1
0
Fork 0
forked from fte/fteqw
fteqw/plugins/quake3/botlib/l_script.c
2022-03-08 05:33:52 +00:00

1453 lines
40 KiB
C

/*
===========================================================================
Copyright (C) 1999-2005 Id Software, Inc.
This file is part of Quake III Arena source code.
Quake III Arena source code is free software; you can redistribute it
and/or modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the License,
or (at your option) any later version.
Quake III Arena source code is distributed in the hope that it will be
useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with Quake III Arena source code; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
===========================================================================
*/
/*****************************************************************************
* name: l_script.c
*
* desc: lexicographical parser
*
* $Archive: /MissionPack/code/botlib/l_script.c $
*
*****************************************************************************/
//#define SCREWUP
//#define BOTLIB
//#define MEQCC
//#define BSPC
#ifdef SCREWUP
#include <stdio.h>
#include <stdlib.h>
#include <limits.h>
#include <string.h>
#include <stdarg.h>
#include "l_memory.h"
#include "l_script.h"
typedef enum {qfalse, qtrue} qboolean;
#endif //SCREWUP
#ifdef BOTLIB
//include files for usage in the bot library
#include "q_shared.h"
#include "botlib.h"
#include "be_interface.h"
#include "l_script.h"
#include "l_memory.h"
#include "l_log.h"
#include "l_libvar.h"
#endif //BOTLIB
#ifdef MEQCC
//include files for usage in MrElusive's QuakeC Compiler
#include "qcc.h"
#include "l_script.h"
#include "l_memory.h"
#include "l_log.h"
#define qtrue true
#define qfalse false
#endif //MEQCC
#ifdef BSPC
//include files for usage in the BSP Converter
#include "../bspc/qbsp.h"
#include "../bspc/l_log.h"
#include "../bspc/l_mem.h"
#define qtrue true
#define qfalse false
#endif //BSPC
#define PUNCTABLE
//longer punctuations first
punctuation_t default_punctuations[] =
{
//binary operators
{">>=",P_RSHIFT_ASSIGN, NULL},
{"<<=",P_LSHIFT_ASSIGN, NULL},
//
{"...",P_PARMS, NULL},
//define merge operator
{"##",P_PRECOMPMERGE, NULL},
//logic operators
{"&&",P_LOGIC_AND, NULL},
{"||",P_LOGIC_OR, NULL},
{">=",P_LOGIC_GEQ, NULL},
{"<=",P_LOGIC_LEQ, NULL},
{"==",P_LOGIC_EQ, NULL},
{"!=",P_LOGIC_UNEQ, NULL},
//arithmatic operators
{"*=",P_MUL_ASSIGN, NULL},
{"/=",P_DIV_ASSIGN, NULL},
{"%=",P_MOD_ASSIGN, NULL},
{"+=",P_ADD_ASSIGN, NULL},
{"-=",P_SUB_ASSIGN, NULL},
{"++",P_INC, NULL},
{"--",P_DEC, NULL},
//binary operators
{"&=",P_BIN_AND_ASSIGN, NULL},
{"|=",P_BIN_OR_ASSIGN, NULL},
{"^=",P_BIN_XOR_ASSIGN, NULL},
{">>",P_RSHIFT, NULL},
{"<<",P_LSHIFT, NULL},
//reference operators
{"->",P_POINTERREF, NULL},
//C++
{"::",P_CPP1, NULL},
{".*",P_CPP2, NULL},
//arithmatic operators
{"*",P_MUL, NULL},
{"/",P_DIV, NULL},
{"%",P_MOD, NULL},
{"+",P_ADD, NULL},
{"-",P_SUB, NULL},
{"=",P_ASSIGN, NULL},
//binary operators
{"&",P_BIN_AND, NULL},
{"|",P_BIN_OR, NULL},
{"^",P_BIN_XOR, NULL},
{"~",P_BIN_NOT, NULL},
//logic operators
{"!",P_LOGIC_NOT, NULL},
{">",P_LOGIC_GREATER, NULL},
{"<",P_LOGIC_LESS, NULL},
//reference operator
{".",P_REF, NULL},
//seperators
{",",P_COMMA, NULL},
{";",P_SEMICOLON, NULL},
//label indication
{":",P_COLON, NULL},
//if statement
{"?",P_QUESTIONMARK, NULL},
//embracements
{"(",P_PARENTHESESOPEN, NULL},
{")",P_PARENTHESESCLOSE, NULL},
{"{",P_BRACEOPEN, NULL},
{"}",P_BRACECLOSE, NULL},
{"[",P_SQBRACKETOPEN, NULL},
{"]",P_SQBRACKETCLOSE, NULL},
//
{"\\",P_BACKSLASH, NULL},
//precompiler operator
{"#",P_PRECOMP, NULL},
#ifdef DOLLAR
{"$",P_DOLLAR, NULL},
#endif //DOLLAR
{NULL, 0}
};
#ifdef BSPC
char basefolder[MAX_PATH];
#else
char basefolder[MAX_QPATH];
#endif
#ifdef _WIN64
#ifdef FTE_SDL
#define vsnprintf linuxlike_vsnprintf
#endif
#endif
//===========================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//===========================================================================
void PS_CreatePunctuationTable(script_t *script, punctuation_t *punctuations)
{
int i;
punctuation_t *p, *lastp, *newp;
//get memory for the table
if (!script->punctuationtable) script->punctuationtable = (punctuation_t **)
GetMemory(256 * sizeof(punctuation_t *));
Com_Memset(script->punctuationtable, 0, 256 * sizeof(punctuation_t *));
//add the punctuations in the list to the punctuation table
for (i = 0; punctuations[i].p; i++)
{
newp = &punctuations[i];
lastp = NULL;
//sort the punctuations in this table entry on length (longer punctuations first)
for (p = script->punctuationtable[(unsigned int) newp->p[0]]; p; p = p->next)
{
if (strlen(p->p) < strlen(newp->p))
{
newp->next = p;
if (lastp) lastp->next = newp;
else script->punctuationtable[(unsigned int) newp->p[0]] = newp;
break;
} //end if
lastp = p;
} //end for
if (!p)
{
newp->next = NULL;
if (lastp) lastp->next = newp;
else script->punctuationtable[(unsigned int) newp->p[0]] = newp;
} //end if
} //end for
} //end of the function PS_CreatePunctuationTable
//===========================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//===========================================================================
char *PunctuationFromNum(script_t *script, int num)
{
int i;
for (i = 0; script->punctuations[i].p; i++)
{
if (script->punctuations[i].n == num) return script->punctuations[i].p;
} //end for
return "unkown punctuation";
} //end of the function PunctuationFromNum
//===========================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//===========================================================================
void QDECL ScriptError(script_t *script, char *str, ...)
{
char text[1024];
va_list ap;
if (script->flags & SCFL_NOERRORS) return;
va_start(ap, str);
vsnprintf(text, sizeof(text), str, ap);
va_end(ap);
#ifdef BOTLIB
botimport.Print(PRT_ERROR, "file %s, line %d: %s\n", script->filename, script->line, text);
#endif //BOTLIB
#ifdef MEQCC
printf("error: file %s, line %d: %s\n", script->filename, script->line, text);
#endif //MEQCC
#ifdef BSPC
Log_Print("error: file %s, line %d: %s\n", script->filename, script->line, text);
#endif //BSPC
} //end of the function ScriptError
//===========================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//===========================================================================
void QDECL ScriptWarning(script_t *script, char *str, ...)
{
char text[1024];
va_list ap;
if (script->flags & SCFL_NOWARNINGS) return;
va_start(ap, str);
vsnprintf(text, sizeof(text), str, ap);
va_end(ap);
#ifdef BOTLIB
botimport.Print(PRT_WARNING, "file %s, line %d: %s\n", script->filename, script->line, text);
#endif //BOTLIB
#ifdef MEQCC
printf("warning: file %s, line %d: %s\n", script->filename, script->line, text);
#endif //MEQCC
#ifdef BSPC
Log_Print("warning: file %s, line %d: %s\n", script->filename, script->line, text);
#endif //BSPC
} //end of the function ScriptWarning
//===========================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//===========================================================================
void SetScriptPunctuations(script_t *script, punctuation_t *p)
{
#ifdef PUNCTABLE
if (p) PS_CreatePunctuationTable(script, p);
else PS_CreatePunctuationTable(script, default_punctuations);
#endif //PUNCTABLE
if (p) script->punctuations = p;
else script->punctuations = default_punctuations;
} //end of the function SetScriptPunctuations
//============================================================================
// Reads spaces, tabs, C-like comments etc.
// When a newline character is found the scripts line counter is increased.
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadWhiteSpace(script_t *script)
{
while(1)
{
//skip white space
while(*script->script_p <= ' ')
{
if (!*script->script_p) return 0;
if (*script->script_p == '\n') script->line++;
script->script_p++;
} //end while
//skip comments
if (*script->script_p == '/')
{
//comments //
if (*(script->script_p+1) == '/')
{
script->script_p++;
do
{
script->script_p++;
if (!*script->script_p) return 0;
} //end do
while(*script->script_p != '\n');
script->line++;
script->script_p++;
if (!*script->script_p) return 0;
continue;
} //end if
//comments /* */
else if (*(script->script_p+1) == '*')
{
script->script_p++;
do
{
script->script_p++;
if (!*script->script_p) return 0;
if (*script->script_p == '\n') script->line++;
} //end do
while(!(*script->script_p == '*' && *(script->script_p+1) == '/'));
script->script_p++;
if (!*script->script_p) return 0;
script->script_p++;
if (!*script->script_p) return 0;
continue;
} //end if
} //end if
break;
} //end while
return 1;
} //end of the function PS_ReadWhiteSpace
//============================================================================
// Reads an escape character.
//
// Parameter: script : script to read from
// ch : place to store the read escape character
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadEscapeCharacter(script_t *script, char *ch)
{
int c, val, i;
//step over the leading '\\'
script->script_p++;
//determine the escape character
switch(*script->script_p)
{
case '\\': c = '\\'; break;
case 'n': c = '\n'; break;
case 'r': c = '\r'; break;
case 't': c = '\t'; break;
case 'v': c = '\v'; break;
case 'b': c = '\b'; break;
case 'f': c = '\f'; break;
case 'a': c = '\a'; break;
case '\'': c = '\''; break;
case '\"': c = '\"'; break;
case '\?': c = '\?'; break;
case 'x':
{
script->script_p++;
for (i = 0, val = 0; ; i++, script->script_p++)
{
c = *script->script_p;
if (c >= '0' && c <= '9') c = c - '0';
else if (c >= 'A' && c <= 'Z') c = c - 'A' + 10;
else if (c >= 'a' && c <= 'z') c = c - 'a' + 10;
else break;
val = (val << 4) + c;
} //end for
script->script_p--;
if (val > 0xFF)
{
ScriptWarning(script, "too large value in escape character");
val = 0xFF;
} //end if
c = val;
break;
} //end case
default: //NOTE: decimal ASCII code, NOT octal
{
if (*script->script_p < '0' || *script->script_p > '9') ScriptError(script, "unknown escape char");
for (i = 0, val = 0; ; i++, script->script_p++)
{
c = *script->script_p;
if (c >= '0' && c <= '9') c = c - '0';
else break;
val = val * 10 + c;
} //end for
script->script_p--;
if (val > 0xFF)
{
ScriptWarning(script, "too large value in escape character");
val = 0xFF;
} //end if
c = val;
break;
} //end default
} //end switch
//step over the escape character or the last digit of the number
script->script_p++;
//store the escape character
*ch = c;
//successfully read escape character
return 1;
} //end of the function PS_ReadEscapeCharacter
//============================================================================
// Reads C-like string. Escape characters are interpretted.
// Quotes are included with the string.
// Reads two strings with a white space between them as one string.
//
// Parameter: script : script to read from
// token : buffer to store the string
// Returns: qtrue when a string was read successfully
// Changes Globals: -
//============================================================================
int PS_ReadString(script_t *script, token_t *token, int quote)
{
int len, tmpline;
char *tmpscript_p;
if (quote == '\"') token->type = TT_STRING;
else token->type = TT_LITERAL;
len = 0;
//leading quote
token->string[len++] = *script->script_p++;
//
while(1)
{
//minus 2 because trailing double quote and zero have to be appended
if (len >= MAX_TOKEN - 2)
{
ScriptError(script, "string longer than MAX_TOKEN = %d", MAX_TOKEN);
return 0;
} //end if
//if there is an escape character and
//if escape characters inside a string are allowed
if (*script->script_p == '\\' && !(script->flags & SCFL_NOSTRINGESCAPECHARS))
{
if (!PS_ReadEscapeCharacter(script, &token->string[len]))
{
token->string[len] = 0;
return 0;
} //end if
len++;
} //end if
//if a trailing quote
else if (*script->script_p == quote)
{
//step over the double quote
script->script_p++;
//if white spaces in a string are not allowed
if (script->flags & SCFL_NOSTRINGWHITESPACES) break;
//
tmpscript_p = script->script_p;
tmpline = script->line;
//read unusefull stuff between possible two following strings
if (!PS_ReadWhiteSpace(script))
{
script->script_p = tmpscript_p;
script->line = tmpline;
break;
} //end if
//if there's no leading double qoute
if (*script->script_p != quote)
{
script->script_p = tmpscript_p;
script->line = tmpline;
break;
} //end if
//step over the new leading double quote
script->script_p++;
} //end if
else
{
if (*script->script_p == '\0')
{
token->string[len] = 0;
ScriptError(script, "missing trailing quote");
return 0;
} //end if
if (*script->script_p == '\n')
{
token->string[len] = 0;
ScriptError(script, "newline inside string %s", token->string);
return 0;
} //end if
token->string[len++] = *script->script_p++;
} //end else
} //end while
//trailing quote
token->string[len++] = quote;
//end string with a zero
token->string[len] = '\0';
//the sub type is the length of the string
token->subtype = len;
return 1;
} //end of the function PS_ReadString
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadName(script_t *script, token_t *token)
{
int len = 0;
char c;
token->type = TT_NAME;
do
{
token->string[len++] = *script->script_p++;
if (len >= MAX_TOKEN)
{
ScriptError(script, "name longer than MAX_TOKEN = %d", MAX_TOKEN);
return 0;
} //end if
c = *script->script_p;
} while ((c >= 'a' && c <= 'z') ||
(c >= 'A' && c <= 'Z') ||
(c >= '0' && c <= '9') ||
c == '_');
token->string[len] = '\0';
//the sub type is the length of the name
token->subtype = len;
return 1;
} //end of the function PS_ReadName
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void NumberValue(char *string, int subtype, unsigned long int *intvalue,
float *floatvalue)
{
unsigned long int dotfound = 0;
*intvalue = 0;
*floatvalue = 0;
//floating point number
if (subtype & TT_FLOAT)
{
while(*string)
{
if (*string == '.')
{
if (dotfound) return;
dotfound = 10;
string++;
} //end if
if (dotfound)
{
*floatvalue = *floatvalue + (float) (*string - '0') /
(float) dotfound;
dotfound *= 10;
} //end if
else
{
*floatvalue = *floatvalue * 10.0 + (float) (*string - '0');
} //end else
string++;
} //end while
*intvalue = (unsigned long) *floatvalue;
} //end if
else if (subtype & TT_DECIMAL)
{
while(*string) *intvalue = *intvalue * 10 + (*string++ - '0');
*floatvalue = *intvalue;
} //end else if
else if (subtype & TT_HEX)
{
//step over the leading 0x or 0X
string += 2;
while(*string)
{
*intvalue <<= 4;
if (*string >= 'a' && *string <= 'f') *intvalue += *string - 'a' + 10;
else if (*string >= 'A' && *string <= 'F') *intvalue += *string - 'A' + 10;
else *intvalue += *string - '0';
string++;
} //end while
*floatvalue = *intvalue;
} //end else if
else if (subtype & TT_OCTAL)
{
//step over the first zero
string += 1;
while(*string) *intvalue = (*intvalue << 3) + (*string++ - '0');
*floatvalue = *intvalue;
} //end else if
else if (subtype & TT_BINARY)
{
//step over the leading 0b or 0B
string += 2;
while(*string) *intvalue = (*intvalue << 1) + (*string++ - '0');
*floatvalue = *intvalue;
} //end else if
} //end of the function NumberValue
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadNumber(script_t *script, token_t *token)
{
int len = 0, i;
int octal, dot;
char c;
// unsigned long int intvalue = 0;
// double floatvalue = 0;
token->type = TT_NUMBER;
//check for a hexadecimal number
if (*script->script_p == '0' &&
(*(script->script_p + 1) == 'x' ||
*(script->script_p + 1) == 'X'))
{
token->string[len++] = *script->script_p++;
token->string[len++] = *script->script_p++;
c = *script->script_p;
//hexadecimal
while((c >= '0' && c <= '9') ||
(c >= 'a' && c <= 'f') ||
(c >= 'A' && c <= 'A'))
{
token->string[len++] = *script->script_p++;
if (len >= MAX_TOKEN)
{
ScriptError(script, "hexadecimal number longer than MAX_TOKEN = %d", MAX_TOKEN);
return 0;
} //end if
c = *script->script_p;
} //end while
token->subtype |= TT_HEX;
} //end if
#ifdef BINARYNUMBERS
//check for a binary number
else if (*script->script_p == '0' &&
(*(script->script_p + 1) == 'b' ||
*(script->script_p + 1) == 'B'))
{
token->string[len++] = *script->script_p++;
token->string[len++] = *script->script_p++;
c = *script->script_p;
//binary
while(c == '0' || c == '1')
{
token->string[len++] = *script->script_p++;
if (len >= MAX_TOKEN)
{
ScriptError(script, "binary number longer than MAX_TOKEN = %d", MAX_TOKEN);
return 0;
} //end if
c = *script->script_p;
} //end while
token->subtype |= TT_BINARY;
} //end if
#endif //BINARYNUMBERS
else //decimal or octal integer or floating point number
{
octal = qfalse;
dot = qfalse;
if (*script->script_p == '0') octal = qtrue;
while(1)
{
c = *script->script_p;
if (c == '.') dot = qtrue;
else if (c == '8' || c == '9') octal = qfalse;
else if (c < '0' || c > '9') break;
token->string[len++] = *script->script_p++;
if (len >= MAX_TOKEN - 1)
{
ScriptError(script, "number longer than MAX_TOKEN = %d", MAX_TOKEN);
return 0;
} //end if
} //end while
if (octal) token->subtype |= TT_OCTAL;
else token->subtype |= TT_DECIMAL;
if (dot) token->subtype |= TT_FLOAT;
} //end else
for (i = 0; i < 2; i++)
{
c = *script->script_p;
//check for a LONG number
if ( (c == 'l' || c == 'L')
&& !(token->subtype & TT_LONG))
{
script->script_p++;
token->subtype |= TT_LONG;
} //end if
//check for an UNSIGNED number
else if ( (c == 'u' || c == 'U')
&& !(token->subtype & (TT_UNSIGNED | TT_FLOAT)))
{
script->script_p++;
token->subtype |= TT_UNSIGNED;
} //end if
} //end for
token->string[len] = '\0';
#ifdef NUMBERVALUE
NumberValue(token->string, token->subtype, &token->intvalue, &token->floatvalue);
#endif //NUMBERVALUE
if (!(token->subtype & TT_FLOAT)) token->subtype |= TT_INTEGER;
return 1;
} //end of the function PS_ReadNumber
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadLiteral(script_t *script, token_t *token)
{
token->type = TT_LITERAL;
//first quote
token->string[0] = *script->script_p++;
//check for end of file
if (!*script->script_p)
{
ScriptError(script, "end of file before trailing \'");
return 0;
} //end if
//if it is an escape character
if (*script->script_p == '\\')
{
if (!PS_ReadEscapeCharacter(script, &token->string[1])) return 0;
} //end if
else
{
token->string[1] = *script->script_p++;
} //end else
//check for trailing quote
if (*script->script_p != '\'')
{
ScriptWarning(script, "too many characters in literal, ignored");
while(*script->script_p &&
*script->script_p != '\'' &&
*script->script_p != '\n')
{
script->script_p++;
} //end while
if (*script->script_p == '\'') script->script_p++;
} //end if
//store the trailing quote
token->string[2] = *script->script_p++;
//store trailing zero to end the string
token->string[3] = '\0';
//the sub type is the integer literal value
token->subtype = token->string[1];
//
return 1;
} //end of the function PS_ReadLiteral
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadPunctuation(script_t *script, token_t *token)
{
int len;
char *p;
punctuation_t *punc;
#ifdef PUNCTABLE
for (punc = script->punctuationtable[(unsigned int)*script->script_p]; punc; punc = punc->next)
{
#else
int i;
for (i = 0; script->punctuations[i].p; i++)
{
punc = &script->punctuations[i];
#endif //PUNCTABLE
p = punc->p;
len = strlen(p);
//if the script contains at least as much characters as the punctuation
if (script->script_p + len <= script->end_p)
{
//if the script contains the punctuation
if (!strncmp(script->script_p, p, len))
{
strncpy(token->string, p, MAX_TOKEN);
script->script_p += len;
token->type = TT_PUNCTUATION;
//sub type is the number of the punctuation
token->subtype = punc->n;
return 1;
} //end if
} //end if
} //end for
return 0;
} //end of the function PS_ReadPunctuation
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadPrimitive(script_t *script, token_t *token)
{
int len;
len = 0;
while(*script->script_p > ' ' && *script->script_p != ';')
{
if (len >= MAX_TOKEN)
{
ScriptError(script, "primitive token longer than MAX_TOKEN = %d", MAX_TOKEN);
return 0;
} //end if
token->string[len++] = *script->script_p++;
} //end while
token->string[len] = 0;
//copy the token into the script structure
Com_Memcpy(&script->token, token, sizeof(token_t));
//primitive reading successfull
return 1;
} //end of the function PS_ReadPrimitive
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ReadToken(script_t *script, token_t *token)
{
//if there is a token available (from UnreadToken)
if (script->tokenavailable)
{
script->tokenavailable = 0;
Com_Memcpy(token, &script->token, sizeof(token_t));
return 1;
} //end if
//save script pointer
script->lastscript_p = script->script_p;
//save line counter
script->lastline = script->line;
//clear the token stuff
Com_Memset(token, 0, sizeof(token_t));
//start of the white space
script->whitespace_p = script->script_p;
token->whitespace_p = script->script_p;
//read unusefull stuff
if (!PS_ReadWhiteSpace(script)) return 0;
//end of the white space
script->endwhitespace_p = script->script_p;
token->endwhitespace_p = script->script_p;
//line the token is on
token->line = script->line;
//number of lines crossed before token
token->linescrossed = script->line - script->lastline;
//if there is a leading double quote
if (*script->script_p == '\"')
{
if (!PS_ReadString(script, token, '\"')) return 0;
} //end if
//if an literal
else if (*script->script_p == '\'')
{
//if (!PS_ReadLiteral(script, token)) return 0;
if (!PS_ReadString(script, token, '\'')) return 0;
} //end if
//if there is a number
else if ((*script->script_p >= '0' && *script->script_p <= '9') ||
(*script->script_p == '.' &&
(*(script->script_p + 1) >= '0' && *(script->script_p + 1) <= '9')))
{
if (!PS_ReadNumber(script, token)) return 0;
} //end if
//if this is a primitive script
else if (script->flags & SCFL_PRIMITIVE)
{
return PS_ReadPrimitive(script, token);
} //end else if
//if there is a name
else if ((*script->script_p >= 'a' && *script->script_p <= 'z') ||
(*script->script_p >= 'A' && *script->script_p <= 'Z') ||
*script->script_p == '_')
{
if (!PS_ReadName(script, token)) return 0;
} //end if
//check for punctuations
else if (!PS_ReadPunctuation(script, token))
{
ScriptError(script, "can't read token");
return 0;
} //end if
//copy the token into the script structure
Com_Memcpy(&script->token, token, sizeof(token_t));
//successfully read a token
return 1;
} //end of the function PS_ReadToken
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ExpectTokenString(script_t *script, char *string)
{
token_t token;
if (!PS_ReadToken(script, &token))
{
ScriptError(script, "couldn't find expected %s", string);
return 0;
} //end if
if (strcmp(token.string, string))
{
ScriptError(script, "expected %s, found %s", string, token.string);
return 0;
} //end if
return 1;
} //end of the function PS_ExpectToken
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ExpectTokenType(script_t *script, int type, int subtype, token_t *token)
{
char str[MAX_TOKEN];
if (!PS_ReadToken(script, token))
{
ScriptError(script, "couldn't read expected token");
return 0;
} //end if
if (token->type != type)
{
if (type == TT_STRING) strcpy(str, "string");
if (type == TT_LITERAL) strcpy(str, "literal");
if (type == TT_NUMBER) strcpy(str, "number");
if (type == TT_NAME) strcpy(str, "name");
if (type == TT_PUNCTUATION) strcpy(str, "punctuation");
ScriptError(script, "expected a %s, found %s", str, token->string);
return 0;
} //end if
if (token->type == TT_NUMBER)
{
if ((token->subtype & subtype) != subtype)
{
if (subtype & TT_DECIMAL) strcpy(str, "decimal");
if (subtype & TT_HEX) strcpy(str, "hex");
if (subtype & TT_OCTAL) strcpy(str, "octal");
if (subtype & TT_BINARY) strcpy(str, "binary");
if (subtype & TT_LONG) strcat(str, " long");
if (subtype & TT_UNSIGNED) strcat(str, " unsigned");
if (subtype & TT_FLOAT) strcat(str, " float");
if (subtype & TT_INTEGER) strcat(str, " integer");
ScriptError(script, "expected %s, found %s", str, token->string);
return 0;
} //end if
} //end if
else if (token->type == TT_PUNCTUATION)
{
if (subtype < 0)
{
ScriptError(script, "BUG: wrong punctuation subtype");
return 0;
} //end if
if (token->subtype != subtype)
{
ScriptError(script, "expected %s, found %s",
script->punctuations[subtype], token->string);
return 0;
} //end if
} //end else if
return 1;
} //end of the function PS_ExpectTokenType
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_ExpectAnyToken(script_t *script, token_t *token)
{
if (!PS_ReadToken(script, token))
{
ScriptError(script, "couldn't read expected token");
return 0;
} //end if
else
{
return 1;
} //end else
} //end of the function PS_ExpectAnyToken
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_CheckTokenString(script_t *script, char *string)
{
token_t tok;
if (!PS_ReadToken(script, &tok)) return 0;
//if the token is available
if (!strcmp(tok.string, string)) return 1;
//token not available
script->script_p = script->lastscript_p;
return 0;
} //end of the function PS_CheckTokenString
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_CheckTokenType(script_t *script, int type, int subtype, token_t *token)
{
token_t tok;
if (!PS_ReadToken(script, &tok)) return 0;
//if the type matches
if (tok.type == type &&
(tok.subtype & subtype) == subtype)
{
Com_Memcpy(token, &tok, sizeof(token_t));
return 1;
} //end if
//token is not available
script->script_p = script->lastscript_p;
return 0;
} //end of the function PS_CheckTokenType
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int PS_SkipUntilString(script_t *script, char *string)
{
token_t token;
while(PS_ReadToken(script, &token))
{
if (!strcmp(token.string, string)) return 1;
} //end while
return 0;
} //end of the function PS_SkipUntilString
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void PS_UnreadLastToken(script_t *script)
{
script->tokenavailable = 1;
} //end of the function UnreadLastToken
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void PS_UnreadToken(script_t *script, token_t *token)
{
Com_Memcpy(&script->token, token, sizeof(token_t));
script->tokenavailable = 1;
} //end of the function UnreadToken
//============================================================================
// returns the next character of the read white space, returns NULL if none
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
char PS_NextWhiteSpaceChar(script_t *script)
{
if (script->whitespace_p != script->endwhitespace_p)
{
return *script->whitespace_p++;
} //end if
else
{
return 0;
} //end else
} //end of the function PS_NextWhiteSpaceChar
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void StripDoubleQuotes(char *string)
{
if (*string == '\"')
{
memmove(string, string+1, strlen(string));
} //end if
if (string[strlen(string)-1] == '\"')
{
string[strlen(string)-1] = '\0';
} //end if
} //end of the function StripDoubleQuotes
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void StripSingleQuotes(char *string)
{
if (*string == '\'')
{
memmove(string, string+1, strlen(string));
} //end if
if (string[strlen(string)-1] == '\'')
{
string[strlen(string)-1] = '\0';
} //end if
} //end of the function StripSingleQuotes
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
float ReadSignedFloat(script_t *script)
{
token_t token;
float sign = 1.0;
PS_ExpectAnyToken(script, &token);
if (!strcmp(token.string, "-"))
{
if(!PS_ExpectAnyToken(script, &token))
{
ScriptError(script, "Missing float value\n", token.string);
return 0;
}
sign = -1.0;
}
if (token.type != TT_NUMBER)
{
ScriptError(script, "expected float value, found %s\n", token.string);
return 0;
}
return sign * token.floatvalue;
} //end of the function ReadSignedFloat
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
signed long int ReadSignedInt(script_t *script)
{
token_t token;
signed long int sign = 1;
PS_ExpectAnyToken(script, &token);
if (!strcmp(token.string, "-"))
{
if(!PS_ExpectAnyToken(script, &token))
{
ScriptError(script, "Missing integer value\n", token.string);
return 0;
}
sign = -1;
}
if (token.type != TT_NUMBER || token.subtype == TT_FLOAT)
{
ScriptError(script, "expected integer value, found %s\n", token.string);
return 0;
}
return sign * token.intvalue;
} //end of the function ReadSignedInt
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void SetScriptFlags(script_t *script, int flags)
{
script->flags = flags;
} //end of the function SetScriptFlags
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int GetScriptFlags(script_t *script)
{
return script->flags;
} //end of the function GetScriptFlags
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void ResetScript(script_t *script)
{
//pointer in script buffer
script->script_p = script->buffer;
//pointer in script buffer before reading token
script->lastscript_p = script->buffer;
//begin of white space
script->whitespace_p = NULL;
//end of white space
script->endwhitespace_p = NULL;
//set if there's a token available in script->token
script->tokenavailable = 0;
//
script->line = 1;
script->lastline = 1;
//clear the saved token
Com_Memset(&script->token, 0, sizeof(token_t));
} //end of the function ResetScript
//============================================================================
// returns true if at the end of the script
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int EndOfScript(script_t *script)
{
return script->script_p >= script->end_p;
} //end of the function EndOfScript
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int NumLinesCrossed(script_t *script)
{
return script->line - script->lastline;
} //end of the function NumLinesCrossed
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int ScriptSkipTo(script_t *script, char *value)
{
int len;
char firstchar;
firstchar = *value;
len = strlen(value);
do
{
if (!PS_ReadWhiteSpace(script)) return 0;
if (*script->script_p == firstchar)
{
if (!strncmp(script->script_p, value, len))
{
return 1;
} //end if
} //end if
script->script_p++;
} while(1);
} //end of the function ScriptSkipTo
#ifndef BOTLIB
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
int FileLength(FILE *fp)
{
int pos;
int end;
pos = ftell(fp);
fseek(fp, 0, SEEK_END);
end = ftell(fp);
fseek(fp, pos, SEEK_SET);
return end;
} //end of the function FileLength
#endif
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
script_t *LoadScriptFile(const char *filename)
{
#ifdef BOTLIB
fileHandle_t fp;
char pathname[MAX_QPATH];
#else
FILE *fp;
#endif
int length;
void *buffer;
script_t *script;
#ifdef BOTLIB
if (strlen(basefolder))
Com_sprintf(pathname, sizeof(pathname), "%s/%s", basefolder, filename);
else
Com_sprintf(pathname, sizeof(pathname), "%s", filename);
length = botimport.FS_FOpenFile( pathname, &fp, FS_READ );
if (!fp) return NULL;
#else
fp = fopen(filename, "rb");
if (!fp) return NULL;
length = FileLength(fp);
#endif
buffer = GetClearedMemory(sizeof(script_t) + length + 1);
script = (script_t *) buffer;
Com_Memset(script, 0, sizeof(script_t));
strcpy(script->filename, filename);
script->buffer = (char *) buffer + sizeof(script_t);
script->buffer[length] = 0;
script->length = length;
//pointer in script buffer
script->script_p = script->buffer;
//pointer in script buffer before reading token
script->lastscript_p = script->buffer;
//pointer to end of script buffer
script->end_p = &script->buffer[length];
//set if there's a token available in script->token
script->tokenavailable = 0;
//
script->line = 1;
script->lastline = 1;
//
SetScriptPunctuations(script, NULL);
//
#ifdef BOTLIB
botimport.FS_Read(script->buffer, length, fp);
botimport.FS_FCloseFile(fp);
#else
if (fread(script->buffer, length, 1, fp) != 1)
{
FreeMemory(buffer);
script = NULL;
} //end if
fclose(fp);
#endif
return script;
} //end of the function LoadScriptFile
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
script_t *LoadScriptMemory(char *ptr, int length, char *name)
{
void *buffer;
script_t *script;
buffer = GetClearedMemory(sizeof(script_t) + length + 1);
script = (script_t *) buffer;
Com_Memset(script, 0, sizeof(script_t));
strcpy(script->filename, name);
script->buffer = (char *) buffer + sizeof(script_t);
script->buffer[length] = 0;
script->length = length;
//pointer in script buffer
script->script_p = script->buffer;
//pointer in script buffer before reading token
script->lastscript_p = script->buffer;
//pointer to end of script buffer
script->end_p = &script->buffer[length];
//set if there's a token available in script->token
script->tokenavailable = 0;
//
script->line = 1;
script->lastline = 1;
//
SetScriptPunctuations(script, NULL);
//
Com_Memcpy(script->buffer, ptr, length);
//
return script;
} //end of the function LoadScriptMemory
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void FreeScript(script_t *script)
{
#ifdef PUNCTABLE
if (script->punctuationtable) FreeMemory(script->punctuationtable);
#endif //PUNCTABLE
FreeMemory(script);
} //end of the function FreeScript
//============================================================================
//
// Parameter: -
// Returns: -
// Changes Globals: -
//============================================================================
void PS_SetBaseFolder(char *path)
{
#ifdef BSPC
sprintf(basefolder, path);
#else
Com_sprintf(basefolder, sizeof(basefolder), "%s", path);
#endif
} //end of the function PS_SetBaseFolder