Files
SingularityViewer/indra/lscript/lscript_byteconvert.h
Lirusaito 6e3f404a1c Rip out old workarounds, hacks and macros for newer C++ features not being supported back in the day.
Adds LL_COMPILE_TIME_MESSAGE support to Linux.

llfinite -> std::isfinite
llisnan -> std::isnan
vector_shrink_to_fit -> vector.shrink_to_fit
2016-02-14 17:37:10 -05:00

1177 lines
28 KiB
C++

/**
* @file lscript_byteconvert.h
* @brief Shared code for compiler and assembler for LSL
*
* $LicenseInfo:firstyear=2002&license=viewergpl$
*
* Copyright (c) 2002-2009, Linden Research, Inc.
*
* Second Life Viewer Source Code
* The source code in this file ("Source Code") is provided by Linden Lab
* to you under the terms of the GNU General Public License, version 2.0
* ("GPL"), unless you have obtained a separate licensing agreement
* ("Other License"), formally executed by you and Linden Lab. Terms of
* the GPL can be found in doc/GPL-license.txt in this distribution, or
* online at http://secondlifegrid.net/programs/open_source/licensing/gplv2
*
* There are special exceptions to the terms and conditions of the GPL as
* it is applied to this Source Code. View the full text of the exception
* in the file doc/FLOSS-exception.txt in this software distribution, or
* online at
* http://secondlifegrid.net/programs/open_source/licensing/flossexception
*
* By copying, modifying or distributing this software, you acknowledge
* that you have read and understood your obligations described above,
* and agree to abide by those obligations.
*
* ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO
* WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY,
* COMPLETENESS OR PERFORMANCE.
* $/LicenseInfo$
*/
// data shared between compiler/assembler
// used to convert data between byte stream and outside data types
#ifndef LL_LSCRIPT_BYTECONVERT_H
#define LL_LSCRIPT_BYTECONVERT_H
#include "stdtypes.h"
#include "v3math.h"
#include "llquaternion.h"
#include "lscript_byteformat.h"
#include "lluuid.h"
void reset_hp_to_safe_spot(const U8 *buffer);
// remember that LScript byte stream is BigEndian
void set_fault(const U8 *stream, LSCRIPTRunTimeFaults fault);
inline S32 bytestream2integer(const U8 *stream, S32 &offset)
{
stream += offset;
offset += 4;
return (*stream<<24) | (*(stream + 1)<<16) | (*(stream + 2)<<8) | *(stream + 3);
}
inline U32 bytestream2unsigned_integer(const U8 *stream, S32 &offset)
{
stream += offset;
offset += 4;
return (*stream<<24) | (*(stream + 1)<<16) | (*(stream + 2)<<8) | *(stream + 3);
}
inline U64 bytestream2u64(const U8 *stream, S32 &offset)
{
stream += offset;
offset += 8;
return ((U64)(*stream)<<56)| ((U64)(*(stream + 1))<<48) | ((U64)(*(stream + 2))<<40) | ((U64)(*(stream + 3))<<32) |
((U64)(*(stream + 4))<<24) | ((U64)(*(stream + 5))<<16) | ((U64)(*(stream + 6))<<8) | (U64)(*(stream + 7));
}
inline void integer2bytestream(U8 *stream, S32 &offset, S32 integer)
{
stream += offset;
offset += 4;
*(stream) = (integer >> 24);
*(stream + 1) = (integer >> 16) & 0xff;
*(stream + 2) = (integer >> 8) & 0xff;
*(stream + 3) = (integer) & 0xff;
}
inline void unsigned_integer2bytestream(U8 *stream, S32 &offset, U32 integer)
{
stream += offset;
offset += 4;
*(stream) = (integer >> 24);
*(stream + 1) = (integer >> 16) & 0xff;
*(stream + 2) = (integer >> 8) & 0xff;
*(stream + 3) = (integer) & 0xff;
}
inline void u642bytestream(U8 *stream, S32 &offset, U64 integer)
{
stream += offset;
offset += 8;
*(stream) = (U8)(integer >> 56);
*(stream + 1) = (U8)((integer >> 48) & 0xff);
*(stream + 2) = (U8)((integer >> 40) & 0xff);
*(stream + 3) = (U8)((integer >> 32) & 0xff);
*(stream + 4) = (U8)((integer >> 24) & 0xff);
*(stream + 5) = (U8)((integer >> 16) & 0xff);
*(stream + 6) = (U8)((integer >> 8) & 0xff);
*(stream + 7) = (U8)((integer) & 0xff);
}
inline S16 bytestream2s16(const U8 *stream, S32 &offset)
{
stream += offset;
offset += 2;
return (*stream<<8) | *(stream + 1);
}
inline void s162bytestream(U8 *stream, S32 &offset, S16 integer)
{
stream += offset;
offset += 2;
*(stream) = (integer >> 8);
*(stream + 1) = (integer) & 0xff;
}
inline U16 bytestream2u16(const U8 *stream, S32 &offset)
{
stream += offset;
offset += 2;
return (*stream<<8) | *(stream + 1);
}
inline void u162bytestream(U8 *stream, S32 &offset, U16 integer)
{
stream += offset;
offset += 2;
*(stream) = (integer >> 8);
*(stream + 1) = (integer) & 0xff;
}
inline F32 bytestream2float(const U8 *stream, S32 &offset)
{
S32 value = bytestream2integer(stream, offset);
F32 fpvalue = *(F32 *)&value;
if (!std::isfinite(fpvalue))
{
fpvalue = 0;
set_fault(stream, LSRF_MATH);
}
return fpvalue;
}
inline void float2bytestream(U8 *stream, S32 &offset, F32 floatingpoint)
{
S32 value = *(S32 *)&floatingpoint;
integer2bytestream(stream, offset, value);
}
inline void bytestream_int2float(U8 *stream, S32 &offset)
{
S32 value = bytestream2integer(stream, offset);
offset -= 4;
F32 fpvalue = (F32)value;
if (!std::isfinite(fpvalue))
{
fpvalue = 0;
set_fault(stream, LSRF_MATH);
}
float2bytestream(stream, offset, fpvalue);
}
// Returns true on success, return false and clip copy on buffer overflow
inline bool bytestream2char(char *buffer, const U8 *stream, S32 &offset, S32 buffsize)
{
S32 source_len = strlen( (const char *)stream+offset );
S32 copy_len = buffsize - 1;
if( copy_len > source_len )
{
copy_len = source_len;
}
// strncpy without \0 padding overhead
memcpy( buffer, stream+offset, copy_len );
buffer[copy_len] = 0;
offset += source_len + 1; // advance past source string, include terminating '\0'
return source_len < buffsize;
}
inline void char2bytestream(U8 *stream, S32 &offset, const char *buffer)
{
while ((*(stream + offset++) = *buffer++))
;
}
inline U8 bytestream2byte(const U8 *stream, S32 &offset)
{
return *(stream + offset++);
}
inline void byte2bytestream(U8 *stream, S32 &offset, U8 byte)
{
*(stream + offset++) = byte;
}
inline void bytestream2bytestream(U8 *dest, S32 &dest_offset, const U8 *src, S32 &src_offset, S32 count)
{
while (count)
{
(*(dest + dest_offset++)) = (*(src + src_offset++));
count--;
}
}
inline void uuid2bytestream(U8 *stream, S32 &offset, const LLUUID &uuid)
{
S32 i;
for (i = 0; i < UUID_BYTES; i++)
{
*(stream + offset++) = uuid.mData[i];
}
}
inline void bytestream2uuid(U8 *stream, S32 &offset, LLUUID &uuid)
{
S32 i;
for (i = 0; i < UUID_BYTES; i++)
{
uuid.mData[i] = *(stream + offset++);
}
}
// vectors and quaternions and encoded in backwards order to match the way in which they are stored on the stack
inline void bytestream2vector(LLVector3 &vector, const U8 *stream, S32 &offset)
{
S32 value = bytestream2integer(stream, offset);
vector.mV[VZ] = *(F32 *)&value;
if (!std::isfinite(vector.mV[VZ]))
{
vector.mV[VZ] = 0;
set_fault(stream, LSRF_MATH);
}
value = bytestream2integer(stream, offset);
vector.mV[VY] = *(F32 *)&value;
if (!std::isfinite(vector.mV[VY]))
{
vector.mV[VY] = 0;
set_fault(stream, LSRF_MATH);
}
value = bytestream2integer(stream, offset);
vector.mV[VX] = *(F32 *)&value;
if (!std::isfinite(vector.mV[VX]))
{
vector.mV[VX] = 0;
set_fault(stream, LSRF_MATH);
}
}
inline void vector2bytestream(U8 *stream, S32 &offset, const LLVector3 &vector)
{
S32 value = *(S32 *)&vector.mV[VZ];
integer2bytestream(stream, offset, value);
value = *(S32 *)&vector.mV[VY];
integer2bytestream(stream, offset, value);
value = *(S32 *)&vector.mV[VX];
integer2bytestream(stream, offset, value);
}
inline void bytestream2quaternion(LLQuaternion &quat, const U8 *stream, S32 &offset)
{
S32 value = bytestream2integer(stream, offset);
quat.mQ[VS] = *(F32 *)&value;
if (!std::isfinite(quat.mQ[VS]))
{
quat.mQ[VS] = 0;
set_fault(stream, LSRF_MATH);
}
value = bytestream2integer(stream, offset);
quat.mQ[VZ] = *(F32 *)&value;
if (!std::isfinite(quat.mQ[VZ]))
{
quat.mQ[VZ] = 0;
set_fault(stream, LSRF_MATH);
}
value = bytestream2integer(stream, offset);
quat.mQ[VY] = *(F32 *)&value;
if (!std::isfinite(quat.mQ[VY]))
{
quat.mQ[VY] = 0;
set_fault(stream, LSRF_MATH);
}
value = bytestream2integer(stream, offset);
quat.mQ[VX] = *(F32 *)&value;
if (!std::isfinite(quat.mQ[VX]))
{
quat.mQ[VX] = 0;
set_fault(stream, LSRF_MATH);
}
}
inline void quaternion2bytestream(U8 *stream, S32 &offset, const LLQuaternion &quat)
{
S32 value = *(S32 *)&quat.mQ[VS];
integer2bytestream(stream, offset, value);
value = *(S32 *)&quat.mQ[VZ];
integer2bytestream(stream, offset, value);
value = *(S32 *)&quat.mQ[VY];
integer2bytestream(stream, offset, value);
value = *(S32 *)&quat.mQ[VX];
integer2bytestream(stream, offset, value);
}
inline S32 get_register(const U8 *stream, LSCRIPTRegisters reg)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
return bytestream2integer(stream, offset);
}
inline F32 get_register_fp(U8 *stream, LSCRIPTRegisters reg)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
F32 value = bytestream2float(stream, offset);
if (!std::isfinite(value))
{
value = 0;
set_fault(stream, LSRF_MATH);
}
return value;
}
inline U64 get_register_u64(U8 *stream, LSCRIPTRegisters reg)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
return bytestream2u64(stream, offset);
}
inline U64 get_event_register(U8 *stream, LSCRIPTRegisters reg, S32 major_version)
{
if (major_version == 1)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
return (U64)bytestream2integer(stream, offset);
}
else if (major_version == 2)
{
S32 offset = gLSCRIPTRegisterAddresses[reg + (LREG_NCE - LREG_CE)];
return bytestream2u64(stream, offset);
}
else
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
return (U64)bytestream2integer(stream, offset);
}
}
inline void set_register(U8 *stream, LSCRIPTRegisters reg, S32 value)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
integer2bytestream(stream, offset, value);
}
inline void set_register_fp(U8 *stream, LSCRIPTRegisters reg, F32 value)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
float2bytestream(stream, offset, value);
}
inline void set_register_u64(U8 *stream, LSCRIPTRegisters reg, U64 value)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
u642bytestream(stream, offset, value);
}
inline void set_event_register(U8 *stream, LSCRIPTRegisters reg, U64 value, S32 major_version)
{
if (major_version == 1)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
integer2bytestream(stream, offset, (S32)value);
}
else if (major_version == 2)
{
S32 offset = gLSCRIPTRegisterAddresses[reg + (LREG_NCE - LREG_CE)];
u642bytestream(stream, offset, value);
}
else
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
integer2bytestream(stream, offset, (S32)value);
}
}
inline F32 add_register_fp(U8 *stream, LSCRIPTRegisters reg, F32 value)
{
S32 offset = gLSCRIPTRegisterAddresses[reg];
F32 newvalue = bytestream2float(stream, offset);
newvalue += value;
if (!std::isfinite(newvalue))
{
newvalue = 0;
set_fault(stream, LSRF_MATH);
}
offset = gLSCRIPTRegisterAddresses[reg];
float2bytestream(stream, offset, newvalue);
return newvalue;
}
void lsa_print_heap(U8 *buffer);
inline void set_fault(const U8 *stream, LSCRIPTRunTimeFaults fault)
{
S32 fr = get_register(stream, LREG_FR);
// record the first error
if (!fr)
{
if ( (fault == LSRF_HEAP_ERROR)
||(fault == LSRF_STACK_HEAP_COLLISION)
||(fault == LSRF_BOUND_CHECK_ERROR))
{
reset_hp_to_safe_spot(stream);
// lsa_print_heap((U8 *)stream);
}
fr = fault;
set_register((U8 *)stream, LREG_FR, fr);
}
}
inline BOOL set_ip(U8 *stream, S32 ip)
{
// Verify that the Instruction Pointer is in a valid
// code area (between the Global Function Register
// and Heap Register).
S32 gfr = get_register(stream, LREG_GFR);
if (ip == 0)
{
set_register(stream, LREG_IP, ip);
return TRUE;
}
if (ip < gfr)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
S32 hr = get_register(stream, LREG_HR);
if (ip >= hr)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
set_register(stream, LREG_IP, ip);
return TRUE;
}
inline BOOL set_bp(U8 *stream, S32 bp)
{
// Verify that the Base Pointer is in a valid
// data area (between the Heap Pointer and
// the Top of Memory, and below the
// Stack Pointer).
S32 hp = get_register(stream, LREG_HP);
if (bp <= hp)
{
set_fault(stream, LSRF_STACK_HEAP_COLLISION);
return FALSE;
}
S32 tm = get_register(stream, LREG_TM);
if (bp >= tm)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
S32 sp = get_register(stream, LREG_SP);
if (bp < sp)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
set_register(stream, LREG_BP, bp);
return TRUE;
}
inline BOOL set_sp(U8 *stream, S32 sp)
{
// Verify that the Stack Pointer is in a valid
// data area (between the Heap Pointer and
// the Top of Memory).
S32 hp = get_register(stream, LREG_HP);
if (sp <= hp)
{
set_fault(stream, LSRF_STACK_HEAP_COLLISION);
return FALSE;
}
S32 tm = get_register(stream, LREG_TM);
if (sp >= tm)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
set_register(stream, LREG_SP, sp);
return TRUE;
}
inline void lscript_push(U8 *stream, U8 value)
{
S32 sp = get_register(stream, LREG_SP);
sp -= 1;
if (set_sp(stream, sp))
{
*(stream + sp) = value;
}
}
inline void lscript_push(U8 *stream, S32 value)
{
S32 sp = get_register(stream, LREG_SP);
sp -= LSCRIPTDataSize[LST_INTEGER];
if (set_sp(stream, sp))
{
integer2bytestream(stream, sp, value);
}
}
inline void lscript_push(U8 *stream, F32 value)
{
S32 sp = get_register(stream, LREG_SP);
sp -= LSCRIPTDataSize[LST_FLOATINGPOINT];
if (set_sp(stream, sp))
{
float2bytestream(stream, sp, value);
}
}
inline void lscript_push(U8 *stream, const LLVector3 &value)
{
S32 sp = get_register(stream, LREG_SP);
sp -= LSCRIPTDataSize[LST_VECTOR];
if (set_sp(stream, sp))
{
vector2bytestream(stream, sp, value);
}
}
inline void lscript_push(U8 *stream, const LLQuaternion &value)
{
S32 sp = get_register(stream, LREG_SP);
sp -= LSCRIPTDataSize[LST_QUATERNION];
if (set_sp(stream, sp))
{
quaternion2bytestream(stream, sp, value);
}
}
inline void lscript_pusharg(U8 *stream, S32 arg)
{
S32 sp = get_register(stream, LREG_SP);
sp -= arg;
set_sp(stream, sp);
}
inline void lscript_poparg(U8 *stream, S32 arg)
{
S32 sp = get_register(stream, LREG_SP);
sp += arg;
set_sp(stream, sp);
}
inline U8 lscript_pop_char(U8 *stream)
{
S32 sp = get_register(stream, LREG_SP);
U8 value = *(stream + sp++);
set_sp(stream, sp);
return value;
}
inline S32 lscript_pop_int(U8 *stream)
{
S32 sp = get_register(stream, LREG_SP);
S32 value = bytestream2integer(stream, sp);
set_sp(stream, sp);
return value;
}
inline F32 lscript_pop_float(U8 *stream)
{
S32 sp = get_register(stream, LREG_SP);
F32 value = bytestream2float(stream, sp);
if (!std::isfinite(value))
{
value = 0;
set_fault(stream, LSRF_MATH);
}
set_sp(stream, sp);
return value;
}
inline void lscript_pop_vector(U8 *stream, LLVector3 &value)
{
S32 sp = get_register(stream, LREG_SP);
bytestream2vector(value, stream, sp);
set_sp(stream, sp);
}
inline void lscript_pop_quaternion(U8 *stream, LLQuaternion &value)
{
S32 sp = get_register(stream, LREG_SP);
bytestream2quaternion(value, stream, sp);
set_sp(stream, sp);
}
inline void lscript_pusharge(U8 *stream, S32 value)
{
S32 sp = get_register(stream, LREG_SP);
sp -= value;
if (set_sp(stream, sp))
{
S32 i;
for (i = 0; i < value; i++)
{
*(stream + sp++) = 0;
}
}
}
inline BOOL lscript_check_local(U8 *stream, S32 &address, S32 size)
{
S32 sp = get_register(stream, LREG_SP);
S32 bp = get_register(stream, LREG_BP);
address += size;
address = bp - address;
if (address < sp - size)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
S32 tm = get_register(stream, LREG_TM);
if (address + size > tm)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
return TRUE;
}
inline BOOL lscript_check_global(U8 *stream, S32 &address, S32 size)
{
S32 gvr = get_register(stream, LREG_GVR);
// Possibility of overwriting registers? -- DK 09/07/04
if (address < 0)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
address += gvr;
S32 gfr = get_register(stream, LREG_GFR);
if (address + size > gfr)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
return TRUE;
}
inline void lscript_local_store(U8 *stream, S32 address, S32 value)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_INTEGER]))
integer2bytestream(stream, address, value);
}
inline void lscript_local_store(U8 *stream, S32 address, F32 value)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_FLOATINGPOINT]))
float2bytestream(stream, address, value);
}
inline void lscript_local_store(U8 *stream, S32 address, const LLVector3 value)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_VECTOR]))
vector2bytestream(stream, address, value);
}
inline void lscript_local_store(U8 *stream, S32 address, const LLQuaternion value)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_QUATERNION]))
quaternion2bytestream(stream, address, value);
}
inline void lscript_global_store(U8 *stream, S32 address, S32 value)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_INTEGER]))
integer2bytestream(stream, address, value);
}
inline void lscript_global_store(U8 *stream, S32 address, F32 value)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_FLOATINGPOINT]))
float2bytestream(stream, address, value);
}
inline void lscript_global_store(U8 *stream, S32 address, const LLVector3 value)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_VECTOR]))
vector2bytestream(stream, address, value);
}
inline void lscript_global_store(U8 *stream, S32 address, const LLQuaternion value)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_QUATERNION]))
quaternion2bytestream(stream, address, value);
}
inline S32 lscript_local_get(U8 *stream, S32 address)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_INTEGER]))
return bytestream2integer(stream, address);
return 0;
}
inline void lscript_local_get(U8 *stream, S32 address, F32 &value)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_FLOATINGPOINT]))
value = bytestream2float(stream, address);
if (!std::isfinite(value))
{
value = 0;
set_fault(stream, LSRF_MATH);
}
}
inline void lscript_local_get(U8 *stream, S32 address, LLVector3 &value)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_VECTOR]))
bytestream2vector(value, stream, address);
}
inline void lscript_local_get(U8 *stream, S32 address, LLQuaternion &value)
{
if (lscript_check_local(stream, address, LSCRIPTDataSize[LST_QUATERNION]))
bytestream2quaternion(value, stream, address);
}
inline S32 lscript_global_get(U8 *stream, S32 address)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_INTEGER]))
return bytestream2integer(stream, address);
return 0;
}
inline void lscript_global_get(U8 *stream, S32 address, F32 &value)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_FLOATINGPOINT]))
value = bytestream2float(stream, address);
if (!std::isfinite(value))
{
value = 0;
set_fault(stream, LSRF_MATH);
}
}
inline void lscript_global_get(U8 *stream, S32 address, LLVector3 &value)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_VECTOR]))
bytestream2vector(value, stream, address);
}
inline void lscript_global_get(U8 *stream, S32 address, LLQuaternion &value)
{
if (lscript_check_global(stream, address, LSCRIPTDataSize[LST_QUATERNION]))
bytestream2quaternion(value, stream, address);
}
inline S32 get_state_event_opcoode_start(U8 *stream, S32 state, LSCRIPTStateEventType event)
{
// get the start of the state table
S32 sr = get_register(stream, LREG_SR);
// get the position of the jump to the desired state
S32 value = get_register(stream, LREG_VN);
S32 state_offset_offset = 0;
S32 major_version = 0;
if (value == LSL2_VERSION1_END_NUMBER)
{
major_version = LSL2_MAJOR_VERSION_ONE;
state_offset_offset = sr + LSCRIPTDataSize[LST_INTEGER] + LSCRIPTDataSize[LST_INTEGER]*2*state;
}
else if (value == LSL2_VERSION_NUMBER)
{
major_version = LSL2_MAJOR_VERSION_TWO;
state_offset_offset = sr + LSCRIPTDataSize[LST_INTEGER] + LSCRIPTDataSize[LST_INTEGER]*3*state;
}
if ( state_offset_offset < 0 || state_offset_offset > TOP_OF_MEMORY )
{
return -1;
}
// get the actual position in memory of the desired state
S32 state_offset = sr + bytestream2integer(stream, state_offset_offset);
if ( state_offset < 0 || state_offset > TOP_OF_MEMORY )
{
return -1;
}
// save that value
S32 state_offset_base = state_offset;
// jump past the state name
S32 event_jump_offset = state_offset_base + bytestream2integer(stream, state_offset);
// get the location of the event offset
S32 event_offset = event_jump_offset + LSCRIPTDataSize[LST_INTEGER]*2*get_event_handler_jump_position(get_event_register(stream, LREG_ER, major_version), event);
if ( event_offset < 0 || event_offset > TOP_OF_MEMORY )
{
return -1;
}
// now, jump to the event
S32 event_start = bytestream2integer(stream, event_offset);
if ( event_start < 0 || event_start > TOP_OF_MEMORY )
{
return -1;
}
event_start += event_jump_offset;
S32 event_start_original = event_start;
// now skip past the parameters
S32 opcode_offset = bytestream2integer(stream, event_start);
if ( opcode_offset < 0 || opcode_offset > TOP_OF_MEMORY )
{
return -1;
}
return opcode_offset + event_start_original;
}
inline U64 get_handled_events(U8 *stream, S32 state)
{
U64 retvalue = 0;
// get the start of the state table
S32 sr = get_register(stream, LREG_SR);
// get the position of the jump to the desired state
S32 value = get_register(stream, LREG_VN);
S32 state_handled_offset = 0;
if (value == LSL2_VERSION1_END_NUMBER)
{
state_handled_offset = sr + LSCRIPTDataSize[LST_INTEGER]*2*state + 2*LSCRIPTDataSize[LST_INTEGER];
retvalue = bytestream2integer(stream, state_handled_offset);
}
else if (value == LSL2_VERSION_NUMBER)
{
state_handled_offset = sr + LSCRIPTDataSize[LST_INTEGER]*3*state + 2*LSCRIPTDataSize[LST_INTEGER];
retvalue = bytestream2u64(stream, state_handled_offset);
}
// get the handled events
return retvalue;
}
// Returns -1 on error
inline S32 get_event_stack_size(U8 *stream, S32 state, LSCRIPTStateEventType event)
{
// get the start of the state table
S32 sr = get_register(stream, LREG_SR);
// get state offset
S32 value = get_register(stream, LREG_VN);
S32 state_offset_offset = 0;
S32 major_version = 0;
if (value == LSL2_VERSION1_END_NUMBER)
{
major_version = LSL2_MAJOR_VERSION_ONE;
state_offset_offset = sr + LSCRIPTDataSize[LST_INTEGER] + LSCRIPTDataSize[LST_INTEGER]*2*state;
}
else if (value == LSL2_VERSION_NUMBER)
{
major_version = LSL2_MAJOR_VERSION_TWO;
state_offset_offset = sr + LSCRIPTDataSize[LST_INTEGER] + LSCRIPTDataSize[LST_INTEGER]*3*state;
}
if ( state_offset_offset < 0 || state_offset_offset > TOP_OF_MEMORY )
{
return -1;
}
S32 state_offset = bytestream2integer(stream, state_offset_offset);
state_offset += sr;
state_offset_offset = state_offset;
if ( state_offset_offset < 0 || state_offset_offset > TOP_OF_MEMORY )
{
return -1;
}
// skip to jump table
S32 jump_table = bytestream2integer(stream, state_offset_offset);
jump_table += state_offset;
if ( jump_table < 0 || jump_table > TOP_OF_MEMORY )
{
return -1;
}
// get the position of the jump to the desired state
S32 stack_size_offset = jump_table + LSCRIPTDataSize[LST_INTEGER]*2*get_event_handler_jump_position(get_event_register(stream, LREG_ER, major_version), event) + LSCRIPTDataSize[LST_INTEGER];
// get the handled events
S32 stack_size = bytestream2integer(stream, stack_size_offset);
if ( stack_size < 0 || stack_size > TOP_OF_MEMORY )
{
return -1;
}
return stack_size;
}
inline LSCRIPTStateEventType return_first_event(S32 event)
{
S32 count = 1;
while (count < LSTT_EOF)
{
if (event & 0x1)
{
return (LSCRIPTStateEventType) count;
}
else
{
event >>= 1;
count++;
}
}
return LSTT_NULL;
}
// the safe instruction versions of these commands will only work if offset is between
// GFR and HR, meaning that it is an instruction (more or less) in global functions or event handlers
inline BOOL safe_instruction_check_address(U8 *stream, S32 offset, S32 size)
{
S32 gfr = get_register(stream, LREG_GFR);
if (offset < gfr)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
else
{
S32 hr = get_register(stream, LREG_HR);
if (offset + size > hr)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
else
{
return TRUE;
}
}
}
inline BOOL safe_heap_check_address(U8 *stream, S32 offset, S32 size)
{
S32 hr = get_register(stream, LREG_HR);
if (offset < hr)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
else
{
S32 hp = get_register(stream, LREG_HP);
if (offset + size > hp)
{
set_fault(stream, LSRF_BOUND_CHECK_ERROR);
return FALSE;
}
else
{
return TRUE;
}
}
}
inline U8 safe_instruction_bytestream2byte(U8 *stream, S32 &offset)
{
if (safe_instruction_check_address(stream, offset, 1))
{
return *(stream + offset++);
}
else
{
return 0;
}
}
inline void safe_instruction_byte2bytestream(U8 *stream, S32 &offset, U8 byte)
{
if (safe_instruction_check_address(stream, offset, 1))
{
*(stream + offset++) = byte;
}
}
inline S32 safe_instruction_bytestream2integer(U8 *stream, S32 &offset)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_INTEGER]))
{
return (bytestream2integer(stream, offset));
}
else
{
return 0;
}
}
inline void safe_instruction_integer2bytestream(U8 *stream, S32 &offset, S32 value)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_INTEGER]))
{
integer2bytestream(stream, offset, value);
}
}
inline U16 safe_instruction_bytestream2u16(U8 *stream, S32 &offset)
{
if (safe_instruction_check_address(stream, offset, 2))
{
return (bytestream2u16(stream, offset));
}
else
{
return 0;
}
}
inline void safe_instruction_u162bytestream(U8 *stream, S32 &offset, U16 value)
{
if (safe_instruction_check_address(stream, offset, 2))
{
u162bytestream(stream, offset, value);
}
}
inline F32 safe_instruction_bytestream2float(U8 *stream, S32 &offset)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_INTEGER]))
{
F32 value = bytestream2float(stream, offset);
if (!std::isfinite(value))
{
value = 0;
set_fault(stream, LSRF_MATH);
}
return value;
}
else
{
return 0;
}
}
inline void safe_instruction_float2bytestream(U8 *stream, S32 &offset, F32 value)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_FLOATINGPOINT]))
{
float2bytestream(stream, offset, value);
}
}
inline void safe_instruction_bytestream2char(char *buffer, U8 *stream, S32 &offset, S32 buffsize)
{
// This varies from the old method. Previously, we would copy up until we got an error,
// then halt the script via safe_isntruction_check_address. Now we don't bother
// copying a thing if there's an error.
if( safe_instruction_check_address(stream, offset, strlen( (const char *)stream + offset ) + 1 ) )
{
// Takes the same parms as this function. Won't overread, per above check.
bytestream2char( buffer, stream, offset, buffsize );
}
else
{
// Truncate - no point in copying
*buffer = 0;
}
}
inline void safe_instruction_bytestream_count_char(U8 *stream, S32 &offset)
{
while ( (safe_instruction_check_address(stream, offset, 1))
&&(*(stream + offset++)))
;
}
inline void safe_heap_bytestream_count_char(U8 *stream, S32 &offset)
{
while ( (safe_heap_check_address(stream, offset, 1))
&&(*(stream + offset++)))
;
}
inline void safe_instruction_char2bytestream(U8 *stream, S32 &offset, const char* buffer)
{
while ( (safe_instruction_check_address(stream, offset, 1))
&&(*(stream + offset++) = *buffer++))
;
}
inline void safe_instruction_bytestream2vector(LLVector3 &value, U8 *stream, S32 &offset)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_VECTOR]))
{
bytestream2vector(value, stream, offset);
}
}
inline void safe_instruction_vector2bytestream(U8 *stream, S32 &offset, const LLVector3 &value)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_VECTOR]))
{
vector2bytestream(stream, offset, value);
}
}
inline void safe_instruction_bytestream2quaternion(LLQuaternion &value, U8 *stream, S32 &offset)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_QUATERNION]))
{
bytestream2quaternion(value, stream, offset);
}
}
inline void safe_instruction_quaternion2bytestream(U8 *stream, S32 &offset, const LLQuaternion &value)
{
if (safe_instruction_check_address(stream, offset, LSCRIPTDataSize[LST_QUATERNION]))
{
quaternion2bytestream(stream, offset, value);
}
}
static inline LSCRIPTType char2type(char type)
{
switch(type)
{
case 'i':
return LST_INTEGER;
case 'f':
return LST_FLOATINGPOINT;
case 's':
return LST_STRING;
case 'k':
return LST_KEY;
case 'v':
return LST_VECTOR;
case 'q':
return LST_QUATERNION;
case 'l':
return LST_LIST;
default:
return LST_NULL;
}
}
#endif