
+ return mTextureList.setFullbright(index, fullbright);
}
S32 LLPrimitive::setTEMediaFlags(const U8 index, const U8 media_flags)
{
- return mTextureList.setMediaFlags(index, media_flags);
+ return mTextureList.setMediaFlags(index, media_flags);
}
S32 LLPrimitive::setTEGlow(const U8 index, const F32 glow)
{
- return mTextureList.setGlow(index, glow);
+ return mTextureList.setGlow(index, glow);
}
void LLPrimitive::setAllTESelected(bool sel)
{
- for (int i = 0, cnt = getNumTEs(); i < cnt; i++)
- {
- setTESelected(i, sel);
- }
+ for (int i = 0, cnt = getNumTEs(); i < cnt; i++)
+ {
+ setTESelected(i, sel);
+ }
}
-
+
void LLPrimitive::setTESelected(const U8 te, bool sel)
{
- LLTextureEntry* tep = getTE(te);
- if ( (tep) && (tep->setSelected(sel)) && (!sel) && (tep->hasPendingMaterialUpdate()) )
- {
- LLMaterialID material_id = tep->getMaterialID();
- setTEMaterialID(te, material_id);
- }
+ LLTextureEntry* tep = getTE(te);
+ if ( (tep) && (tep->setSelected(sel)) && (!sel) && (tep->hasPendingMaterialUpdate()) )
+ {
+ LLMaterialID material_id = tep->getMaterialID();
+ setTEMaterialID(te, material_id);
+ }
}
LLPCode LLPrimitive::legacyToPCode(const U8 legacy)
{
- // TODO: Should this default to something valid?
- // Maybe volume?
- LLPCode pcode = 0;
-
- switch (legacy)
- {
- /*
- case BOX:
- pcode = LL_PCODE_CUBE;
- break;
- case CYLINDER:
- pcode = LL_PCODE_CYLINDER;
- break;
- case CONE:
- pcode = LL_PCODE_CONE;
- break;
- case HALF_CONE:
- pcode = LL_PCODE_CONE_HEMI;
- break;
- case HALF_CYLINDER:
- pcode = LL_PCODE_CYLINDER_HEMI;
- break;
- case HALF_SPHERE:
- pcode = LL_PCODE_SPHERE_HEMI;
- break;
- case PRISM:
- pcode = LL_PCODE_PRISM;
- break;
- case PYRAMID:
- pcode = LL_PCODE_PYRAMID;
- break;
- case SPHERE:
- pcode = LL_PCODE_SPHERE;
- break;
- case TETRAHEDRON:
- pcode = LL_PCODE_TETRAHEDRON;
- break;
- case DEMON:
- pcode = LL_PCODE_LEGACY_DEMON;
- break;
- case LSL_TEST:
- pcode = LL_PCODE_LEGACY_LSL_TEST;
- break;
- case ORACLE:
- pcode = LL_PCODE_LEGACY_ORACLE;
- break;
- case TEXTBUBBLE:
- pcode = LL_PCODE_LEGACY_TEXT_BUBBLE;
- break;
- case ATOR:
- pcode = LL_PCODE_LEGACY_ATOR;
- break;
- case BASIC_SHOT:
- pcode = LL_PCODE_LEGACY_SHOT;
- break;
- case BIG_SHOT:
- pcode = LL_PCODE_LEGACY_SHOT_BIG;
- break;
- case BIRD:
- pcode = LL_PCODE_LEGACY_BIRD;
- break;
- case ROCK:
- pcode = LL_PCODE_LEGACY_ROCK;
- break;
- case SMOKE:
- pcode = LL_PCODE_LEGACY_SMOKE;
- break;
- case SPARK:
- pcode = LL_PCODE_LEGACY_SPARK;
- break;
- */
- case PRIMITIVE_VOLUME:
- pcode = LL_PCODE_VOLUME;
- break;
- case GRASS:
- pcode = LL_PCODE_LEGACY_GRASS;
- break;
- case PART_SYS:
- pcode = LL_PCODE_LEGACY_PART_SYS;
- break;
- case PLAYER:
- pcode = LL_PCODE_LEGACY_AVATAR;
- break;
- case TREE:
- pcode = LL_PCODE_LEGACY_TREE;
- break;
- case TREE_NEW:
- pcode = LL_PCODE_TREE_NEW;
- break;
- default:
- LL_WARNS() << "Unknown legacy code " << legacy << " [" << (S32)legacy << "]!" << LL_ENDL;
- }
-
- return pcode;
+ // TODO: Should this default to something valid?
+ // Maybe volume?
+ LLPCode pcode = 0;
+
+ switch (legacy)
+ {
+ /*
+ case BOX:
+ pcode = LL_PCODE_CUBE;
+ break;
+ case CYLINDER:
+ pcode = LL_PCODE_CYLINDER;
+ break;
+ case CONE:
+ pcode = LL_PCODE_CONE;
+ break;
+ case HALF_CONE:
+ pcode = LL_PCODE_CONE_HEMI;
+ break;
+ case HALF_CYLINDER:
+ pcode = LL_PCODE_CYLINDER_HEMI;
+ break;
+ case HALF_SPHERE:
+ pcode = LL_PCODE_SPHERE_HEMI;
+ break;
+ case PRISM:
+ pcode = LL_PCODE_PRISM;
+ break;
+ case PYRAMID:
+ pcode = LL_PCODE_PYRAMID;
+ break;
+ case SPHERE:
+ pcode = LL_PCODE_SPHERE;
+ break;
+ case TETRAHEDRON:
+ pcode = LL_PCODE_TETRAHEDRON;
+ break;
+ case DEMON:
+ pcode = LL_PCODE_LEGACY_DEMON;
+ break;
+ case LSL_TEST:
+ pcode = LL_PCODE_LEGACY_LSL_TEST;
+ break;
+ case ORACLE:
+ pcode = LL_PCODE_LEGACY_ORACLE;
+ break;
+ case TEXTBUBBLE:
+ pcode = LL_PCODE_LEGACY_TEXT_BUBBLE;
+ break;
+ case ATOR:
+ pcode = LL_PCODE_LEGACY_ATOR;
+ break;
+ case BASIC_SHOT:
+ pcode = LL_PCODE_LEGACY_SHOT;
+ break;
+ case BIG_SHOT:
+ pcode = LL_PCODE_LEGACY_SHOT_BIG;
+ break;
+ case BIRD:
+ pcode = LL_PCODE_LEGACY_BIRD;
+ break;
+ case ROCK:
+ pcode = LL_PCODE_LEGACY_ROCK;
+ break;
+ case SMOKE:
+ pcode = LL_PCODE_LEGACY_SMOKE;
+ break;
+ case SPARK:
+ pcode = LL_PCODE_LEGACY_SPARK;
+ break;
+ */
+ case PRIMITIVE_VOLUME:
+ pcode = LL_PCODE_VOLUME;
+ break;
+ case GRASS:
+ pcode = LL_PCODE_LEGACY_GRASS;
+ break;
+ case PART_SYS:
+ pcode = LL_PCODE_LEGACY_PART_SYS;
+ break;
+ case PLAYER:
+ pcode = LL_PCODE_LEGACY_AVATAR;
+ break;
+ case TREE:
+ pcode = LL_PCODE_LEGACY_TREE;
+ break;
+ case TREE_NEW:
+ pcode = LL_PCODE_TREE_NEW;
+ break;
+ default:
+ LL_WARNS() << "Unknown legacy code " << legacy << " [" << (S32)legacy << "]!" << LL_ENDL;
+ }
+
+ return pcode;
}
U8 LLPrimitive::pCodeToLegacy(const LLPCode pcode)
{
- U8 legacy;
- switch (pcode)
- {
+ U8 legacy;
+ switch (pcode)
+ {
/*
- case LL_PCODE_CUBE:
- legacy = BOX;
- break;
- case LL_PCODE_CYLINDER:
- legacy = CYLINDER;
- break;
- case LL_PCODE_CONE:
- legacy = CONE;
- break;
- case LL_PCODE_CONE_HEMI:
- legacy = HALF_CONE;
- break;
- case LL_PCODE_CYLINDER_HEMI:
- legacy = HALF_CYLINDER;
- break;
- case LL_PCODE_SPHERE_HEMI:
- legacy = HALF_SPHERE;
- break;
- case LL_PCODE_PRISM:
- legacy = PRISM;
- break;
- case LL_PCODE_PYRAMID:
- legacy = PYRAMID;
- break;
- case LL_PCODE_SPHERE:
- legacy = SPHERE;
- break;
- case LL_PCODE_TETRAHEDRON:
- legacy = TETRAHEDRON;
- break;
- case LL_PCODE_LEGACY_ATOR:
- legacy = ATOR;
- break;
- case LL_PCODE_LEGACY_SHOT:
- legacy = BASIC_SHOT;
- break;
- case LL_PCODE_LEGACY_SHOT_BIG:
- legacy = BIG_SHOT;
- break;
- case LL_PCODE_LEGACY_BIRD:
- legacy = BIRD;
- break;
- case LL_PCODE_LEGACY_DEMON:
- legacy = DEMON;
- break;
- case LL_PCODE_LEGACY_LSL_TEST:
- legacy = LSL_TEST;
- break;
- case LL_PCODE_LEGACY_ORACLE:
- legacy = ORACLE;
- break;
- case LL_PCODE_LEGACY_ROCK:
- legacy = ROCK;
- break;
- case LL_PCODE_LEGACY_TEXT_BUBBLE:
- legacy = TEXTBUBBLE;
- break;
- case LL_PCODE_LEGACY_SMOKE:
- legacy = SMOKE;
- break;
- case LL_PCODE_LEGACY_SPARK:
- legacy = SPARK;
- break;
+ case LL_PCODE_CUBE:
+ legacy = BOX;
+ break;
+ case LL_PCODE_CYLINDER:
+ legacy = CYLINDER;
+ break;
+ case LL_PCODE_CONE:
+ legacy = CONE;
+ break;
+ case LL_PCODE_CONE_HEMI:
+ legacy = HALF_CONE;
+ break;
+ case LL_PCODE_CYLINDER_HEMI:
+ legacy = HALF_CYLINDER;
+ break;
+ case LL_PCODE_SPHERE_HEMI:
+ legacy = HALF_SPHERE;
+ break;
+ case LL_PCODE_PRISM:
+ legacy = PRISM;
+ break;
+ case LL_PCODE_PYRAMID:
+ legacy = PYRAMID;
+ break;
+ case LL_PCODE_SPHERE:
+ legacy = SPHERE;
+ break;
+ case LL_PCODE_TETRAHEDRON:
+ legacy = TETRAHEDRON;
+ break;
+ case LL_PCODE_LEGACY_ATOR:
+ legacy = ATOR;
+ break;
+ case LL_PCODE_LEGACY_SHOT:
+ legacy = BASIC_SHOT;
+ break;
+ case LL_PCODE_LEGACY_SHOT_BIG:
+ legacy = BIG_SHOT;
+ break;
+ case LL_PCODE_LEGACY_BIRD:
+ legacy = BIRD;
+ break;
+ case LL_PCODE_LEGACY_DEMON:
+ legacy = DEMON;
+ break;
+ case LL_PCODE_LEGACY_LSL_TEST:
+ legacy = LSL_TEST;
+ break;
+ case LL_PCODE_LEGACY_ORACLE:
+ legacy = ORACLE;
+ break;
+ case LL_PCODE_LEGACY_ROCK:
+ legacy = ROCK;
+ break;
+ case LL_PCODE_LEGACY_TEXT_BUBBLE:
+ legacy = TEXTBUBBLE;
+ break;
+ case LL_PCODE_LEGACY_SMOKE:
+ legacy = SMOKE;
+ break;
+ case LL_PCODE_LEGACY_SPARK:
+ legacy = SPARK;
+ break;
*/
- case LL_PCODE_VOLUME:
- legacy = PRIMITIVE_VOLUME;
- break;
- case LL_PCODE_LEGACY_GRASS:
- legacy = GRASS;
- break;
- case LL_PCODE_LEGACY_PART_SYS:
- legacy = PART_SYS;
- break;
- case LL_PCODE_LEGACY_AVATAR:
- legacy = PLAYER;
- break;
- case LL_PCODE_LEGACY_TREE:
- legacy = TREE;
- break;
- case LL_PCODE_TREE_NEW:
- legacy = TREE_NEW;
- break;
- default:
- LL_WARNS() << "Unknown pcode " << (S32)pcode << ":" << pcode << "!" << LL_ENDL;
- return 0;
- }
- return legacy;
+ case LL_PCODE_VOLUME:
+ legacy = PRIMITIVE_VOLUME;
+ break;
+ case LL_PCODE_LEGACY_GRASS:
+ legacy = GRASS;
+ break;
+ case LL_PCODE_LEGACY_PART_SYS:
+ legacy = PART_SYS;
+ break;
+ case LL_PCODE_LEGACY_AVATAR:
+ legacy = PLAYER;
+ break;
+ case LL_PCODE_LEGACY_TREE:
+ legacy = TREE;
+ break;
+ case LL_PCODE_TREE_NEW:
+ legacy = TREE_NEW;
+ break;
+ default:
+ LL_WARNS() << "Unknown pcode " << (S32)pcode << ":" << pcode << "!" << LL_ENDL;
+ return 0;
+ }
+ return legacy;
}
@@ -632,491 +632,491 @@ U8 LLPrimitive::pCodeToLegacy(const LLPCode pcode)
// Don't crash or LL_ERRS() here! This function is used for debug strings.
std::string LLPrimitive::pCodeToString(const LLPCode pcode)
{
- std::string pcode_string;
-
- U8 base_code = pcode & LL_PCODE_BASE_MASK;
- if (!pcode)
- {
- pcode_string = "null";
- }
- else if ((base_code) == LL_PCODE_LEGACY)
- {
- // It's a legacy object
- switch (pcode)
- {
- case LL_PCODE_LEGACY_GRASS:
- pcode_string = "grass";
- break;
- case LL_PCODE_LEGACY_PART_SYS:
- pcode_string = "particle system";
- break;
- case LL_PCODE_LEGACY_AVATAR:
- pcode_string = "avatar";
- break;
- case LL_PCODE_LEGACY_TEXT_BUBBLE:
- pcode_string = "text bubble";
- break;
- case LL_PCODE_LEGACY_TREE:
- pcode_string = "tree";
- break;
- case LL_PCODE_TREE_NEW:
- pcode_string = "tree_new";
- break;
- default:
- pcode_string = llformat( "unknown legacy pcode %i",(U32)pcode);
- }
- }
- else
- {
- std::string shape;
- std::string mask;
- if (base_code == LL_PCODE_CUBE)
- {
- shape = "cube";
- }
- else if (base_code == LL_PCODE_CYLINDER)
- {
- shape = "cylinder";
- }
- else if (base_code == LL_PCODE_CONE)
- {
- shape = "cone";
- }
- else if (base_code == LL_PCODE_PRISM)
- {
- shape = "prism";
- }
- else if (base_code == LL_PCODE_PYRAMID)
- {
- shape = "pyramid";
- }
- else if (base_code == LL_PCODE_SPHERE)
- {
- shape = "sphere";
- }
- else if (base_code == LL_PCODE_TETRAHEDRON)
- {
- shape = "tetrahedron";
- }
- else if (base_code == LL_PCODE_VOLUME)
- {
- shape = "volume";
- }
- else if (base_code == LL_PCODE_APP)
- {
- shape = "app";
- }
- else
- {
- LL_WARNS() << "Unknown base mask for pcode: " << base_code << LL_ENDL;
- }
-
- U8 mask_code = pcode & (~LL_PCODE_BASE_MASK);
- if (base_code == LL_PCODE_APP)
- {
- mask = llformat( "%x", mask_code);
- }
- else if (mask_code & LL_PCODE_HEMI_MASK)
- {
- mask = "hemi";
- }
- else
- {
- mask = llformat( "%x", mask_code);
- }
-
- if (mask[0])
- {
- pcode_string = llformat( "%s-%s", shape.c_str(), mask.c_str());
- }
- else
- {
- pcode_string = llformat( "%s", shape.c_str());
- }
- }
-
- return pcode_string;
+ std::string pcode_string;
+
+ U8 base_code = pcode & LL_PCODE_BASE_MASK;
+ if (!pcode)
+ {
+ pcode_string = "null";
+ }
+ else if ((base_code) == LL_PCODE_LEGACY)
+ {
+ // It's a legacy object
+ switch (pcode)
+ {
+ case LL_PCODE_LEGACY_GRASS:
+ pcode_string = "grass";
+ break;
+ case LL_PCODE_LEGACY_PART_SYS:
+ pcode_string = "particle system";
+ break;
+ case LL_PCODE_LEGACY_AVATAR:
+ pcode_string = "avatar";
+ break;
+ case LL_PCODE_LEGACY_TEXT_BUBBLE:
+ pcode_string = "text bubble";
+ break;
+ case LL_PCODE_LEGACY_TREE:
+ pcode_string = "tree";
+ break;
+ case LL_PCODE_TREE_NEW:
+ pcode_string = "tree_new";
+ break;
+ default:
+ pcode_string = llformat( "unknown legacy pcode %i",(U32)pcode);
+ }
+ }
+ else
+ {
+ std::string shape;
+ std::string mask;
+ if (base_code == LL_PCODE_CUBE)
+ {
+ shape = "cube";
+ }
+ else if (base_code == LL_PCODE_CYLINDER)
+ {
+ shape = "cylinder";
+ }
+ else if (base_code == LL_PCODE_CONE)
+ {
+ shape = "cone";
+ }
+ else if (base_code == LL_PCODE_PRISM)
+ {
+ shape = "prism";
+ }
+ else if (base_code == LL_PCODE_PYRAMID)
+ {
+ shape = "pyramid";
+ }
+ else if (base_code == LL_PCODE_SPHERE)
+ {
+ shape = "sphere";
+ }
+ else if (base_code == LL_PCODE_TETRAHEDRON)
+ {
+ shape = "tetrahedron";
+ }
+ else if (base_code == LL_PCODE_VOLUME)
+ {
+ shape = "volume";
+ }
+ else if (base_code == LL_PCODE_APP)
+ {
+ shape = "app";
+ }
+ else
+ {
+ LL_WARNS() << "Unknown base mask for pcode: " << base_code << LL_ENDL;
+ }
+
+ U8 mask_code = pcode & (~LL_PCODE_BASE_MASK);
+ if (base_code == LL_PCODE_APP)
+ {
+ mask = llformat( "%x", mask_code);
+ }
+ else if (mask_code & LL_PCODE_HEMI_MASK)
+ {
+ mask = "hemi";
+ }
+ else
+ {
+ mask = llformat( "%x", mask_code);
+ }
+
+ if (mask[0])
+ {
+ pcode_string = llformat( "%s-%s", shape.c_str(), mask.c_str());
+ }
+ else
+ {
+ pcode_string = llformat( "%s", shape.c_str());
+ }
+ }
+
+ return pcode_string;
}
void LLPrimitive::copyTEs(const LLPrimitive *primitivep)
{
- U32 i;
- if (primitivep->getExpectedNumTEs() != getExpectedNumTEs())
- {
- LL_WARNS() << "Primitives don't have same expected number of TE's" << LL_ENDL;
- }
- U32 num_tes = llmin(primitivep->getExpectedNumTEs(), getExpectedNumTEs());
- if (mTextureList.size() < getExpectedNumTEs())
- {
- mTextureList.setSize(getExpectedNumTEs());
- }
- for (i = 0; i < num_tes; i++)
- {
- mTextureList.copyTexture(i, *(primitivep->getTE(i)));
- }
-}
-
-S32 face_index_from_id(LLFaceID face_ID, const std::vector<LLProfile::Face>& faceArray)
-{
- S32 i;
- for (i = 0; i < (S32)faceArray.size(); i++)
- {
- if (faceArray[i].mFaceID == face_ID)
- {
- return i;
- }
- }
- return -1;
+ U32 i;
+ if (primitivep->getExpectedNumTEs() != getExpectedNumTEs())
+ {
+ LL_WARNS() << "Primitives don't have same expected number of TE's" << LL_ENDL;
+ }
+ U32 num_tes = llmin(primitivep->getExpectedNumTEs(), getExpectedNumTEs());
+ if (mTextureList.size() < getExpectedNumTEs())
+ {
+ mTextureList.setSize(getExpectedNumTEs());
+ }
+ for (i = 0; i < num_tes; i++)
+ {
+ mTextureList.copyTexture(i, *(primitivep->getTE(i)));
+ }
+}
+
+S32 face_index_from_id(LLFaceID face_ID, const std::vector<LLProfile::Face>& faceArray)
+{
+ S32 i;
+ for (i = 0; i < (S32)faceArray.size(); i++)
+ {
+ if (faceArray[i].mFaceID == face_ID)
+ {
+ return i;
+ }
+ }
+ return -1;
}
BOOL LLPrimitive::setVolume(const LLVolumeParams &volume_params, const S32 detail, bool unique_volume)
{
- if (NO_LOD == detail)
- {
- // build the new object
- setChanged(GEOMETRY);
- sVolumeManager->unrefVolume(mVolumep);
- mVolumep = new LLVolume(volume_params, 1, TRUE, TRUE);
- setNumTEs(mVolumep->getNumFaces());
- return FALSE;
- }
-
- LLVolume *volumep;
- if (unique_volume)
- {
- F32 volume_detail = LLVolumeLODGroup::getVolumeScaleFromDetail(detail);
- if (mVolumep.notNull() && volume_params == mVolumep->getParams() && (volume_detail == mVolumep->getDetail()))
- {
- return FALSE;
- }
- volumep = new LLVolume(volume_params, volume_detail, FALSE, TRUE);
- }
- else
- {
- if (mVolumep.notNull())
- {
- F32 volume_detail = LLVolumeLODGroup::getVolumeScaleFromDetail(detail);
- if (volume_params == mVolumep->getParams() && (volume_detail == mVolumep->getDetail()))
- {
- return FALSE;
- }
- }
-
- volumep = sVolumeManager->refVolume(volume_params, detail);
- if (volumep == mVolumep)
- {
- sVolumeManager->unrefVolume( volumep ); // LLVolumeMgr::refVolume() creates a reference, but we don't need a second one.
- return TRUE;
- }
- }
-
- setChanged(GEOMETRY);
-
-
- if (!mVolumep)
- {
- mVolumep = volumep;
- //mFaceMask = mVolumep->generateFaceMask();
- setNumTEs(mVolumep->getNumFaces());
- return TRUE;
- }
-
-#if 0
- // #if 0'd out by davep
- // this is a lot of cruft to set texture entry values that just stay the same for LOD switch
- // or immediately get overridden by an object update message, also crashes occasionally
- U32 old_face_mask = mVolumep->mFaceMask;
-
- S32 face_bit = 0;
- S32 cur_mask = 0;
-
- // Grab copies of the old faces from the original shape, ordered by type.
- // We will use these to figure out what old texture info gets mapped to new
- // faces in the new shape.
- std::vector<LLProfile::Face> old_faces;
- for (S32 face = 0; face < mVolumep->getNumFaces(); face++)
- {
- old_faces.push_back(mVolumep->getProfile().mFaces[face]);
- }
-
- // Copy the old texture info off to the side, but not in the order in which
- // they live in the mTextureList, rather in order of ther "face id" which
- // is the corresponding value of LLVolueParams::LLProfile::mFaces::mIndex.
- //
- // Hence, some elements of old_tes::mEntryList will be invalid. It is
- // initialized to a size of 9 (max number of possible faces on a volume?)
- // and only the ones with valid types are filled in.
- LLPrimTextureList old_tes;
- old_tes.setSize(9);
- for (face_bit = 0; face_bit < 9; face_bit++)
- {
- cur_mask = 0x1 << face_bit;
- if (old_face_mask & cur_mask)
- {
- S32 te_index = face_index_from_id(cur_mask, old_faces);
- old_tes.copyTexture(face_bit, *(getTE(te_index)));
- //LL_INFOS() << face_bit << ":" << te_index << ":" << old_tes[face_bit].getID() << LL_ENDL;
- }
- }
-
-
- // build the new object
- sVolumeManager->unrefVolume(mVolumep);
- mVolumep = volumep;
-
- U32 new_face_mask = mVolumep->mFaceMask;
- S32 i;
-
- if (old_face_mask == new_face_mask)
- {
- // nothing to do
- return TRUE;
- }
-
- if (mVolumep->getNumFaces() == 0 && new_face_mask != 0)
- {
- LL_WARNS() << "Object with 0 faces found...INCORRECT!" << LL_ENDL;
- setNumTEs(mVolumep->getNumFaces());
- return TRUE;
- }
-
- // initialize face_mapping
- S32 face_mapping[9];
- for (face_bit = 0; face_bit < 9; face_bit++)
- {
- face_mapping[face_bit] = face_bit;
- }
-
- // The new shape may have more faces than the original, but we can't just
- // add them to the end -- the ordering matters and it may be that we must
- // insert the new faces in the middle of the list. When we add a face it
- // will pick up the texture/color info of one of the old faces an so we
- // now figure out which old face info gets mapped to each new face, and
- // store in the face_mapping lookup table.
- for (face_bit = 0; face_bit < 9; face_bit++)
- {
- cur_mask = 0x1 << face_bit;
- if (!(new_face_mask & cur_mask))
- {
- // Face doesn't exist in new map.
- face_mapping[face_bit] = -1;
- continue;
- }
- else if (old_face_mask & cur_mask)
- {
- // Face exists in new and old map.
- face_mapping[face_bit] = face_bit;
- continue;
- }
-
- // OK, how we've got a mismatch, where we have to fill a new face with one from
- // the old face.
- if (cur_mask & (LL_FACE_PATH_BEGIN | LL_FACE_PATH_END | LL_FACE_INNER_SIDE))
- {
- // It's a top/bottom/hollow interior face.
- if (old_face_mask & LL_FACE_PATH_END)
- {
- face_mapping[face_bit] = 1;
- continue;
- }
- else
- {
- S32 cur_outer_mask = LL_FACE_OUTER_SIDE_0;
- for (i = 0; i < 4; i++)
- {
- if (old_face_mask & cur_outer_mask)
- {
- face_mapping[face_bit] = 5 + i;
- break;
- }
- cur_outer_mask <<= 1;
- }
- if (i == 4)
- {
- LL_WARNS() << "No path end or outer face in volume!" << LL_ENDL;
- }
- continue;
- }
- }
-
- if (cur_mask & (LL_FACE_PROFILE_BEGIN | LL_FACE_PROFILE_END))
- {
- // A cut slice. Use the hollow interior if we have it.
- if (old_face_mask & LL_FACE_INNER_SIDE)
- {
- face_mapping[face_bit] = 2;
- continue;
- }
-
- // No interior, use the bottom face.
- // Could figure out which of the outer faces was nearest, but that would be harder.
- if (old_face_mask & LL_FACE_PATH_END)
- {
- face_mapping[face_bit] = 1;
- continue;
- }
- else
- {
- S32 cur_outer_mask = LL_FACE_OUTER_SIDE_0;
- for (i = 0; i < 4; i++)
- {
- if (old_face_mask & cur_outer_mask)
- {
- face_mapping[face_bit] = 5 + i;
- break;
- }
- cur_outer_mask <<= 1;
- }
- if (i == 4)
- {
- LL_WARNS() << "No path end or outer face in volume!" << LL_ENDL;
- }
- continue;
- }
- }
-
- // OK, the face that's missing is an outer face...
- // Pull from the nearest adjacent outer face (there's always guaranteed to be one...
- S32 cur_outer = face_bit - 5;
- S32 min_dist = 5;
- S32 min_outer_bit = -1;
- S32 i;
- for (i = 0; i < 4; i++)
- {
- if (old_face_mask & (LL_FACE_OUTER_SIDE_0 << i))
- {
- S32 dist = abs(i - cur_outer);
- if (dist < min_dist)
- {
- min_dist = dist;
- min_outer_bit = i + 5;
- }
- }
- }
- if (-1 == min_outer_bit)
- {
- LL_INFOS() << (LLVolume *)mVolumep << LL_ENDL;
- LL_WARNS() << "Bad! No outer faces, impossible!" << LL_ENDL;
- }
- face_mapping[face_bit] = min_outer_bit;
- }
-
-
- setNumTEs(mVolumep->getNumFaces());
- for (face_bit = 0; face_bit < 9; face_bit++)
- {
- // For each possible face type on the new shape we check to see if that
- // face exists and if it does we create a texture entry that is a copy
- // of one of the originals. Since the originals might not have a
- // matching face, we use the face_mapping lookup table to figure out
- // which face information to copy.
- cur_mask = 0x1 << face_bit;
- if (new_face_mask & cur_mask)
- {
- if (-1 == face_mapping[face_bit])
- {
- LL_WARNS() << "No mapping from old face to new face!" << LL_ENDL;
- }
-
- S32 te_num = face_index_from_id(cur_mask, mVolumep->getProfile().mFaces);
- setTE(te_num, *(old_tes.getTexture(face_mapping[face_bit])));
- }
- }
+ if (NO_LOD == detail)
+ {
+ // build the new object
+ setChanged(GEOMETRY);
+ sVolumeManager->unrefVolume(mVolumep);
+ mVolumep = new LLVolume(volume_params, 1, TRUE, TRUE);
+ setNumTEs(mVolumep->getNumFaces());
+ return FALSE;
+ }
+
+ LLVolume *volumep;
+ if (unique_volume)
+ {
+ F32 volume_detail = LLVolumeLODGroup::getVolumeScaleFromDetail(detail);
+ if (mVolumep.notNull() && volume_params == mVolumep->getParams() && (volume_detail == mVolumep->getDetail()))
+ {
+ return FALSE;
+ }
+ volumep = new LLVolume(volume_params, volume_detail, FALSE, TRUE);
+ }
+ else
+ {
+ if (mVolumep.notNull())
+ {
+ F32 volume_detail = LLVolumeLODGroup::getVolumeScaleFromDetail(detail);
+ if (volume_params == mVolumep->getParams() && (volume_detail == mVolumep->getDetail()))
+ {
+ return FALSE;
+ }
+ }
+
+ volumep = sVolumeManager->refVolume(volume_params, detail);
+ if (volumep == mVolumep)
+ {
+ sVolumeManager->unrefVolume( volumep ); // LLVolumeMgr::refVolume() creates a reference, but we don't need a second one.
+ return TRUE;
+ }
+ }
+
+ setChanged(GEOMETRY);
+
+
+ if (!mVolumep)
+ {
+ mVolumep = volumep;
+ //mFaceMask = mVolumep->generateFaceMask();
+ setNumTEs(mVolumep->getNumFaces());
+ return TRUE;
+ }
+
+#if 0
+ // #if 0'd out by davep
+ // this is a lot of cruft to set texture entry values that just stay the same for LOD switch
+ // or immediately get overridden by an object update message, also crashes occasionally
+ U32 old_face_mask = mVolumep->mFaceMask;
+
+ S32 face_bit = 0;
+ S32 cur_mask = 0;
+
+ // Grab copies of the old faces from the original shape, ordered by type.
+ // We will use these to figure out what old texture info gets mapped to new
+ // faces in the new shape.
+ std::vector<LLProfile::Face> old_faces;
+ for (S32 face = 0; face < mVolumep->getNumFaces(); face++)
+ {
+ old_faces.push_back(mVolumep->getProfile().mFaces[face]);
+ }
+
+ // Copy the old texture info off to the side, but not in the order in which
+ // they live in the mTextureList, rather in order of ther "face id" which
+ // is the corresponding value of LLVolueParams::LLProfile::mFaces::mIndex.
+ //
+ // Hence, some elements of old_tes::mEntryList will be invalid. It is
+ // initialized to a size of 9 (max number of possible faces on a volume?)
+ // and only the ones with valid types are filled in.
+ LLPrimTextureList old_tes;
+ old_tes.setSize(9);
+ for (face_bit = 0; face_bit < 9; face_bit++)
+ {
+ cur_mask = 0x1 << face_bit;
+ if (old_face_mask & cur_mask)
+ {
+ S32 te_index = face_index_from_id(cur_mask, old_faces);
+ old_tes.copyTexture(face_bit, *(getTE(te_index)));
+ //LL_INFOS() << face_bit << ":" << te_index << ":" << old_tes[face_bit].getID() << LL_ENDL;
+ }
+ }
+
+
+ // build the new object
+ sVolumeManager->unrefVolume(mVolumep);
+ mVolumep = volumep;
+
+ U32 new_face_mask = mVolumep->mFaceMask;
+ S32 i;
+
+ if (old_face_mask == new_face_mask)
+ {
+ // nothing to do
+ return TRUE;
+ }
+
+ if (mVolumep->getNumFaces() == 0 && new_face_mask != 0)
+ {
+ LL_WARNS() << "Object with 0 faces found...INCORRECT!" << LL_ENDL;
+ setNumTEs(mVolumep->getNumFaces());
+ return TRUE;
+ }
+
+ // initialize face_mapping
+ S32 face_mapping[9];
+ for (face_bit = 0; face_bit < 9; face_bit++)
+ {
+ face_mapping[face_bit] = face_bit;
+ }
+
+ // The new shape may have more faces than the original, but we can't just
+ // add them to the end -- the ordering matters and it may be that we must
+ // insert the new faces in the middle of the list. When we add a face it
+ // will pick up the texture/color info of one of the old faces an so we
+ // now figure out which old face info gets mapped to each new face, and
+ // store in the face_mapping lookup table.
+ for (face_bit = 0; face_bit < 9; face_bit++)
+ {
+ cur_mask = 0x1 << face_bit;
+ if (!(new_face_mask & cur_mask))
+ {
+ // Face doesn't exist in new map.
+ face_mapping[face_bit] = -1;
+ continue;
+ }
+ else if (old_face_mask & cur_mask)
+ {
+ // Face exists in new and old map.
+ face_mapping[face_bit] = face_bit;
+ continue;
+ }
+
+ // OK, how we've got a mismatch, where we have to fill a new face with one from
+ // the old face.
+ if (cur_mask & (LL_FACE_PATH_BEGIN | LL_FACE_PATH_END | LL_FACE_INNER_SIDE))
+ {
+ // It's a top/bottom/hollow interior face.
+ if (old_face_mask & LL_FACE_PATH_END)
+ {
+ face_mapping[face_bit] = 1;
+ continue;
+ }
+ else
+ {
+ S32 cur_outer_mask = LL_FACE_OUTER_SIDE_0;
+ for (i = 0; i < 4; i++)
+ {
+ if (old_face_mask & cur_outer_mask)
+ {
+ face_mapping[face_bit] = 5 + i;
+ break;
+ }
+ cur_outer_mask <<= 1;
+ }
+ if (i == 4)
+ {
+ LL_WARNS() << "No path end or outer face in volume!" << LL_ENDL;
+ }
+ continue;
+ }
+ }
+
+ if (cur_mask & (LL_FACE_PROFILE_BEGIN | LL_FACE_PROFILE_END))
+ {
+ // A cut slice. Use the hollow interior if we have it.
+ if (old_face_mask & LL_FACE_INNER_SIDE)
+ {
+ face_mapping[face_bit] = 2;
+ continue;
+ }
+
+ // No interior, use the bottom face.
+ // Could figure out which of the outer faces was nearest, but that would be harder.
+ if (old_face_mask & LL_FACE_PATH_END)
+ {
+ face_mapping[face_bit] = 1;
+ continue;
+ }
+ else
+ {
+ S32 cur_outer_mask = LL_FACE_OUTER_SIDE_0;
+ for (i = 0; i < 4; i++)
+ {
+ if (old_face_mask & cur_outer_mask)
+ {
+ face_mapping[face_bit] = 5 + i;
+ break;
+ }
+ cur_outer_mask <<= 1;
+ }
+ if (i == 4)
+ {
+ LL_WARNS() << "No path end or outer face in volume!" << LL_ENDL;
+ }
+ continue;
+ }
+ }
+
+ // OK, the face that's missing is an outer face...
+ // Pull from the nearest adjacent outer face (there's always guaranteed to be one...
+ S32 cur_outer = face_bit - 5;
+ S32 min_dist = 5;
+ S32 min_outer_bit = -1;
+ S32 i;
+ for (i = 0; i < 4; i++)
+ {
+ if (old_face_mask & (LL_FACE_OUTER_SIDE_0 << i))
+ {
+ S32 dist = abs(i - cur_outer);
+ if (dist < min_dist)
+ {
+ min_dist = dist;
+ min_outer_bit = i + 5;
+ }
+ }
+ }
+ if (-1 == min_outer_bit)
+ {
+ LL_INFOS() << (LLVolume *)mVolumep << LL_ENDL;
+ LL_WARNS() << "Bad! No outer faces, impossible!" << LL_ENDL;
+ }
+ face_mapping[face_bit] = min_outer_bit;
+ }
+
+
+ setNumTEs(mVolumep->getNumFaces());
+ for (face_bit = 0; face_bit < 9; face_bit++)
+ {
+ // For each possible face type on the new shape we check to see if that
+ // face exists and if it does we create a texture entry that is a copy
+ // of one of the originals. Since the originals might not have a
+ // matching face, we use the face_mapping lookup table to figure out
+ // which face information to copy.
+ cur_mask = 0x1 << face_bit;
+ if (new_face_mask & cur_mask)
+ {
+ if (-1 == face_mapping[face_bit])
+ {
+ LL_WARNS() << "No mapping from old face to new face!" << LL_ENDL;
+ }
+
+ S32 te_num = face_index_from_id(cur_mask, mVolumep->getProfile().mFaces);
+ setTE(te_num, *(old_tes.getTexture(face_mapping[face_bit])));
+ }
+ }
#else
- // build the new object
- sVolumeManager->unrefVolume(mVolumep);
- mVolumep = volumep;
+ // build the new object
+ sVolumeManager->unrefVolume(mVolumep);
+ mVolumep = volumep;
- setNumTEs(mVolumep->getNumFaces());
+ setNumTEs(mVolumep->getNumFaces());
#endif
- return TRUE;
+ return TRUE;
}
BOOL LLPrimitive::setMaterial(U8 material)
{
- if (material != mMaterial)
- {
- mMaterial = material;
- return TRUE;
- }
- else
- {
- return FALSE;
- }
+ if (material != mMaterial)
+ {
+ mMaterial = material;
+ return TRUE;
+ }
+ else
+ {
+ return FALSE;
+ }
}
S32 LLPrimitive::packTEField(U8 *cur_ptr, U8 *data_ptr, U8 data_size, U8 last_face_index, EMsgVariableType type) const
{
- S32 face_index;
- S32 i;
- U64 exception_faces;
- U8 *start_loc = cur_ptr;
-
- htolememcpy(cur_ptr,data_ptr + (last_face_index * data_size), type, data_size);
- cur_ptr += data_size;
-
- for (face_index = last_face_index-1; face_index >= 0; face_index--)
- {
- BOOL already_sent = FALSE;
- for (i = face_index+1; i <= last_face_index; i++)
- {
- if (!memcmp(data_ptr+(data_size *face_index), data_ptr+(data_size *i), data_size))
- {
- already_sent = TRUE;
- break;
- }
- }
-
- if (!already_sent)
- {
- exception_faces = 0;
- for (i = face_index; i >= 0; i--)
- {
- if (!memcmp(data_ptr+(data_size *face_index), data_ptr+(data_size *i), data_size))
- {
- exception_faces |= ((U64)1 << i);
- }
- }
-
- //assign exception faces to cur_ptr
- if (exception_faces >= ((U64)0x1 << 7))
- {
- if (exception_faces >= ((U64)0x1 << 14))
- {
- if (exception_faces >= ((U64)0x1 << 21))
- {
- if (exception_faces >= ((U64)0x1 << 28))
- {
- if (exception_faces >= ((U64)0x1 << 35))
- {
- if (exception_faces >= ((U64)0x1 << 42))
- {
- if (exception_faces >= ((U64)0x1 << 49))
- {
- *cur_ptr++ = (U8)(((exception_faces >> 49) & 0x7F) | 0x80);
- }
- *cur_ptr++ = (U8)(((exception_faces >> 42) & 0x7F) | 0x80);
- }
- *cur_ptr++ = (U8)(((exception_faces >> 35) & 0x7F) | 0x80);
- }
- *cur_ptr++ = (U8)(((exception_faces >> 28) & 0x7F) | 0x80);
- }
- *cur_ptr++ = (U8)(((exception_faces >> 21) & 0x7F) | 0x80);
- }
- *cur_ptr++ = (U8)(((exception_faces >> 14) & 0x7F) | 0x80);
- }
- *cur_ptr++ = (U8)(((exception_faces >> 7) & 0x7F) | 0x80);
- }
-
-
- *cur_ptr++ = (U8)(exception_faces & 0x7F);
-
- htolememcpy(cur_ptr,data_ptr + (face_index * data_size), type, data_size);
- cur_ptr += data_size;
- }
- }
- return (S32)(cur_ptr - start_loc);
+ S32 face_index;
+ S32 i;
+ U64 exception_faces;
+ U8 *start_loc = cur_ptr;
+
+ htolememcpy(cur_ptr,data_ptr + (last_face_index * data_size), type, data_size);
+ cur_ptr += data_size;
+
+ for (face_index = last_face_index-1; face_index >= 0; face_index--)
+ {
+ BOOL already_sent = FALSE;
+ for (i = face_index+1; i <= last_face_index; i++)
+ {
+ if (!memcmp(data_ptr+(data_size *face_index), data_ptr+(data_size *i), data_size))
+ {
+ already_sent = TRUE;
+ break;
+ }
+ }
+
+ if (!already_sent)
+ {
+ exception_faces = 0;
+ for (i = face_index; i >= 0; i--)
+ {
+ if (!memcmp(data_ptr+(data_size *face_index), data_ptr+(data_size *i), data_size))
+ {
+ exception_faces |= ((U64)1 << i);
+ }
+ }
+
+ //assign exception faces to cur_ptr
+ if (exception_faces >= ((U64)0x1 << 7))
+ {
+ if (exception_faces >= ((U64)0x1 << 14))
+ {
+ if (exception_faces >= ((U64)0x1 << 21))
+ {
+ if (exception_faces >= ((U64)0x1 << 28))
+ {
+ if (exception_faces >= ((U64)0x1 << 35))
+ {
+ if (exception_faces >= ((U64)0x1 << 42))
+ {
+ if (exception_faces >= ((U64)0x1 << 49))
+ {
+ *cur_ptr++ = (U8)(((exception_faces >> 49) & 0x7F) | 0x80);
+ }
+ *cur_ptr++ = (U8)(((exception_faces >> 42) & 0x7F) | 0x80);
+ }
+ *cur_ptr++ = (U8)(((exception_faces >> 35) & 0x7F) | 0x80);
+ }
+ *cur_ptr++ = (U8)(((exception_faces >> 28) & 0x7F) | 0x80);
+ }
+ *cur_ptr++ = (U8)(((exception_faces >> 21) & 0x7F) | 0x80);
+ }
+ *cur_ptr++ = (U8)(((exception_faces >> 14) & 0x7F) | 0x80);
+ }
+ *cur_ptr++ = (U8)(((exception_faces >> 7) & 0x7F) | 0x80);
+ }
+
+
+ *cur_ptr++ = (U8)(exception_faces & 0x7F);
+
+ htolememcpy(cur_ptr,data_ptr + (face_index * data_size), type, data_size);
+ cur_ptr += data_size;
+ }
+ }
+ return (S32)(cur_ptr - start_loc);
}
namespace
@@ -1136,7 +1136,7 @@ namespace
return false;
}
- // Extract the default value and fill the array.
+ // Extract the default value and fill the array.
htolememcpy(dest, source, type, size);
source += size;
for (S32 idx = 1; idx < dest_count; ++idx)
@@ -1149,7 +1149,7 @@ namespace
U64 index_flags(0);
U8 sbit(0);
- // Unpack the variable length bitfield. Each bit represents whether the following
+ // Unpack the variable length bitfield. Each bit represents whether the following
// value will be placed at the corresponding array index.
do
{
@@ -1203,194 +1203,194 @@ namespace
// Includes information about image ID, color, scale S,T, offset S,T and rotation
BOOL LLPrimitive::packTEMessage(LLMessageSystem *mesgsys) const
{
- const U32 MAX_TES = 45;
-
- U8 image_ids[MAX_TES*16];
- U8 colors[MAX_TES*4];
- F32 scale_s[MAX_TES];
- F32 scale_t[MAX_TES];
- S16 offset_s[MAX_TES];
- S16 offset_t[MAX_TES];
- S16 image_rot[MAX_TES];
- U8 bump[MAX_TES];
- U8 media_flags[MAX_TES];
+ const U32 MAX_TES = 45;
+
+ U8 image_ids[MAX_TES*16];
+ U8 colors[MAX_TES*4];
+ F32 scale_s[MAX_TES];
+ F32 scale_t[MAX_TES];
+ S16 offset_s[MAX_TES];
+ S16 offset_t[MAX_TES];
+ S16 image_rot[MAX_TES];
+ U8 bump[MAX_TES];
+ U8 media_flags[MAX_TES];
U8 glow[MAX_TES];
- U8 material_data[MAX_TES*16];
-
- const U32 MAX_TE_BUFFER = 4096;
- U8 packed_buffer[MAX_TE_BUFFER];
- U8 *cur_ptr = packed_buffer;
-
- S32 last_face_index = llmin((U32) getNumTEs(), MAX_TES) - 1;
-
- if (last_face_index > -1)
- {
- // ...if we hit the front, send one image id
- S8 face_index;
- LLColor4U coloru;
- for (face_index = 0; face_index <= last_face_index; face_index++)
- {
- // Directly sending image_ids is not safe!
- memcpy(&image_ids[face_index*16],getTE(face_index)->getID().mData,16); /* Flawfinder: ignore */
-
- // Cast LLColor4 to LLColor4U
- coloru.setVec( getTE(face_index)->getColor() );
-
- // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
- // as all zeros. However, the subtraction and addition must be done in unsigned
- // byte space, not in float space, otherwise off-by-one errors occur. JC
- colors[4*face_index] = 255 - coloru.mV[0];
- colors[4*face_index + 1] = 255 - coloru.mV[1];
- colors[4*face_index + 2] = 255 - coloru.mV[2];
- colors[4*face_index + 3] = 255 - coloru.mV[3];
-
- const LLTextureEntry* te = getTE(face_index);
- scale_s[face_index] = (F32) te->mScaleS;
- scale_t[face_index] = (F32) te->mScaleT;
- offset_s[face_index] = (S16) ll_round((llclamp(te->mOffsetS,-1.0f,1.0f) * (F32)0x7FFF)) ;
- offset_t[face_index] = (S16) ll_round((llclamp(te->mOffsetT,-1.0f,1.0f) * (F32)0x7FFF)) ;
- image_rot[face_index] = (S16) ll_round(((fmod(te->mRotation, F_TWO_PI)/F_TWO_PI) * TEXTURE_ROTATION_PACK_FACTOR));
- bump[face_index] = te->getBumpShinyFullbright();
- media_flags[face_index] = te->getMediaTexGen();
- glow[face_index] = (U8) ll_round((llclamp(te->getGlow(), 0.0f, 1.0f) * (F32)0xFF));
-
- // Directly sending material_ids is not safe!
- memcpy(&material_data[face_index*16],getTE(face_index)->getMaterialID().get(),16); /* Flawfinder: ignore */
- }
-
- cur_ptr += packTEField(cur_ptr, (U8 *)image_ids, sizeof(LLUUID),last_face_index, MVT_LLUUID);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)colors, 4 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)scale_s, 4 ,last_face_index, MVT_F32);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)scale_t, 4 ,last_face_index, MVT_F32);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)offset_s, 2 ,last_face_index, MVT_S16Array);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)offset_t, 2 ,last_face_index, MVT_S16Array);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)image_rot, 2 ,last_face_index, MVT_S16Array);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)bump, 1 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)media_flags, 1 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)glow, 1 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)material_data, 16, last_face_index, MVT_LLUUID);
- }
- mesgsys->addBinaryDataFast(_PREHASH_TextureEntry, packed_buffer, (S32)(cur_ptr - packed_buffer));
-
- return FALSE;
+ U8 material_data[MAX_TES*16];
+
+ const U32 MAX_TE_BUFFER = 4096;
+ U8 packed_buffer[MAX_TE_BUFFER];
+ U8 *cur_ptr = packed_buffer;
+
+ S32 last_face_index = llmin((U32) getNumTEs(), MAX_TES) - 1;
+
+ if (last_face_index > -1)
+ {
+ // ...if we hit the front, send one image id
+ S8 face_index;
+ LLColor4U coloru;
+ for (face_index = 0; face_index <= last_face_index; face_index++)
+ {
+ // Directly sending image_ids is not safe!
+ memcpy(&image_ids[face_index*16],getTE(face_index)->getID().mData,16); /* Flawfinder: ignore */
+
+ // Cast LLColor4 to LLColor4U
+ coloru.setVec( getTE(face_index)->getColor() );
+
+ // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
+ // as all zeros. However, the subtraction and addition must be done in unsigned
+ // byte space, not in float space, otherwise off-by-one errors occur. JC
+ colors[4*face_index] = 255 - coloru.mV[0];
+ colors[4*face_index + 1] = 255 - coloru.mV[1];
+ colors[4*face_index + 2] = 255 - coloru.mV[2];
+ colors[4*face_index + 3] = 255 - coloru.mV[3];
+
+ const LLTextureEntry* te = getTE(face_index);
+ scale_s[face_index] = (F32) te->mScaleS;
+ scale_t[face_index] = (F32) te->mScaleT;
+ offset_s[face_index] = (S16) ll_round((llclamp(te->mOffsetS,-1.0f,1.0f) * (F32)0x7FFF)) ;
+ offset_t[face_index] = (S16) ll_round((llclamp(te->mOffsetT,-1.0f,1.0f) * (F32)0x7FFF)) ;
+ image_rot[face_index] = (S16) ll_round(((fmod(te->mRotation, F_TWO_PI)/F_TWO_PI) * TEXTURE_ROTATION_PACK_FACTOR));
+ bump[face_index] = te->getBumpShinyFullbright();
+ media_flags[face_index] = te->getMediaTexGen();
+ glow[face_index] = (U8) ll_round((llclamp(te->getGlow(), 0.0f, 1.0f) * (F32)0xFF));
+
+ // Directly sending material_ids is not safe!
+ memcpy(&material_data[face_index*16],getTE(face_index)->getMaterialID().get(),16); /* Flawfinder: ignore */
+ }
+
+ cur_ptr += packTEField(cur_ptr, (U8 *)image_ids, sizeof(LLUUID),last_face_index, MVT_LLUUID);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)colors, 4 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)scale_s, 4 ,last_face_index, MVT_F32);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)scale_t, 4 ,last_face_index, MVT_F32);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)offset_s, 2 ,last_face_index, MVT_S16Array);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)offset_t, 2 ,last_face_index, MVT_S16Array);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)image_rot, 2 ,last_face_index, MVT_S16Array);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)bump, 1 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)media_flags, 1 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)glow, 1 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)material_data, 16, last_face_index, MVT_LLUUID);
+ }
+ mesgsys->addBinaryDataFast(_PREHASH_TextureEntry, packed_buffer, (S32)(cur_ptr - packed_buffer));
+
+ return FALSE;
}
BOOL LLPrimitive::packTEMessage(LLDataPacker &dp) const
{
- const U32 MAX_TES = 45;
-
- U8 image_ids[MAX_TES*16];
- U8 colors[MAX_TES*4];
- F32 scale_s[MAX_TES];
- F32 scale_t[MAX_TES];
- S16 offset_s[MAX_TES];
- S16 offset_t[MAX_TES];
- S16 image_rot[MAX_TES];
- U8 bump[MAX_TES];
- U8 media_flags[MAX_TES];
+ const U32 MAX_TES = 45;
+
+ U8 image_ids[MAX_TES*16];
+ U8 colors[MAX_TES*4];
+ F32 scale_s[MAX_TES];
+ F32 scale_t[MAX_TES];
+ S16 offset_s[MAX_TES];
+ S16 offset_t[MAX_TES];
+ S16 image_rot[MAX_TES];
+ U8 bump[MAX_TES];
+ U8 media_flags[MAX_TES];
U8 glow[MAX_TES];
- U8 material_data[MAX_TES*16];
-
- const U32 MAX_TE_BUFFER = 4096;
- U8 packed_buffer[MAX_TE_BUFFER];
- U8 *cur_ptr = packed_buffer;
-
- S32 last_face_index = getNumTEs() - 1;
-
- if (last_face_index > -1)
- {
- // ...if we hit the front, send one image id
- S8 face_index;
- LLColor4U coloru;
- for (face_index = 0; face_index <= last_face_index; face_index++)
- {
- // Directly sending image_ids is not safe!
- memcpy(&image_ids[face_index*16],getTE(face_index)->getID().mData,16); /* Flawfinder: ignore */
-
- // Cast LLColor4 to LLColor4U
- coloru.setVec( getTE(face_index)->getColor() );
-
- // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
- // as all zeros. However, the subtraction and addition must be done in unsigned
- // byte space, not in float space, otherwise off-by-one errors occur. JC
- colors[4*face_index] = 255 - coloru.mV[0];
- colors[4*face_index + 1] = 255 - coloru.mV[1];
- colors[4*face_index + 2] = 255 - coloru.mV[2];
- colors[4*face_index + 3] = 255 - coloru.mV[3];
-
- const LLTextureEntry* te = getTE(face_index);
- scale_s[face_index] = (F32) te->mScaleS;
- scale_t[face_index] = (F32) te->mScaleT;
- offset_s[face_index] = (S16) ll_round((llclamp(te->mOffsetS,-1.0f,1.0f) * (F32)0x7FFF)) ;
- offset_t[face_index] = (S16) ll_round((llclamp(te->mOffsetT,-1.0f,1.0f) * (F32)0x7FFF)) ;
- image_rot[face_index] = (S16) ll_round(((fmod(te->mRotation, F_TWO_PI)/F_TWO_PI) * TEXTURE_ROTATION_PACK_FACTOR));
- bump[face_index] = te->getBumpShinyFullbright();
- media_flags[face_index] = te->getMediaTexGen();
+ U8 material_data[MAX_TES*16];
+
+ const U32 MAX_TE_BUFFER = 4096;
+ U8 packed_buffer[MAX_TE_BUFFER];
+ U8 *cur_ptr = packed_buffer;
+
+ S32 last_face_index = getNumTEs() - 1;
+
+ if (last_face_index > -1)
+ {
+ // ...if we hit the front, send one image id
+ S8 face_index;
+ LLColor4U coloru;
+ for (face_index = 0; face_index <= last_face_index; face_index++)
+ {
+ // Directly sending image_ids is not safe!
+ memcpy(&image_ids[face_index*16],getTE(face_index)->getID().mData,16); /* Flawfinder: ignore */
+
+ // Cast LLColor4 to LLColor4U
+ coloru.setVec( getTE(face_index)->getColor() );
+
+ // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
+ // as all zeros. However, the subtraction and addition must be done in unsigned
+ // byte space, not in float space, otherwise off-by-one errors occur. JC
+ colors[4*face_index] = 255 - coloru.mV[0];
+ colors[4*face_index + 1] = 255 - coloru.mV[1];
+ colors[4*face_index + 2] = 255 - coloru.mV[2];
+ colors[4*face_index + 3] = 255 - coloru.mV[3];
+
+ const LLTextureEntry* te = getTE(face_index);
+ scale_s[face_index] = (F32) te->mScaleS;
+ scale_t[face_index] = (F32) te->mScaleT;
+ offset_s[face_index] = (S16) ll_round((llclamp(te->mOffsetS,-1.0f,1.0f) * (F32)0x7FFF)) ;
+ offset_t[face_index] = (S16) ll_round((llclamp(te->mOffsetT,-1.0f,1.0f) * (F32)0x7FFF)) ;
+ image_rot[face_index] = (S16) ll_round(((fmod(te->mRotation, F_TWO_PI)/F_TWO_PI) * TEXTURE_ROTATION_PACK_FACTOR));
+ bump[face_index] = te->getBumpShinyFullbright();
+ media_flags[face_index] = te->getMediaTexGen();
glow[face_index] = (U8) ll_round((llclamp(te->getGlow(), 0.0f, 1.0f) * (F32)0xFF));
- // Directly sending material_ids is not safe!
- memcpy(&material_data[face_index*16],getTE(face_index)->getMaterialID().get(),16); /* Flawfinder: ignore */
- }
-
- cur_ptr += packTEField(cur_ptr, (U8 *)image_ids, sizeof(LLUUID),last_face_index, MVT_LLUUID);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)colors, 4 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)scale_s, 4 ,last_face_index, MVT_F32);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)scale_t, 4 ,last_face_index, MVT_F32);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)offset_s, 2 ,last_face_index, MVT_S16Array);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)offset_t, 2 ,last_face_index, MVT_S16Array);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)image_rot, 2 ,last_face_index, MVT_S16Array);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)bump, 1 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)media_flags, 1 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)glow, 1 ,last_face_index, MVT_U8);
- *cur_ptr++ = 0;
- cur_ptr += packTEField(cur_ptr, (U8 *)material_data, 16, last_face_index, MVT_LLUUID);
- }
-
- dp.packBinaryData(packed_buffer, (S32)(cur_ptr - packed_buffer), "TextureEntry");
- return FALSE;
+ // Directly sending material_ids is not safe!
+ memcpy(&material_data[face_index*16],getTE(face_index)->getMaterialID().get(),16); /* Flawfinder: ignore */
+ }
+
+ cur_ptr += packTEField(cur_ptr, (U8 *)image_ids, sizeof(LLUUID),last_face_index, MVT_LLUUID);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)colors, 4 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)scale_s, 4 ,last_face_index, MVT_F32);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)scale_t, 4 ,last_face_index, MVT_F32);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)offset_s, 2 ,last_face_index, MVT_S16Array);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)offset_t, 2 ,last_face_index, MVT_S16Array);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)image_rot, 2 ,last_face_index, MVT_S16Array);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)bump, 1 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)media_flags, 1 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)glow, 1 ,last_face_index, MVT_U8);
+ *cur_ptr++ = 0;
+ cur_ptr += packTEField(cur_ptr, (U8 *)material_data, 16, last_face_index, MVT_LLUUID);
+ }
+
+ dp.packBinaryData(packed_buffer, (S32)(cur_ptr - packed_buffer), "TextureEntry");
+ return FALSE;
}
S32 LLPrimitive::parseTEMessage(LLMessageSystem* mesgsys, char const* block_name, const S32 block_num, LLTEContents& tec)
{
- S32 retval = 0;
+ S32 retval = 0;
// temp buffer for material ID processing
- // data will end up in tec.material_id[]
+ // data will end up in tec.material_id[]
material_id_type material_data[LLTEContents::MAX_TES];
- if (block_num < 0)
- {
- tec.size = mesgsys->getSizeFast(block_name, _PREHASH_TextureEntry);
- }
- else
- {
- tec.size = mesgsys->getSizeFast(block_name, block_num, _PREHASH_TextureEntry);
- }
-
- if (tec.size == 0)
- {
- tec.face_count = 0;
- return retval;
- }
+ if (block_num < 0)
+ {
+ tec.size = mesgsys->getSizeFast(block_name, _PREHASH_TextureEntry);
+ }
+ else
+ {
+ tec.size = mesgsys->getSizeFast(block_name, block_num, _PREHASH_TextureEntry);
+ }
+
+ if (tec.size == 0)
+ {
+ tec.face_count = 0;
+ return retval;
+ }
else if (tec.size >= LLTEContents::MAX_TE_BUFFER)
{
LL_WARNS("TEXTUREENTRY") << "Excessive buffer size detected in Texture Entry! Truncating." << LL_ENDL;
@@ -1400,14 +1400,14 @@ S32 LLPrimitive::parseTEMessage(LLMessageSystem* mesgsys, char const* block_name
// if block_num < 0 ask for block 0
mesgsys->getBinaryDataFast(block_name, _PREHASH_TextureEntry, tec.packed_buffer, 0, std::max(block_num, 0), LLTEContents::MAX_TE_BUFFER - 1);
- // The last field is not zero terminated.
+ // The last field is not zero terminated.
// Rather than special case the upack functions. Just make it 0x00 terminated.
tec.packed_buffer[tec.size] = 0x00;
++tec.size;
- tec.face_count = llmin((U32)getNumTEs(),(U32)LLTEContents::MAX_TES);
+ tec.face_count = llmin((U32)getNumTEs(),(U32)LLTEContents::MAX_TES);
- U8 *cur_ptr = tec.packed_buffer;
+ U8 *cur_ptr = tec.packed_buffer;
LL_DEBUGS("TEXTUREENTRY") << "Texture Entry with buffere sized: " << tec.size << LL_ENDL;
U8 *buffer_end = tec.packed_buffer + tec.size;
@@ -1426,68 +1426,68 @@ S32 LLPrimitive::parseTEMessage(LLMessageSystem* mesgsys, char const* block_name
return 0;
}
- if (cur_ptr >= buffer_end || !unpack_TEField<material_id_type>(material_data, tec.face_count, cur_ptr, buffer_end, MVT_LLUUID))
- {
- memset((void*)material_data, 0, sizeof(material_data));
- }
-
- for (U32 i = 0; i < tec.face_count; i++)
- {
- tec.material_ids[i].set(&(material_data[i]));
- }
-
- retval = 1;
- return retval;
- }
-
+ if (cur_ptr >= buffer_end || !unpack_TEField<material_id_type>(material_data, tec.face_count, cur_ptr, buffer_end, MVT_LLUUID))
+ {
+ memset((void*)material_data, 0, sizeof(material_data));
+ }
+
+ for (U32 i = 0; i < tec.face_count; i++)
+ {
+ tec.material_ids[i].set(&(material_data[i]));
+ }
+
+ retval = 1;
+ return retval;
+ }
+
S32 LLPrimitive::applyParsedTEMessage(LLTEContents& tec)
{
- S32 retval = 0;
-
- LLColor4 color;
- for (U32 i = 0; i < tec.face_count; i++)
- {
- LLUUID& req_id = ((LLUUID*)tec.image_data)[i];
- retval |= setTETexture(i, req_id);
- retval |= setTEScale(i, tec.scale_s[i], tec.scale_t[i]);
- retval |= setTEOffset(i, (F32)tec.offset_s[i] / (F32)0x7FFF, (F32) tec.offset_t[i] / (F32) 0x7FFF);
- retval |= setTERotation(i, ((F32)tec.image_rot[i] / TEXTURE_ROTATION_PACK_FACTOR) * F_TWO_PI);
- retval |= setTEBumpShinyFullbright(i, tec.bump[i]);
- retval |= setTEMediaTexGen(i, tec.media_flags[i]);
- retval |= setTEGlow(i, (F32)tec.glow[i] / (F32)0xFF);
- retval |= setTEMaterialID(i, tec.material_ids[i]);
-
- // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
- // as all zeros. However, the subtraction and addition must be done in unsigned
- // byte space, not in float space, otherwise off-by-one errors occur. JC
- color.mV[VRED] = F32(255 - tec.colors[i].mV[VRED]) / 255.f;
- color.mV[VGREEN] = F32(255 - tec.colors[i].mV[VGREEN]) / 255.f;
- color.mV[VBLUE] = F32(255 - tec.colors[i].mV[VBLUE]) / 255.f;
- color.mV[VALPHA] = F32(255 - tec.colors[i].mV[VALPHA]) / 255.f;
-
- retval |= setTEColor(i, color);
- }
-
- return retval;
+ S32 retval = 0;
+
+ LLColor4 color;
+ for (U32 i = 0; i < tec.face_count; i++)
+ {
+ LLUUID& req_id = ((LLUUID*)tec.image_data)[i];
+ retval |= setTETexture(i, req_id);
+ retval |= setTEScale(i, tec.scale_s[i], tec.scale_t[i]);
+ retval |= setTEOffset(i, (F32)tec.offset_s[i] / (F32)0x7FFF, (F32) tec.offset_t[i] / (F32) 0x7FFF);
+ retval |= setTERotation(i, ((F32)tec.image_rot[i] / TEXTURE_ROTATION_PACK_FACTOR) * F_TWO_PI);
+ retval |= setTEBumpShinyFullbright(i, tec.bump[i]);
+ retval |= setTEMediaTexGen(i, tec.media_flags[i]);
+ retval |= setTEGlow(i, (F32)tec.glow[i] / (F32)0xFF);
+ retval |= setTEMaterialID(i, tec.material_ids[i]);
+
+ // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
+ // as all zeros. However, the subtraction and addition must be done in unsigned
+ // byte space, not in float space, otherwise off-by-one errors occur. JC
+ color.mV[VRED] = F32(255 - tec.colors[i].mV[VRED]) / 255.f;
+ color.mV[VGREEN] = F32(255 - tec.colors[i].mV[VGREEN]) / 255.f;
+ color.mV[VBLUE] = F32(255 - tec.colors[i].mV[VBLUE]) / 255.f;
+ color.mV[VALPHA] = F32(255 - tec.colors[i].mV[VALPHA]) / 255.f;
+
+ retval |= setTEColor(i, color);
+ }
+
+ return retval;
}
S32 LLPrimitive::unpackTEMessage(LLMessageSystem* mesgsys, char const* block_name, const S32 block_num)
{
- LLTEContents tec;
- S32 retval = parseTEMessage(mesgsys, block_name, block_num, tec);
- if (!retval)
- return retval;
- return applyParsedTEMessage(tec);
+ LLTEContents tec;
+ S32 retval = parseTEMessage(mesgsys, block_name, block_num, tec);
+ if (!retval)
+ return retval;
+ return applyParsedTEMessage(tec);
}
S32 LLPrimitive::unpackTEMessage(LLDataPacker &dp)
{
- // use a negative block_num to indicate a single-block read (a non-variable block)
- S32 retval = 0;
- const U32 MAX_TES = 45;
+ // use a negative block_num to indicate a single-block read (a non-variable block)
+ S32 retval = 0;
+ const U32 MAX_TES = 45;
- // Avoid construction of 32 UUIDs per call
- static LLMaterialID material_ids[MAX_TES];
+ // Avoid construction of 32 UUIDs per call
+ static LLMaterialID material_ids[MAX_TES];
const U32 MAX_TE_BUFFER = 4096;
U8 packed_buffer[MAX_TE_BUFFER];
@@ -1504,7 +1504,7 @@ S32 LLPrimitive::unpackTEMessage(LLDataPacker &dp)
U8 media_flags[MAX_TES];
U8 glow[MAX_TES];
material_id_type material_data[MAX_TES];
-
+
memset((void*)scale_s, 0, sizeof(scale_s));
memset((void*)scale_t, 0, sizeof(scale_t));
memset((void*)offset_s, 0, sizeof(offset_s));
@@ -1514,32 +1514,32 @@ S32 LLPrimitive::unpackTEMessage(LLDataPacker &dp)
memset((void*)media_flags, 0, sizeof(media_flags));
memset((void*)glow, 0, sizeof(glow));
- S32 size;
- U32 face_count = 0;
+ S32 size;
+ U32 face_count = 0;
- if (!dp.unpackBinaryData(packed_buffer, size, "TextureEntry"))
- {
- retval = TEM_INVALID;
- LL_WARNS() << "Bad texture entry block! Abort!" << LL_ENDL;
- return retval;
- }
+ if (!dp.unpackBinaryData(packed_buffer, size, "TextureEntry"))
+ {
+ retval = TEM_INVALID;
+ LL_WARNS() << "Bad texture entry block! Abort!" << LL_ENDL;
+ return retval;
+ }
- if (size == 0)
- {
- return retval;
- }
+ if (size == 0)
+ {
+ return retval;
+ }
else if (size >= MAX_TE_BUFFER)
{
LL_WARNS("TEXTUREENTRY") << "Excessive buffer size detected in Texture Entry! Truncating." << LL_ENDL;
size = MAX_TE_BUFFER - 1;
}
- // The last field is not zero terminated.
+ // The last field is not zero terminated.
// Rather than special case the upack functions. Just make it 0x00 terminated.
packed_buffer[size] = 0x00;
++size;
- face_count = llmin((U32) getNumTEs(), MAX_TES);
- U32 i;
+ face_count = llmin((U32) getNumTEs(), MAX_TES);
+ U32 i;
U8 *cur_ptr = packed_buffer;
LL_DEBUGS("TEXTUREENTRY") << "Texture Entry with buffer sized: " << size << LL_ENDL;
@@ -1560,81 +1560,81 @@ S32 LLPrimitive::unpackTEMessage(LLDataPacker &dp)
return 0;
}
- if (cur_ptr >= buffer_end || !unpack_TEField<material_id_type>(material_data, face_count, cur_ptr, buffer_end, MVT_LLUUID))
- {
- memset((void*)material_data, 0, sizeof(material_data));
- }
-
- for (i = 0; i < face_count; i++)
- {
- material_ids[i].set(&(material_data[i]));
- }
-
- LLColor4 color;
- for (i = 0; i < face_count; i++)
- {
- retval |= setTETexture(i, ((LLUUID*)image_data)[i]);
- retval |= setTEScale(i, scale_s[i], scale_t[i]);
- retval |= setTEOffset(i, (F32)offset_s[i] / (F32)0x7FFF, (F32) offset_t[i] / (F32) 0x7FFF);
- retval |= setTERotation(i, ((F32)image_rot[i] / TEXTURE_ROTATION_PACK_FACTOR) * F_TWO_PI);
- retval |= setTEBumpShinyFullbright(i, bump[i]);
- retval |= setTEMediaTexGen(i, media_flags[i]);
- retval |= setTEGlow(i, (F32)glow[i] / (F32)0xFF);
- retval |= setTEMaterialID(i, material_ids[i]);
+ if (cur_ptr >= buffer_end || !unpack_TEField<material_id_type>(material_data, face_count, cur_ptr, buffer_end, MVT_LLUUID))
+ {
+ memset((void*)material_data, 0, sizeof(material_data));
+ }
- // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
- // as all zeros. However, the subtraction and addition must be done in unsigned
- // byte space, not in float space, otherwise off-by-one errors occur. JC
- color.mV[VRED] = F32(255 - colors[i].mV[VRED]) / 255.f;
- color.mV[VGREEN] = F32(255 - colors[i].mV[VGREEN]) / 255.f;
- color.mV[VBLUE] = F32(255 - colors[i].mV[VBLUE]) / 255.f;
- color.mV[VALPHA] = F32(255 - colors[i].mV[VALPHA]) / 255.f;
+ for (i = 0; i < face_count; i++)
+ {
+ material_ids[i].set(&(material_data[i]));
+ }
- retval |= setTEColor(i, color);
- }
+ LLColor4 color;
+ for (i = 0; i < face_count; i++)
+ {
+ retval |= setTETexture(i, ((LLUUID*)image_data)[i]);
+ retval |= setTEScale(i, scale_s[i], scale_t[i]);
+ retval |= setTEOffset(i, (F32)offset_s[i] / (F32)0x7FFF, (F32) offset_t[i] / (F32) 0x7FFF);
+ retval |= setTERotation(i, ((F32)image_rot[i] / TEXTURE_ROTATION_PACK_FACTOR) * F_TWO_PI);
+ retval |= setTEBumpShinyFullbright(i, bump[i]);
+ retval |= setTEMediaTexGen(i, media_flags[i]);
+ retval |= setTEGlow(i, (F32)glow[i] / (F32)0xFF);
+ retval |= setTEMaterialID(i, material_ids[i]);
+
+ // Note: This is an optimization to send common colors (1.f, 1.f, 1.f, 1.f)
+ // as all zeros. However, the subtraction and addition must be done in unsigned
+ // byte space, not in float space, otherwise off-by-one errors occur. JC
+ color.mV[VRED] = F32(255 - colors[i].mV[VRED]) / 255.f;
+ color.mV[VGREEN] = F32(255 - colors[i].mV[VGREEN]) / 255.f;
+ color.mV[VBLUE] = F32(255 - colors[i].mV[VBLUE]) / 255.f;
+ color.mV[VALPHA] = F32(255 - colors[i].mV[VALPHA]) / 255.f;
+
+ retval |= setTEColor(i, color);
+ }
- return retval;
+ return retval;
}
-U8 LLPrimitive::getExpectedNumTEs() const
+U8 LLPrimitive::getExpectedNumTEs() const
{
- U8 expected_face_count = 0;
- if (mVolumep)
- {
- expected_face_count = mVolumep->getNumFaces();
- }
- return expected_face_count;
+ U8 expected_face_count = 0;
+ if (mVolumep)
+ {
+ expected_face_count = mVolumep->getNumFaces();
+ }
+ return expected_face_count;
}
void LLPrimitive::copyTextureList(const LLPrimTextureList& other_list)
{
- mTextureList.copy(other_list);
+ mTextureList.copy(other_list);
}
void LLPrimitive::takeTextureList(LLPrimTextureList& other_list)
{
- mTextureList.take(other_list);
+ mTextureList.take(other_list);
}
void LLPrimitive::updateNumBumpmap(const U8 index, const U8 bump)
{
- LLTextureEntry* te = getTE(index);
- if(!te)
- {
- return;
- }
+ LLTextureEntry* te = getTE(index);
+ if(!te)
+ {
+ return;
+ }
- U8 old_bump = te->getBumpmap();
- if(old_bump > 0)
- {
- mNumBumpmapTEs--;
- }
- if((bump & TEM_BUMP_MASK) > 0)
- {
- mNumBumpmapTEs++;
- }
+ U8 old_bump = te->getBumpmap();
+ if(old_bump > 0)
+ {
+ mNumBumpmapTEs--;
+ }
+ if((bump & TEM_BUMP_MASK) > 0)
+ {
+ mNumBumpmapTEs++;
+ }
- return;
+ return;
}
//============================================================================
@@ -1645,180 +1645,180 @@ void LLPrimitive::updateNumBumpmap(const U8 index, const U8 bump)
// static
bool LLPrimitive::getTESTAxes(const U8 face, U32* s_axis, U32* t_axis)
{
- if (face == 0)
- {
- *s_axis = VX; *t_axis = VY;
- return true;
- }
- else if (face == 1)
- {
- *s_axis = VX; *t_axis = VZ;
- return true;
- }
- else if (face == 2)
- {
- *s_axis = VY; *t_axis = VZ;
- return true;
- }
- else if (face == 3)
- {
- *s_axis = VX; *t_axis = VZ;
- return true;
- }
- else if (face == 4)
- {
- *s_axis = VY; *t_axis = VZ;
- return true;
- }
- else if (face >= 5)
- {
- *s_axis = VX; *t_axis = VY;
- return true;
- }
- else
- {
- // unknown face
- return false;
- }
+ if (face == 0)
+ {
+ *s_axis = VX; *t_axis = VY;
+ return true;
+ }
+ else if (face == 1)
+ {
+ *s_axis = VX; *t_axis = VZ;
+ return true;
+ }
+ else if (face == 2)
+ {
+ *s_axis = VY; *t_axis = VZ;
+ return true;
+ }
+ else if (face == 3)
+ {
+ *s_axis = VX; *t_axis = VZ;
+ return true;
+ }
+ else if (face == 4)
+ {
+ *s_axis = VY; *t_axis = VZ;
+ return true;
+ }
+ else if (face >= 5)
+ {
+ *s_axis = VX; *t_axis = VY;
+ return true;
+ }
+ else
+ {
+ // unknown face
+ return false;
+ }
}
//============================================================================
-//static
+//static
BOOL LLNetworkData::isValid(U16 param_type, U32 size)
{
- // ew - better mechanism needed
-
- switch(param_type)
- {
- case PARAMS_FLEXIBLE:
- return (size == 16);
- case PARAMS_LIGHT:
- return (size == 16);
- case PARAMS_SCULPT:
- return (size == 17);
- case PARAMS_LIGHT_IMAGE:
- return (size == 28);
+ // ew - better mechanism needed
+
+ switch(param_type)
+ {
+ case PARAMS_FLEXIBLE:
+ return (size == 16);
+ case PARAMS_LIGHT:
+ return (size == 16);
+ case PARAMS_SCULPT:
+ return (size == 17);
+ case PARAMS_LIGHT_IMAGE:
+ return (size == 28);
case PARAMS_EXTENDED_MESH:
return (size == 4);
case PARAMS_RENDER_MATERIAL:
return (size > 1);
case PARAMS_REFLECTION_PROBE:
return (size == 9);
- }
-
- return FALSE;
+ }
+
+ return FALSE;
}
//============================================================================
LLLightParams::LLLightParams()
{
- mColor.setToWhite();
- mRadius = 10.f;
- mCutoff = 0.0f;
- mFalloff = 0.75f;
+ mColor.setToWhite();
+ mRadius = 10.f;
+ mCutoff = 0.0f;
+ mFalloff = 0.75f;
- mType = PARAMS_LIGHT;
+ mType = PARAMS_LIGHT;
}
BOOL LLLightParams::pack(LLDataPacker &dp) const
{
- LLColor4U color4u(mColor);
- dp.packColor4U(color4u, "color");
- dp.packF32(mRadius, "radius");
- dp.packF32(mCutoff, "cutoff");
- dp.packF32(mFalloff, "falloff");
- return TRUE;
+ LLColor4U color4u(mColor);
+ dp.packColor4U(color4u, "color");
+ dp.packF32(mRadius, "radius");
+ dp.packF32(mCutoff, "cutoff");
+ dp.packF32(mFalloff, "falloff");
+ return TRUE;
}
BOOL LLLightParams::unpack(LLDataPacker &dp)
{
- LLColor4U color;
- dp.unpackColor4U(color, "color");
- setLinearColor(LLColor4(color));
+ LLColor4U color;
+ dp.unpackColor4U(color, "color");
+ setLinearColor(LLColor4(color));
- F32 radius;
- dp.unpackF32(radius, "radius");
- setRadius(radius);
+ F32 radius;
+ dp.unpackF32(radius, "radius");
+ setRadius(radius);
- F32 cutoff;
- dp.unpackF32(cutoff, "cutoff");
- setCutoff(cutoff);
+ F32 cutoff;
+ dp.unpackF32(cutoff, "cutoff");
+ setCutoff(cutoff);
- F32 falloff;
- dp.unpackF32(falloff, "falloff");
- setFalloff(falloff);
-
- return TRUE;
+ F32 falloff;
+ dp.unpackF32(falloff, "falloff");
+ setFalloff(falloff);
+
+ return TRUE;
}
bool LLLightParams::operator==(const LLNetworkData& data) const
{
- if (data.mType != PARAMS_LIGHT)
- {
- return false;
- }
- const LLLightParams *param = (const LLLightParams*)&data;
- if (param->mColor != mColor ||
- param->mRadius != mRadius ||
- param->mCutoff != mCutoff ||
- param->mFalloff != mFalloff)
- {
- return false;
- }
- return true;
+ if (data.mType != PARAMS_LIGHT)
+ {
+ return false;
+ }
+ const LLLightParams *param = (const LLLightParams*)&data;
+ if (param->mColor != mColor ||
+ param->mRadius != mRadius ||
+ param->mCutoff != mCutoff ||
+ param->mFalloff != mFalloff)
+ {
+ return false;
+ }
+ return true;
}
void LLLightParams::copy(const LLNetworkData& data)
{
- const LLLightParams *param = (LLLightParams*)&data;
- mType = param->mType;
- mColor = param->mColor;
- mRadius = param->mRadius;
- mCutoff = param->mCutoff;
- mFalloff = param->mFalloff;
+ const LLLightParams *param = (LLLightParams*)&data;
+ mType = param->mType;
+ mColor = param->mColor;
+ mRadius = param->mRadius;
+ mCutoff = param->mCutoff;
+ mFalloff = param->mFalloff;
}
LLSD LLLightParams::asLLSD() const
{
- LLSD sd;
-
- sd["color"] = ll_sd_from_color4(getLinearColor());
- sd["radius"] = getRadius();
- sd["falloff"] = getFalloff();
- sd["cutoff"] = getCutoff();
-
- return sd;
+ LLSD sd;
+
+ sd["color"] = ll_sd_from_color4(getLinearColor());
+ sd["radius"] = getRadius();
+ sd["falloff"] = getFalloff();
+ sd["cutoff"] = getCutoff();
+
+ return sd;
}
bool LLLightParams::fromLLSD(LLSD& sd)
{
- const char *w;
- w = "color";
- if (sd.has(w))
- {
- setLinearColor( ll_color4_from_sd(sd["color"]) );
- } else goto fail;
- w = "radius";
- if (sd.has(w))
- {
- setRadius( (F32)sd[w].asReal() );
- } else goto fail;
- w = "falloff";
- if (sd.has(w))
- {
- setFalloff( (F32)sd[w].asReal() );
- } else goto fail;
- w = "cutoff";
- if (sd.has(w))
- {
- setCutoff( (F32)sd[w].asReal() );
- } else goto fail;
-
- return true;
+ const char *w;
+ w = "color";
+ if (sd.has(w))
+ {
+ setLinearColor( ll_color4_from_sd(sd["color"]) );
+ } else goto fail;
+ w = "radius";
+ if (sd.has(w))
+ {
+ setRadius( (F32)sd[w].asReal() );
+ } else goto fail;
+ w = "falloff";
+ if (sd.has(w))
+ {
+ setFalloff( (F32)sd[w].asReal() );
+ } else goto fail;
+ w = "cutoff";
+ if (sd.has(w))
+ {
+ setCutoff( (F32)sd[w].asReal() );
+ } else goto fail;
+
+ return true;
fail:
- return false;
+ return false;
}
//============================================================================
@@ -1832,39 +1832,39 @@ LLReflectionProbeParams::LLReflectionProbeParams()
BOOL LLReflectionProbeParams::pack(LLDataPacker &dp) const
{
- dp.packF32(mAmbiance, "ambiance");
+ dp.packF32(mAmbiance, "ambiance");
dp.packF32(mClipDistance, "clip_distance");
dp.packU8(mFlags, "flags");
- return TRUE;
+ return TRUE;
}
BOOL LLReflectionProbeParams::unpack(LLDataPacker &dp)
{
- F32 ambiance;
+ F32 ambiance;
F32 clip_distance;
- dp.unpackF32(ambiance, "ambiance");
- setAmbiance(ambiance);
-
+ dp.unpackF32(ambiance, "ambiance");
+ setAmbiance(ambiance);
+
dp.unpackF32(clip_distance, "clip_distance");
- setClipDistance(clip_distance);
-
+ setClipDistance(clip_distance);
+
dp.unpackU8(mFlags, "flags");
-
- return TRUE;
+
+ return TRUE;
}
bool LLReflectionProbeParams::operator==(const LLNetworkData& data) const
{
- if (data.mType != PARAMS_REFLECTION_PROBE)
- {
- return false;
- }
- const LLReflectionProbeParams *param = (const LLReflectionProbeParams*)&data;
- if (param->mAmbiance != mAmbiance)
- {
- return false;
- }
+ if (data.mType != PARAMS_REFLECTION_PROBE)
+ {
+ return false;
+ }
+ const LLReflectionProbeParams *param = (const LLReflectionProbeParams*)&data;
+ if (param->mAmbiance != mAmbiance)
+ {
+ return false;
+ }
if (param->mClipDistance != mClipDistance)
{
return false;
@@ -1873,25 +1873,25 @@ bool LLReflectionProbeParams::operator==(const LLNetworkData& data) const
{
return false;
}
- return true;
+ return true;
}
void LLReflectionProbeParams::copy(const LLNetworkData& data)
{
- const LLReflectionProbeParams *param = (LLReflectionProbeParams*)&data;
- mType = param->mType;
- mAmbiance = param->mAmbiance;
+ const LLReflectionProbeParams *param = (LLReflectionProbeParams*)&data;
+ mType = param->mType;
+ mAmbiance = param->mAmbiance;
mClipDistance = param->mClipDistance;
mFlags = param->mFlags;
}
LLSD LLReflectionProbeParams::asLLSD() const
{
- LLSD sd;
- sd["ambiance"] = getAmbiance();
+ LLSD sd;
+ sd["ambiance"] = getAmbiance();
sd["clip_distance"] = getClipDistance();
sd["flags"] = mFlags;
- return sd;
+ return sd;
}
bool LLReflectionProbeParams::fromLLSD(LLSD& sd)
@@ -1903,10 +1903,10 @@ bool LLReflectionProbeParams::fromLLSD(LLSD& sd)
return false;
}
- setAmbiance((F32)sd["ambiance"].asReal());
+ setAmbiance((F32)sd["ambiance"].asReal());
setClipDistance((F32)sd["clip_distance"].asReal());
mFlags = (U8) sd["flags"].asInteger();
-
+
return true;
}
@@ -1937,312 +1937,312 @@ void LLReflectionProbeParams::setIsDynamic(bool is_dynamic)
//============================================================================
LLFlexibleObjectData::LLFlexibleObjectData()
{
- mSimulateLOD = FLEXIBLE_OBJECT_DEFAULT_NUM_SECTIONS;
- mGravity = FLEXIBLE_OBJECT_DEFAULT_GRAVITY;
- mAirFriction = FLEXIBLE_OBJECT_DEFAULT_AIR_FRICTION;
- mWindSensitivity = FLEXIBLE_OBJECT_DEFAULT_WIND_SENSITIVITY;
- mTension = FLEXIBLE_OBJECT_DEFAULT_TENSION;
- //mUsingCollisionSphere = FLEXIBLE_OBJECT_DEFAULT_USING_COLLISION_SPHERE;
- //mRenderingCollisionSphere = FLEXIBLE_OBJECT_DEFAULT_RENDERING_COLLISION_SPHERE;
- mUserForce = LLVector3(0.f, 0.f, 0.f);
+ mSimulateLOD = FLEXIBLE_OBJECT_DEFAULT_NUM_SECTIONS;
+ mGravity = FLEXIBLE_OBJECT_DEFAULT_GRAVITY;
+ mAirFriction = FLEXIBLE_OBJECT_DEFAULT_AIR_FRICTION;
+ mWindSensitivity = FLEXIBLE_OBJECT_DEFAULT_WIND_SENSITIVITY;
+ mTension = FLEXIBLE_OBJECT_DEFAULT_TENSION;
+ //mUsingCollisionSphere = FLEXIBLE_OBJECT_DEFAULT_USING_COLLISION_SPHERE;
+ //mRenderingCollisionSphere = FLEXIBLE_OBJECT_DEFAULT_RENDERING_COLLISION_SPHERE;
+ mUserForce = LLVector3(0.f, 0.f, 0.f);
- mType = PARAMS_FLEXIBLE;
+ mType = PARAMS_FLEXIBLE;
}
BOOL LLFlexibleObjectData::pack(LLDataPacker &dp) const
{
- // Custom, uber-svelte pack "softness" in upper bits of tension & drag
- U8 bit1 = (mSimulateLOD & 2) << 6;
- U8 bit2 = (mSimulateLOD & 1) << 7;
- dp.packU8((U8)(mTension*10.01f) + bit1, "tension");
- dp.packU8((U8)(mAirFriction*10.01f) + bit2, "drag");
- dp.packU8((U8)((mGravity+10.f)*10.01f), "gravity");
- dp.packU8((U8)(mWindSensitivity*10.01f), "wind");
- dp.packVector3(mUserForce, "userforce");
- return TRUE;
+ // Custom, uber-svelte pack "softness" in upper bits of tension & drag
+ U8 bit1 = (mSimulateLOD & 2) << 6;
+ U8 bit2 = (mSimulateLOD & 1) << 7;
+ dp.packU8((U8)(mTension*10.01f) + bit1, "tension");
+ dp.packU8((U8)(mAirFriction*10.01f) + bit2, "drag");
+ dp.packU8((U8)((mGravity+10.f)*10.01f), "gravity");
+ dp.packU8((U8)(mWindSensitivity*10.01f), "wind");
+ dp.packVector3(mUserForce, "userforce");
+ return TRUE;
}
BOOL LLFlexibleObjectData::unpack(LLDataPacker &dp)
{
- U8 tension, friction, gravity, wind;
- U8 bit1, bit2;
- dp.unpackU8(tension, "tension"); bit1 = (tension >> 6) & 2;
- mTension = ((F32)(tension&0x7f))/10.f;
- dp.unpackU8(friction, "drag"); bit2 = (friction >> 7) & 1;
- mAirFriction = ((F32)(friction&0x7f))/10.f;
- mSimulateLOD = bit1 | bit2;
- dp.unpackU8(gravity, "gravity"); mGravity = ((F32)gravity)/10.f - 10.f;
- dp.unpackU8(wind, "wind"); mWindSensitivity = ((F32)wind)/10.f;
- if (dp.hasNext())
- {
- dp.unpackVector3(mUserForce, "userforce");
- }
- else
- {
- mUserForce.setVec(0.f, 0.f, 0.f);
- }
- return TRUE;
+ U8 tension, friction, gravity, wind;
+ U8 bit1, bit2;
+ dp.unpackU8(tension, "tension"); bit1 = (tension >> 6) & 2;
+ mTension = ((F32)(tension&0x7f))/10.f;
+ dp.unpackU8(friction, "drag"); bit2 = (friction >> 7) & 1;
+ mAirFriction = ((F32)(friction&0x7f))/10.f;
+ mSimulateLOD = bit1 | bit2;
+ dp.unpackU8(gravity, "gravity"); mGravity = ((F32)gravity)/10.f - 10.f;
+ dp.unpackU8(wind, "wind"); mWindSensitivity = ((F32)wind)/10.f;
+ if (dp.hasNext())
+ {
+ dp.unpackVector3(mUserForce, "userforce");
+ }
+ else
+ {
+ mUserForce.setVec(0.f, 0.f, 0.f);
+ }
+ return TRUE;
}
bool LLFlexibleObjectData::operator==(const LLNetworkData& data) const
{
- if (data.mType != PARAMS_FLEXIBLE)
- {
- return false;
- }
- LLFlexibleObjectData *flex_data = (LLFlexibleObjectData*)&data;
- return (mSimulateLOD == flex_data->mSimulateLOD &&
- mGravity == flex_data->mGravity &&
- mAirFriction == flex_data->mAirFriction &&
- mWindSensitivity == flex_data->mWindSensitivity &&
- mTension == flex_data->mTension &&
- mUserForce == flex_data->mUserForce);
- //mUsingCollisionSphere == flex_data->mUsingCollisionSphere &&
- //mRenderingCollisionSphere == flex_data->mRenderingCollisionSphere
+ if (data.mType != PARAMS_FLEXIBLE)
+ {
+ return false;
+ }
+ LLFlexibleObjectData *flex_data = (LLFlexibleObjectData*)&data;
+ return (mSimulateLOD == flex_data->mSimulateLOD &&
+ mGravity == flex_data->mGravity &&
+ mAirFriction == flex_data->mAirFriction &&
+ mWindSensitivity == flex_data->mWindSensitivity &&
+ mTension == flex_data->mTension &&
+ mUserForce == flex_data->mUserForce);
+ //mUsingCollisionSphere == flex_data->mUsingCollisionSphere &&
+ //mRenderingCollisionSphere == flex_data->mRenderingCollisionSphere
}
void LLFlexibleObjectData::copy(const LLNetworkData& data)
{
- const LLFlexibleObjectData *flex_data = (LLFlexibleObjectData*)&data;
- mSimulateLOD = flex_data->mSimulateLOD;
- mGravity = flex_data->mGravity;
- mAirFriction = flex_data->mAirFriction;
- mWindSensitivity = flex_data->mWindSensitivity;
- mTension = flex_data->mTension;
- mUserForce = flex_data->mUserForce;
- //mUsingCollisionSphere = flex_data->mUsingCollisionSphere;
- //mRenderingCollisionSphere = flex_data->mRenderingCollisionSphere;
+ const LLFlexibleObjectData *flex_data = (LLFlexibleObjectData*)&data;
+ mSimulateLOD = flex_data->mSimulateLOD;
+ mGravity = flex_data->mGravity;
+ mAirFriction = flex_data->mAirFriction;
+ mWindSensitivity = flex_data->mWindSensitivity;
+ mTension = flex_data->mTension;
+ mUserForce = flex_data->mUserForce;
+ //mUsingCollisionSphere = flex_data->mUsingCollisionSphere;
+ //mRenderingCollisionSphere = flex_data->mRenderingCollisionSphere;
}
LLSD LLFlexibleObjectData::asLLSD() const
{
- LLSD sd;
+ LLSD sd;
- sd["air_friction"] = getAirFriction();
- sd["gravity"] = getGravity();
- sd["simulate_lod"] = getSimulateLOD();
- sd["tension"] = getTension();
- sd["user_force"] = getUserForce().getValue();
- sd["wind_sensitivity"] = getWindSensitivity();
-
- return sd;
+ sd["air_friction"] = getAirFriction();
+ sd["gravity"] = getGravity();
+ sd["simulate_lod"] = getSimulateLOD();
+ sd["tension"] = getTension();
+ sd["user_force"] = getUserForce().getValue();
+ sd["wind_sensitivity"] = getWindSensitivity();
+
+ return sd;
}
bool LLFlexibleObjectData::fromLLSD(LLSD& sd)
{
- const char *w;
- w = "air_friction";
- if (sd.has(w))
- {
- setAirFriction( (F32)sd[w].asReal() );
- } else goto fail;
- w = "gravity";
- if (sd.has(w))
- {
- setGravity( (F32)sd[w].asReal() );
- } else goto fail;
- w = "simulate_lod";
- if (sd.has(w))
- {
- setSimulateLOD( sd[w].asInteger() );
- } else goto fail;
- w = "tension";
- if (sd.has(w))
- {
- setTension( (F32)sd[w].asReal() );
- } else goto fail;
- w = "user_force";
- if (sd.has(w))
- {
- LLVector3 user_force = ll_vector3_from_sd(sd[w], 0);
- setUserForce( user_force );
- } else goto fail;
- w = "wind_sensitivity";
- if (sd.has(w))
- {
- setWindSensitivity( (F32)sd[w].asReal() );
- } else goto fail;
-
- return true;
+ const char *w;
+ w = "air_friction";
+ if (sd.has(w))
+ {
+ setAirFriction( (F32)sd[w].asReal() );
+ } else goto fail;
+ w = "gravity";
+ if (sd.has(w))
+ {
+ setGravity( (F32)sd[w].asReal() );
+ } else goto fail;
+ w = "simulate_lod";
+ if (sd.has(w))
+ {
+ setSimulateLOD( sd[w].asInteger() );
+ } else goto fail;
+ w = "tension";
+ if (sd.has(w))
+ {
+ setTension( (F32)sd[w].asReal() );
+ } else goto fail;
+ w = "user_force";
+ if (sd.has(w))
+ {
+ LLVector3 user_force = ll_vector3_from_sd(sd[w], 0);
+ setUserForce( user_force );
+ } else goto fail;
+ w = "wind_sensitivity";
+ if (sd.has(w))
+ {
+ setWindSensitivity( (F32)sd[w].asReal() );
+ } else goto fail;
+
+ return true;
fail:
- return false;
+ return false;
}
//============================================================================
LLSculptParams::LLSculptParams()
{
- mType = PARAMS_SCULPT;
- mSculptTexture = SCULPT_DEFAULT_TEXTURE;
- mSculptType = LL_SCULPT_TYPE_SPHERE;
+ mType = PARAMS_SCULPT;
+ mSculptTexture = SCULPT_DEFAULT_TEXTURE;
+ mSculptType = LL_SCULPT_TYPE_SPHERE;
}
BOOL LLSculptParams::pack(LLDataPacker &dp) const
{
- dp.packUUID(mSculptTexture, "texture");
- dp.packU8(mSculptType, "type");
-
- return TRUE;
+ dp.packUUID(mSculptTexture, "texture");
+ dp.packU8(mSculptType, "type");
+
+ return TRUE;
}
BOOL LLSculptParams::unpack(LLDataPacker &dp)
{
- U8 type;
- LLUUID id;
- dp.unpackUUID(id, "texture");
- dp.unpackU8(type, "type");
+ U8 type;
+ LLUUID id;
+ dp.unpackUUID(id, "texture");
+ dp.unpackU8(type, "type");
- setSculptTexture(id, type);
- return TRUE;
+ setSculptTexture(id, type);
+ return TRUE;
}
bool LLSculptParams::operator==(const LLNetworkData& data) const
{
- if (data.mType != PARAMS_SCULPT)
- {
- return false;
- }
-
- const LLSculptParams *param = (const LLSculptParams*)&data;
- if ( (param->mSculptTexture != mSculptTexture) ||
- (param->mSculptType != mSculptType) )
-
- {
- return false;
- }
-
- return true;
+ if (data.mType != PARAMS_SCULPT)
+ {
+ return false;
+ }
+
+ const LLSculptParams *param = (const LLSculptParams*)&data;
+ if ( (param->mSculptTexture != mSculptTexture) ||
+ (param->mSculptType != mSculptType) )
+
+ {
+ return false;
+ }
+
+ return true;
}
void LLSculptParams::copy(const LLNetworkData& data)
{
- const LLSculptParams *param = (LLSculptParams*)&data;
- setSculptTexture(param->mSculptTexture, param->mSculptType);
+ const LLSculptParams *param = (LLSculptParams*)&data;
+ setSculptTexture(param->mSculptTexture, param->mSculptType);
}
LLSD LLSculptParams::asLLSD() const
{
- LLSD sd;
-
- sd["texture"] = mSculptTexture;
- sd["type"] = mSculptType;
-
- return sd;
+ LLSD sd;
+
+ sd["texture"] = mSculptTexture;
+ sd["type"] = mSculptType;
+
+ return sd;
}
bool LLSculptParams::fromLLSD(LLSD& sd)
{
- const char *w;
- U8 type;
- w = "type";
- if (sd.has(w))
- {
- type = sd[w].asInteger();
- }
- else return false;
+ const char *w;
+ U8 type;
+ w = "type";
+ if (sd.has(w))
+ {
+ type = sd[w].asInteger();
+ }
+ else return false;
- w = "texture";
- if (sd.has(w))
- {
- setSculptTexture(sd[w], type);
- }
- else return false;
+ w = "texture";
+ if (sd.has(w))
+ {
+ setSculptTexture(sd[w], type);
+ }
+ else return false;
- return true;
+ return true;
}
void LLSculptParams::setSculptTexture(const LLUUID& texture_id, U8 sculpt_type)
{
- U8 type = sculpt_type & LL_SCULPT_TYPE_MASK;
- U8 flags = sculpt_type & LL_SCULPT_FLAG_MASK;
- if (sculpt_type != (type | flags) || type > LL_SCULPT_TYPE_MAX)
- {
- mSculptTexture = SCULPT_DEFAULT_TEXTURE;
- mSculptType = LL_SCULPT_TYPE_SPHERE;
- }
- else
- {
- mSculptTexture = texture_id;
- mSculptType = sculpt_type;
- }
+ U8 type = sculpt_type & LL_SCULPT_TYPE_MASK;
+ U8 flags = sculpt_type & LL_SCULPT_FLAG_MASK;
+ if (sculpt_type != (type | flags) || type > LL_SCULPT_TYPE_MAX)
+ {
+ mSculptTexture = SCULPT_DEFAULT_TEXTURE;
+ mSculptType = LL_SCULPT_TYPE_SPHERE;
+ }
+ else
+ {
+ mSculptTexture = texture_id;
+ mSculptType = sculpt_type;
+ }
}
//============================================================================
LLLightImageParams::LLLightImageParams()
{
- mType = PARAMS_LIGHT_IMAGE;
- mParams.setVec(F_PI*0.5f, 0.f, 0.f);
+ mType = PARAMS_LIGHT_IMAGE;
+ mParams.setVec(F_PI*0.5f, 0.f, 0.f);
}
BOOL LLLightImageParams::pack(LLDataPacker &dp) const
{
- dp.packUUID(mLightTexture, "texture");
- dp.packVector3(mParams, "params");
+ dp.packUUID(mLightTexture, "texture");
+ dp.packVector3(mParams, "params");
- return TRUE;
+ return TRUE;
}
BOOL LLLightImageParams::unpack(LLDataPacker &dp)
{
- dp.unpackUUID(mLightTexture, "texture");
- dp.unpackVector3(mParams, "params");
-
- return TRUE;
+ dp.unpackUUID(mLightTexture, "texture");
+ dp.unpackVector3(mParams, "params");
+
+ return TRUE;
}
bool LLLightImageParams::operator==(const LLNetworkData& data) const
{
- if (data.mType != PARAMS_LIGHT_IMAGE)
- {
- return false;
- }
-
- const LLLightImageParams *param = (const LLLightImageParams*)&data;
- if ( (param->mLightTexture != mLightTexture) )
- {
- return false;
- }
-
- if ( (param->mParams != mParams ) )
- {
- return false;
- }
-
- return true;
+ if (data.mType != PARAMS_LIGHT_IMAGE)
+ {
+ return false;
+ }
+
+ const LLLightImageParams *param = (const LLLightImageParams*)&data;
+ if ( (param->mLightTexture != mLightTexture) )
+ {
+ return false;
+ }
+
+ if ( (param->mParams != mParams ) )
+ {
+ return false;
+ }
+
+ return true;
}
void LLLightImageParams::copy(const LLNetworkData& data)
{
- const LLLightImageParams *param = (LLLightImageParams*)&data;
- mLightTexture = param->mLightTexture;
- mParams = param->mParams;
+ const LLLightImageParams *param = (LLLightImageParams*)&data;
+ mLightTexture = param->mLightTexture;
+ mParams = param->mParams;
}
LLSD LLLightImageParams::asLLSD() const
{
- LLSD sd;
-
- sd["texture"] = mLightTexture;
- sd["params"] = mParams.getValue();
-
- return sd;
+ LLSD sd;
+
+ sd["texture"] = mLightTexture;
+ sd["params"] = mParams.getValue();
+
+ return sd;
}
bool LLLightImageParams::fromLLSD(LLSD& sd)
{
- if (sd.has("texture"))
- {
- setLightTexture( sd["texture"] );
- setParams( LLVector3( sd["params"] ) );
- return true;
- }
-
- return false;
+ if (sd.has("texture"))
+ {
+ setLightTexture( sd["texture"] );
+ setParams( LLVector3( sd["params"] ) );
+ return true;
+ }
+
+ return false;
}
//============================================================================
@@ -2250,63 +2250,63 @@ bool LLLightImageParams::fromLLSD(LLSD& sd)
LLExtendedMeshParams::LLExtendedMeshParams()
{
mType = PARAMS_EXTENDED_MESH;
- mFlags = 0;
+ mFlags = 0;
}
BOOL LLExtendedMeshParams::pack(LLDataPacker &dp) const
{
- dp.packU32(mFlags, "flags");
+ dp.packU32(mFlags, "flags");
- return TRUE;
+ return TRUE;
}
BOOL LLExtendedMeshParams::unpack(LLDataPacker &dp)
{
- dp.unpackU32(mFlags, "flags");
-
- return TRUE;
+ dp.unpackU32(mFlags, "flags");
+
+ return TRUE;
}
bool LLExtendedMeshParams::operator==(const LLNetworkData& data) const
{
- if (data.mType != PARAMS_EXTENDED_MESH)
- {
- return false;
- }
-
- const LLExtendedMeshParams *param = (const LLExtendedMeshParams*)&data;
- if ( (param->mFlags != mFlags) )
- {
- return false;
- }
+ if (data.mType != PARAMS_EXTENDED_MESH)
+ {
+ return false;
+ }
- return true;
+ const LLExtendedMeshParams *param = (const LLExtendedMeshParams*)&data;
+ if ( (param->mFlags != mFlags) )
+ {
+ return false;
+ }
+
+ return true;
}
void LLExtendedMeshParams::copy(const LLNetworkData& data)
{
- const LLExtendedMeshParams *param = (LLExtendedMeshParams*)&data;
- mFlags = param->mFlags;
+ const LLExtendedMeshParams *param = (LLExtendedMeshParams*)&data;
+ mFlags = param->mFlags;
}
LLSD LLExtendedMeshParams::asLLSD() const
{
- LLSD sd;
-
- sd["flags"] = LLSD::Integer(mFlags);
-
- return sd;
+ LLSD sd;
+
+ sd["flags"] = LLSD::Integer(mFlags);
+
+ return sd;
}
bool LLExtendedMeshParams::fromLLSD(LLSD& sd)
{
- if (sd.has("flags"))
- {
- setFlags( sd["flags"].asInteger());
- return true;
- }
-
- return false;
+ if (sd.has("flags"))
+ {
+ setFlags( sd["flags"].asInteger());
+ return true;
+ }
+
+ return false;
}
//============================================================================