phoenix-firestorm/indra/llcommon/llsdutil.cpp

281 lines
5.5 KiB
C++

/**
* @file llsdutil.cpp
* @author Phoenix
* @date 2006-05-24
* @brief Implementation of classes, functions, etc, for using structured data.
*
* Copyright (c) 2006-$CurrentYear$, Linden Research, Inc.
* $License$
*/
#include "linden_common.h"
#include "llsdutil.h"
#if LL_WINDOWS
# define WIN32_LEAN_AND_MEAN
# include <winsock2.h> // for htonl
#elif LL_LINUX || LL_SOLARIS
# include <netinet/in.h>
#elif LL_DARWIN
# include <arpa/inet.h>
#endif
#include "llsdserialize.h"
// vector3
LLSD ll_sd_from_vector3(const LLVector3& vec)
{
LLSD rv;
rv.append((F64)vec.mV[VX]);
rv.append((F64)vec.mV[VY]);
rv.append((F64)vec.mV[VZ]);
return rv;
}
LLVector3 ll_vector3_from_sd(const LLSD& sd, S32 start_index)
{
LLVector3 rv;
rv.mV[VX] = (F32)sd[start_index].asReal();
rv.mV[VY] = (F32)sd[++start_index].asReal();
rv.mV[VZ] = (F32)sd[++start_index].asReal();
return rv;
}
// vector4
LLSD ll_sd_from_vector4(const LLVector4& vec)
{
LLSD rv;
rv.append((F64)vec.mV[VX]);
rv.append((F64)vec.mV[VY]);
rv.append((F64)vec.mV[VZ]);
rv.append((F64)vec.mV[VW]);
return rv;
}
LLVector4 ll_vector4_from_sd(const LLSD& sd, S32 start_index)
{
LLVector4 rv;
rv.mV[VX] = (F32)sd[start_index].asReal();
rv.mV[VY] = (F32)sd[++start_index].asReal();
rv.mV[VZ] = (F32)sd[++start_index].asReal();
rv.mV[VW] = (F32)sd[++start_index].asReal();
return rv;
}
// vector3d
LLSD ll_sd_from_vector3d(const LLVector3d& vec)
{
LLSD rv;
rv.append(vec.mdV[VX]);
rv.append(vec.mdV[VY]);
rv.append(vec.mdV[VZ]);
return rv;
}
LLVector3d ll_vector3d_from_sd(const LLSD& sd, S32 start_index)
{
LLVector3d rv;
rv.mdV[VX] = sd[start_index].asReal();
rv.mdV[VY] = sd[++start_index].asReal();
rv.mdV[VZ] = sd[++start_index].asReal();
return rv;
}
//vector2
LLSD ll_sd_from_vector2(const LLVector2& vec)
{
LLSD rv;
rv.append((F64)vec.mV[VX]);
rv.append((F64)vec.mV[VY]);
return rv;
}
LLVector2 ll_vector2_from_sd(const LLSD& sd)
{
LLVector2 rv;
rv.mV[VX] = (F32)sd[0].asReal();
rv.mV[VY] = (F32)sd[1].asReal();
return rv;
}
// Quaternion
LLSD ll_sd_from_quaternion(const LLQuaternion& quat)
{
LLSD rv;
rv.append((F64)quat.mQ[VX]);
rv.append((F64)quat.mQ[VY]);
rv.append((F64)quat.mQ[VZ]);
rv.append((F64)quat.mQ[VW]);
return rv;
}
LLQuaternion ll_quaternion_from_sd(const LLSD& sd)
{
LLQuaternion quat;
quat.mQ[VX] = (F32)sd[0].asReal();
quat.mQ[VY] = (F32)sd[1].asReal();
quat.mQ[VZ] = (F32)sd[2].asReal();
quat.mQ[VW] = (F32)sd[3].asReal();
return quat;
}
// color4
LLSD ll_sd_from_color4(const LLColor4& c)
{
LLSD rv;
rv.append(c.mV[0]);
rv.append(c.mV[1]);
rv.append(c.mV[2]);
rv.append(c.mV[3]);
return rv;
}
LLColor4 ll_color4_from_sd(const LLSD& sd)
{
LLColor4 c;
c.mV[0] = (F32)sd[0].asReal();
c.mV[1] = (F32)sd[1].asReal();
c.mV[2] = (F32)sd[2].asReal();
c.mV[3] = (F32)sd[3].asReal();
return c;
}
// U32
LLSD ll_sd_from_U32(const U32 val)
{
std::vector<U8> v;
U32 net_order = htonl(val);
v.resize(4);
memcpy(&(v[0]), &net_order, 4); /* Flawfinder: ignore */
return LLSD(v);
}
U32 ll_U32_from_sd(const LLSD& sd)
{
U32 ret;
std::vector<U8> v = sd.asBinary();
if (v.size() < 4)
{
return 0;
}
memcpy(&ret, &(v[0]), 4); /* Flawfinder: ignore */
ret = ntohl(ret);
return ret;
}
//U64
LLSD ll_sd_from_U64(const U64 val)
{
std::vector<U8> v;
U32 high, low;
high = (U32)(val >> 32);
low = (U32)val;
high = htonl(high);
low = htonl(low);
v.resize(8);
memcpy(&(v[0]), &high, 4); /* Flawfinder: ignore */
memcpy(&(v[4]), &low, 4); /* Flawfinder: ignore */
return LLSD(v);
}
U64 ll_U64_from_sd(const LLSD& sd)
{
U32 high, low;
std::vector<U8> v = sd.asBinary();
if (v.size() < 8)
{
return 0;
}
memcpy(&high, &(v[0]), 4); /* Flawfinder: ignore */
memcpy(&low, &(v[4]), 4); /* Flawfinder: ignore */
high = ntohl(high);
low = ntohl(low);
return ((U64)high) << 32 | low;
}
// IP Address (stored in net order in a U32, so don't need swizzling)
LLSD ll_sd_from_ipaddr(const U32 val)
{
std::vector<U8> v;
v.resize(4);
memcpy(&(v[0]), &val, 4); /* Flawfinder: ignore */
return LLSD(v);
}
U32 ll_ipaddr_from_sd(const LLSD& sd)
{
U32 ret;
std::vector<U8> v = sd.asBinary();
if (v.size() < 4)
{
return 0;
}
memcpy(&ret, &(v[0]), 4); /* Flawfinder: ignore */
return ret;
}
// Converts an LLSD binary to an LLSD string
LLSD ll_string_from_binary(const LLSD& sd)
{
std::vector<U8> value = sd.asBinary();
std::string str;
str.resize(value.size());
memcpy(&str[0], &value[0], value.size());
return str;
}
// Converts an LLSD string to an LLSD binary
LLSD ll_binary_from_string(const LLSD& sd)
{
std::vector<U8> binary_value;
LLString string_value = sd.asString();
const char* string_p = string_value.c_str();
while (*string_p)
{
binary_value.push_back(*string_p);
string_p++;
}
binary_value.push_back('\0');
return binary_value;
}
char* ll_print_sd(const LLSD& sd)
{
const U32 bufferSize = 10 * 1024;
static char buffer[bufferSize];
std::ostringstream stream;
//stream.rdbuf()->pubsetbuf(buffer, bufferSize);
stream << LLSDOStreamer<LLSDXMLFormatter>(sd);
stream << std::ends;
strncpy(buffer, stream.str().c_str(), bufferSize);
buffer[bufferSize - 1] = '\0';
return buffer;
}
char* ll_pretty_print_sd(const LLSD& sd)
{
const U32 bufferSize = 10 * 1024;
static char buffer[bufferSize];
std::ostringstream stream;
//stream.rdbuf()->pubsetbuf(buffer, bufferSize);
stream << LLSDOStreamer<LLSDXMLFormatter>(sd, LLSDFormatter::OPTIONS_PRETTY);
stream << std::ends;
strncpy(buffer, stream.str().c_str(), bufferSize);
buffer[bufferSize - 1] = '\0';
return buffer;
}