virtualx-engine/modules/gdscript/gd_tokenizer.cpp

974 lines
22 KiB
C++
Raw Normal View History

2014-02-10 02:10:30 +01:00
/*************************************************************************/
/* gd_tokenizer.cpp */
/*************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* http://www.godotengine.org */
/*************************************************************************/
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/*************************************************************************/
#include "gd_tokenizer.h"
#include "print_string.h"
#include "gd_functions.h"
const char* GDTokenizer::token_names[TK_MAX]={
"Empty",
"Identifier",
"Constant",
"Self",
"Built-In Type",
"Built-In Func",
"In",
"'=='",
"'!='",
"'<'",
"'<='",
"'>'",
"'>='",
"'and'",
"'or'",
"'not'",
"'+'",
"'-'",
"'*'",
"'/'",
"'%'",
"'<<'",
"'>>'",
"'='",
"'+='",
"'-='",
"'*='",
"'/='",
"'%='",
"'<<='",
"'>>='",
"'&='",
"'|='",
"'^='",
"'&'",
"'|'",
"'^'",
"'~'",
//"Plus Plus",
//"Minus Minus",
"if",
"elif",
"else",
"for",
"do",
"while",
"switch",
"case",
"break",
"continue",
"pass",
"return",
"func",
"class",
"extends",
"tool",
"static",
"export",
"const",
"var",
"preload",
"assert",
"'['",
"']'",
"'{'",
"'}'",
"'('",
"')'",
"','",
"';'",
"'.'",
"'?'",
"':'",
"'\\n'",
"Error",
"EOF"};
const char *GDTokenizer::get_token_name(Token p_token) {
ERR_FAIL_INDEX_V(p_token,TK_MAX,"<error>");
return token_names[p_token];
}
static bool _is_text_char(CharType c) {
return (c>='a' && c<='z') || (c>='A' && c<='Z') || (c>='0' && c<='9') || c=='_';
}
static bool _is_number(CharType c) {
return (c>='0' && c<='9');
}
static bool _is_hex(CharType c) {
return (c>='0' && c<='9') || (c>='a' && c<='f') || (c>='A' && c<='F');
}
void GDTokenizer::_make_token(Token p_type) {
TokenData &tk=tk_rb[tk_rb_pos];
tk.type=p_type;
tk.line=line;
tk.col=column;
tk_rb_pos=(tk_rb_pos+1)%TK_RB_SIZE;
}
void GDTokenizer::_make_identifier(const StringName& p_identifier) {
TokenData &tk=tk_rb[tk_rb_pos];
tk.type=TK_IDENTIFIER;
tk.identifier=p_identifier;
tk.line=line;
tk.col=column;
tk_rb_pos=(tk_rb_pos+1)%TK_RB_SIZE;
}
void GDTokenizer::_make_built_in_func(GDFunctions::Function p_func) {
TokenData &tk=tk_rb[tk_rb_pos];
tk.type=TK_BUILT_IN_FUNC;
tk.func=p_func;
tk.line=line;
tk.col=column;
tk_rb_pos=(tk_rb_pos+1)%TK_RB_SIZE;
}
void GDTokenizer::_make_constant(const Variant& p_constant) {
TokenData &tk=tk_rb[tk_rb_pos];
tk.type=TK_CONSTANT;
tk.constant=p_constant;
tk.line=line;
tk.col=column;
tk_rb_pos=(tk_rb_pos+1)%TK_RB_SIZE;
}
void GDTokenizer::_make_type(const Variant::Type& p_type) {
TokenData &tk=tk_rb[tk_rb_pos];
tk.type=TK_BUILT_IN_TYPE;
tk.vtype=p_type;
tk.line=line;
tk.col=column;
tk_rb_pos=(tk_rb_pos+1)%TK_RB_SIZE;
}
void GDTokenizer::_make_error(const String& p_error) {
error_flag=true;
last_error=p_error;
TokenData &tk=tk_rb[tk_rb_pos];
tk.type=TK_ERROR;
tk.constant=p_error;
tk.line=line;
tk.col=column;
tk_rb_pos=(tk_rb_pos+1)%TK_RB_SIZE;
}
void GDTokenizer::_make_newline(int p_spaces) {
TokenData &tk=tk_rb[tk_rb_pos];
tk.type=TK_NEWLINE;
tk.constant=p_spaces;
tk.line=line;
tk.col=column;
tk_rb_pos=(tk_rb_pos+1)%TK_RB_SIZE;
}
void GDTokenizer::_advance() {
if (error_flag) {
//parser broke
_make_error(last_error);
return;
}
if (code_pos>=len) {
_make_token(TK_EOF);
return;
}
#define GETCHAR(m_ofs) ((m_ofs+code_pos)>=len?0:_code[m_ofs+code_pos])
#define INCPOS(m_amount) { code_pos+=m_amount; column+=m_amount; }
while (true) {
bool is_node_path=false;
switch(GETCHAR(0)) {
case 0:
_make_token(TK_EOF);
break;
case '\t':
case '\r':
case ' ':
INCPOS(1);
continue;
case '\n': {
line++;
INCPOS(1);
column=0;
int i=0;
while(GETCHAR(i)==' ' || GETCHAR(i)=='\t') {
i++;
}
_make_newline(i);
return;
}
#if 1 //py style tokenizer
case '#': { // line comment skip
while(GETCHAR(0)!='\n') {
code_pos++;
if (GETCHAR(0)==0) { //end of file
_make_error("Unterminated Comment");
return;
}
}
INCPOS(1);
column=0;
line++;
int i=0;
while(GETCHAR(i)==' ' || GETCHAR(i)=='\t') {
i++;
}
_make_newline(i);
return;
} break;
#endif
case '/': {
switch(GETCHAR(1)) {
#if 0 // c style tokenizer
case '*': { // block comment
int pos = code_pos+2;
int new_line=line;
int new_col=column+2;
while(true) {
if (_code[pos]=='0') {
_make_error("Unterminated Comment");
code_pos=pos;
return;
}
if (_code[pos]=='*' && _code[pos+1]=='/') {
new_col+=2;
pos+=2; //compensate
break;
} else if (_code[pos]=='\n') {
new_line++;
new_col=0;
} else {
new_col++;
}
pos++;
}
column=new_col;
line=new_line;
code_pos=pos;
continue;
} break;
case '/': { // line comment skip
while(GETCHAR(0)!='\n') {
code_pos++;
if (GETCHAR(0)==0) { //end of file
_make_error("Unterminated Comment");
return;
}
}
INCPOS(1);
column=0;
line++;
continue;
} break;
#endif
case '=': { // diveq
_make_token(TK_OP_ASSIGN_DIV);
INCPOS(1);
} break;
default:
_make_token(TK_OP_DIV);
}
} break;
case '=': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_EQUAL);
INCPOS(1);
} else
_make_token(TK_OP_ASSIGN);
} break;
case '<': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_LESS_EQUAL);
INCPOS(1);
} else if (GETCHAR(1)=='<') {
if (GETCHAR(2)=='=') {
_make_token(TK_OP_ASSIGN_SHIFT_LEFT);
INCPOS(1);
} else {
_make_token(TK_OP_SHIFT_LEFT);
}
INCPOS(1);
} else
_make_token(TK_OP_LESS);
} break;
case '>': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_GREATER_EQUAL);
INCPOS(1);
} else if (GETCHAR(1)=='>') {
if (GETCHAR(2)=='=') {
_make_token(TK_OP_ASSIGN_SHIFT_RIGHT);
INCPOS(1);
} else {
_make_token(TK_OP_SHIFT_RIGHT);
}
INCPOS(1);
} else {
_make_token(TK_OP_GREATER);
}
} break;
case '!': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_NOT_EQUAL);
INCPOS(1);
} else {
_make_token(TK_OP_NOT);
}
} break;
//case '"' //string - no strings in shader
//case '\'' //string - no strings in shader
case '{':
_make_token(TK_CURLY_BRACKET_OPEN);
break;
case '}':
_make_token(TK_CURLY_BRACKET_CLOSE);
break;
case '[':
_make_token(TK_BRACKET_OPEN);
break;
case ']':
_make_token(TK_BRACKET_CLOSE);
break;
case '(':
_make_token(TK_PARENTHESIS_OPEN);
break;
case ')':
_make_token(TK_PARENTHESIS_CLOSE);
break;
case ',':
_make_token(TK_COMMA);
break;
case ';':
_make_token(TK_SEMICOLON);
break;
case '?':
_make_token(TK_QUESTION_MARK);
break;
case ':':
_make_token(TK_COLON); //for methods maybe but now useless.
break;
case '^': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_ASSIGN_BIT_XOR);
INCPOS(1);
} else {
_make_token(TK_OP_BIT_XOR);
}
} break;
case '~':
_make_token(TK_OP_BIT_INVERT);
break;
case '&': {
if (GETCHAR(1)=='&') {
_make_token(TK_OP_AND);
INCPOS(1);
} else if (GETCHAR(1)=='=') {
_make_token(TK_OP_ASSIGN_BIT_AND);
INCPOS(1);
} else {
_make_token(TK_OP_BIT_AND);
}
} break;
case '|': {
if (GETCHAR(1)=='|') {
_make_token(TK_OP_OR);
INCPOS(1);
} else if (GETCHAR(1)=='=') {
_make_token(TK_OP_ASSIGN_BIT_OR);
INCPOS(1);
} else {
_make_token(TK_OP_BIT_OR);
}
} break;
case '*': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_ASSIGN_MUL);
INCPOS(1);
} else {
_make_token(TK_OP_MUL);
}
} break;
case '+': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_ASSIGN_ADD);
INCPOS(1);
//} else if (GETCHAR(1)=='+') {
// _make_token(TK_OP_PLUS_PLUS);
// INCPOS(1);
} else {
_make_token(TK_OP_ADD);
}
} break;
case '-': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_ASSIGN_SUB);
INCPOS(1);
//} else if (GETCHAR(1)=='-') {
// _make_token(TK_OP_MINUS_MINUS);
// INCPOS(1);
} else {
_make_token(TK_OP_SUB);
}
} break;
case '%': {
if (GETCHAR(1)=='=') {
_make_token(TK_OP_ASSIGN_MOD);
INCPOS(1);
} else {
_make_token(TK_OP_MOD);
}
} break;
case '@':
if (CharType(GETCHAR(1))!='"') {
_make_error("Unexpected '@'");
return;
}
INCPOS(1);
is_node_path=true;
case '"': {
int i=1;
String str;
while(true) {
if (CharType(GETCHAR(i)==0)) {
_make_error("Unterminated String");
return;
} else if (CharType(GETCHAR(i)=='"')) {
break;
} else if (CharType(GETCHAR(i)=='\\')) {
//escaped characters...
i++;
CharType next = GETCHAR(i);
if (next==0) {
_make_error("Unterminated String");
return;
}
CharType res=0;
switch(next) {
case 'a': res=7; break;
case 'b': res=8; break;
case 't': res=9; break;
case 'n': res=10; break;
case 'v': res=11; break;
case 'f': res=12; break;
case 'r': res=13; break;
case '\'': res='\''; break;
case '\"': res='\"'; break;
case '\\': res='\\'; break;
case 'x': {
//hexnumbarh - oct is deprecated
int read=0;
for(int j=0;j<4;j++) {
CharType c = GETCHAR(i+j);
if (c==0) {
_make_error("Unterminated String");
return;
}
if (!_is_hex(c)) {
if (j==0 || !(j&1)) {
_make_error("Malformed hex constant in string");
return;
} else
break;
}
CharType v;
if (c>='0' && c<='9') {
v=c-'0';
} else if (c>='a' && c<='f') {
v=c-'a';
v+=10;
} else if (c>='A' && c<='F') {
v=c-'A';
v+=10;
} else {
ERR_PRINT("BUG");
v=0;
}
res<<=4;
res|=v;
read++;
}
i+=read-1;
} break;
default: {
_make_error("Invalid escape sequence");
return;
} break;
}
str+=res;
} else {
str+=CharType(GETCHAR(i));
}
i++;
}
INCPOS(i);
if (is_node_path) {
_make_constant(NodePath(str));
} else {
_make_constant(str);
}
} break;
default: {
if (_is_number(GETCHAR(0)) || (GETCHAR(0)=='.' && _is_number(GETCHAR(1)))) {
// parse number
bool period_found=false;
bool exponent_found=false;
bool hexa_found=false;
bool sign_found=false;
String str;
int i=0;
while(true) {
if (GETCHAR(i)=='.') {
if (period_found || exponent_found) {
_make_error("Invalid numeric constant at '.'");
return;
}
period_found=true;
} else if (GETCHAR(i)=='x') {
if (hexa_found || str.length()!=1 || !( (i==1 && str[0]=='0') || (i==2 && str[1]=='0' && str[0]=='-') ) ) {
_make_error("Invalid numeric constant at 'x'");
return;
}
hexa_found=true;
} else if (!hexa_found && GETCHAR(i)=='e') {
if (hexa_found || exponent_found) {
_make_error("Invalid numeric constant at 'e'");
return;
}
exponent_found=true;
} else if (_is_number(GETCHAR(i))) {
//all ok
} else if (hexa_found && _is_hex(GETCHAR(i))) {
} else if ((GETCHAR(i)=='-' || GETCHAR(i)=='+') && exponent_found) {
if (sign_found) {
_make_error("Invalid numeric constant at '-'");
return;
}
sign_found=true;
} else
break;
str+=CharType(GETCHAR(i));
i++;
}
if (!( _is_number(str[str.length()-1]) || (hexa_found && _is_hex(str[str.length()-1])))) {
_make_error("Invalid numeric constant: "+str);
return;
}
INCPOS(str.length());
if (hexa_found) {
int val = str.hex_to_int();
_make_constant(val);
} else if (period_found) {
real_t val = str.to_double();
//print_line("*%*%*%*% to convert: "+str+" result: "+rtos(val));
_make_constant(val);
} else {
int val = str.to_int();
_make_constant(val);
}
return;
}
if (GETCHAR(0)=='.') {
//parse period
_make_token(TK_PERIOD);
break;
}
if (_is_text_char(GETCHAR(0))) {
// parse identifier
String str;
str+=CharType(GETCHAR(0));
int i=1;
while(_is_text_char(GETCHAR(i))) {
str+=CharType(GETCHAR(i));
i++;
}
bool identifier=false;
if (str=="null") {
_make_constant(Variant());
} else if (str=="true") {
_make_constant(true);
} else if (str=="false") {
_make_constant(false);
} else {
bool found=false;
struct _bit { Variant::Type type; const char *text;};
//built in types
static const _bit type_list[]={
//types
{Variant::BOOL,"bool"},
{Variant::INT,"int"},
{Variant::REAL,"float"},
{Variant::STRING,"String"},
{Variant::VECTOR2,"vec2"},
{Variant::VECTOR2,"Vector2"},
{Variant::RECT2,"Rect2"},
{Variant::MATRIX32,"Matrix32"},
{Variant::MATRIX32,"mat32"},
{Variant::VECTOR3,"vec3"},
{Variant::VECTOR3,"Vector3"},
{Variant::_AABB,"AABB"},
{Variant::_AABB,"Rect3"},
{Variant::PLANE,"Plane"},
{Variant::QUAT,"Quat"},
{Variant::MATRIX3,"mat3"},
{Variant::MATRIX3,"Matrix3"},
{Variant::TRANSFORM,"trn"},
{Variant::TRANSFORM,"Transform"},
{Variant::COLOR,"Color"},
{Variant::IMAGE,"Image"},
{Variant::_RID,"RID"},
{Variant::OBJECT,"Object"},
{Variant::INPUT_EVENT,"InputEvent"},
{Variant::DICTIONARY,"dict"},
{Variant::DICTIONARY,"Dictionary"},
{Variant::ARRAY,"Array"},
{Variant::RAW_ARRAY,"RawArray"},
{Variant::INT_ARRAY,"IntArray"},
{Variant::REAL_ARRAY,"FloatArray"},
{Variant::STRING_ARRAY,"StringArray"},
{Variant::VECTOR2_ARRAY,"Vector2Array"},
{Variant::VECTOR3_ARRAY,"Vector3Array"},
{Variant::COLOR_ARRAY,"ColorArray"},
{Variant::VARIANT_MAX,NULL},
};
{
int idx=0;
while(type_list[idx].text) {
if (str==type_list[idx].text) {
_make_type(type_list[idx].type);
found=true;
break;
}
idx++;
}
}
if (!found) {
//built in func?
for(int i=0;i<GDFunctions::FUNC_MAX;i++) {
if (str==GDFunctions::get_func_name(GDFunctions::Function(i))) {
_make_built_in_func(GDFunctions::Function(i));
found=true;
break;
}
}
//keywor
}
if (!found) {
struct _kws { Token token; const char *text;};
static const _kws keyword_list[]={
//ops
{TK_OP_IN,"in"},
{TK_OP_NOT,"not"},
{TK_OP_OR,"or"},
{TK_OP_AND,"and"},
//func
{TK_PR_FUNCTION,"func"},
{TK_PR_FUNCTION,"function"},
{TK_PR_CLASS,"class"},
{TK_PR_EXTENDS,"extends"},
{TK_PR_TOOL,"tool"},
{TK_PR_STATIC,"static"},
{TK_PR_EXPORT,"export"},
{TK_PR_VAR,"var"},
{TK_PR_PRELOAD,"preload"},
{TK_PR_ASSERT,"assert"},
{TK_PR_CONST,"const"},
//controlflow
{TK_CF_IF,"if"},
{TK_CF_ELIF,"elif"},
{TK_CF_ELSE,"else"},
{TK_CF_FOR,"for"},
{TK_CF_WHILE,"while"},
{TK_CF_DO,"do"},
{TK_CF_SWITCH,"switch"},
{TK_CF_BREAK,"break"},
{TK_CF_CONTINUE,"continue"},
{TK_CF_RETURN,"return"},
{TK_CF_PASS,"pass"},
{TK_SELF,"self"},
{TK_ERROR,NULL}
};
int idx=0;
found=false;
while(keyword_list[idx].text) {
if (str==keyword_list[idx].text) {
_make_token(keyword_list[idx].token);
found=true;
break;
}
idx++;
}
}
if (!found)
identifier=true;
}
if (identifier) {
_make_identifier(str);
}
INCPOS(str.length());
return;
}
_make_error("Unknown character");
return;
} break;
}
INCPOS(1);
break;
}
}
void GDTokenizer::set_code(const String& p_code) {
code=p_code;
len = p_code.length();
if (len) {
_code=&code[0];
} else {
_code=NULL;
}
code_pos=0;
line=1; //it is stand-ar-ized that lines begin in 1 in code..
column=0;
tk_rb_pos=0;
error_flag=false;
last_error="";
for(int i=0;i<MAX_LOOKAHEAD+1;i++)
_advance();
}
GDTokenizer::Token GDTokenizer::get_token(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, TK_ERROR);
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, TK_ERROR);
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
return tk_rb[ofs].type;
}
int GDTokenizer::get_token_line(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, -1);
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, -1);
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
return tk_rb[ofs].line;
}
int GDTokenizer::get_token_column(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, -1);
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, -1);
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
return tk_rb[ofs].col;
}
const Variant& GDTokenizer::get_token_constant(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, tk_rb[0].constant);
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, tk_rb[0].constant);
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
ERR_FAIL_COND_V(tk_rb[ofs].type!=TK_CONSTANT,tk_rb[0].constant);
return tk_rb[ofs].constant;
}
StringName GDTokenizer::get_token_identifier(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, StringName());
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, StringName());
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
ERR_FAIL_COND_V(tk_rb[ofs].type!=TK_IDENTIFIER,StringName());
return tk_rb[ofs].identifier;
}
GDFunctions::Function GDTokenizer::get_token_built_in_func(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, GDFunctions::FUNC_MAX);
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, GDFunctions::FUNC_MAX);
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
ERR_FAIL_COND_V(tk_rb[ofs].type!=TK_BUILT_IN_FUNC,GDFunctions::FUNC_MAX);
return tk_rb[ofs].func;
}
Variant::Type GDTokenizer::get_token_type(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, Variant::NIL);
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, Variant::NIL);
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
ERR_FAIL_COND_V(tk_rb[ofs].type!=TK_BUILT_IN_TYPE,Variant::NIL);
return tk_rb[ofs].vtype;
}
int GDTokenizer::get_token_line_indent(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, 0);
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, 0);
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
ERR_FAIL_COND_V(tk_rb[ofs].type!=TK_NEWLINE,0);
return tk_rb[ofs].constant;
}
String GDTokenizer::get_token_error(int p_offset) const {
ERR_FAIL_COND_V( p_offset <= -MAX_LOOKAHEAD, String());
ERR_FAIL_COND_V( p_offset >= MAX_LOOKAHEAD, String());
int ofs = (TK_RB_SIZE + tk_rb_pos + p_offset - MAX_LOOKAHEAD -1)%TK_RB_SIZE;
ERR_FAIL_COND_V(tk_rb[ofs].type!=TK_ERROR,String());
return tk_rb[ofs].constant;
}
void GDTokenizer::advance(int p_amount) {
ERR_FAIL_COND( p_amount <=0 );
for(int i=0;i<p_amount;i++)
_advance();
}