[692] | 1 | /*
|
---|
| 2 | -----------------------------------------------------------------------------
|
---|
| 3 | This source file is part of OGRE
|
---|
| 4 | (Object-oriented Graphics Rendering Engine)
|
---|
| 5 | For the latest info, see http://www.stevestreeting.com/ogre/
|
---|
| 6 |
|
---|
| 7 | Copyright (c) 2000-2005 The OGRE Team
|
---|
| 8 | Also see acknowledgements in Readme.html
|
---|
| 9 |
|
---|
| 10 | This program is free software; you can redistribute it and/or modify it under
|
---|
| 11 | the terms of the GNU General Public License as published by the Free Software
|
---|
| 12 | Foundation; either version 2 of the License, or (at your option) any later
|
---|
| 13 | version.
|
---|
| 14 |
|
---|
| 15 | This program is distributed in the hope that it will be useful, but WITHOUT
|
---|
| 16 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
|
---|
| 17 | FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
|
---|
| 18 |
|
---|
| 19 | You should have received a copy of the GNU General Public License along with
|
---|
| 20 | this program; if not, write to the Free Software Foundation, Inc., 59 Temple
|
---|
| 21 | Place - Suite 330, Boston, MA 02111-1307, USA, or go to
|
---|
| 22 | http://www.gnu.org/copyleft/gpl.html.
|
---|
| 23 | -----------------------------------------------------------------------------
|
---|
| 24 | */
|
---|
| 25 |
|
---|
| 26 |
|
---|
| 27 | #ifndef __Compiler2Pass_H__
|
---|
| 28 | #define __Compiler2Pass_H__
|
---|
| 29 |
|
---|
| 30 | #include <vector>
|
---|
| 31 | #include "OgrePrerequisites.h"
|
---|
| 32 |
|
---|
| 33 | namespace Ogre {
|
---|
| 34 |
|
---|
| 35 | /** Compiler2Pass is a generic 2 pass compiler/assembler
|
---|
| 36 | @remarks
|
---|
| 37 | provides a tokenizer in pass 1 and relies on the subclass to provide the virtual method for pass 2
|
---|
| 38 |
|
---|
| 39 | PASS 1 - tokenize source: this is a simple brute force lexical scanner/analyzer that also parses
|
---|
| 40 | the formed token for proper semantics and context in one pass
|
---|
| 41 | it uses Look Ahead Left-Right (LALR) ruling based on Backus - Naur Form (BNF) notation for semantic
|
---|
| 42 | checking and also performs context checking allowing for language dialects
|
---|
| 43 |
|
---|
| 44 | PASS 2 - generate application specific instructions ie native instructions based on the tokens in the instruction container.
|
---|
| 45 |
|
---|
| 46 | @par
|
---|
| 47 | this class must be subclassed with the subclass providing some implementation details for Pass 2. The subclass
|
---|
| 48 | is responsible for setting up the token libraries along with defining the language syntax and handling
|
---|
| 49 | token actions during the second pass.
|
---|
| 50 |
|
---|
| 51 | @par
|
---|
| 52 | The sub class normally supplies a simplified BNF text description in its constructor prior to doing any parsing/tokenizing of source.
|
---|
| 53 | The simplified BNF text description defines the language syntax and rule structure.
|
---|
| 54 | The meta-symbols used in the BNF text description are:
|
---|
| 55 |
|
---|
| 56 | ::= meaning "is defined as". "::=" starts the definition of a rule. The left side of ::= must contain an <identifier>
|
---|
| 57 |
|
---|
| 58 | <> angle brackets are used to surround syntax rule names. A syntax rule name is also called a non-terminal in that
|
---|
| 59 | it does not generate a terminal token in the instruction container for pass 2 processing.
|
---|
| 60 |
|
---|
| 61 | | meaning "or". if the item on the left of the | fails then the item on the right is tested.
|
---|
| 62 | Example: <true_false> ::= 'true' | 'false';
|
---|
| 63 | whitespace is used to imply AND operation between left and right items.
|
---|
| 64 | Example: <terrain_shadaws> ::= 'terrain_shadows' <true_false>
|
---|
| 65 | the 'terrain_shadows' terminal token must be found and <true_false> rule must pass in order for <terrain_shadows> rule
|
---|
| 66 | to pass.
|
---|
| 67 |
|
---|
| 68 |
|
---|
| 69 | [] optional rule identifier is enclosed in meta symbols [ and ].
|
---|
| 70 | Note that only one identifier or terminal token can take [] modifier.
|
---|
| 71 | {} repetitive identifier (zero or more times) is enclosed in meta symbols { and }
|
---|
| 72 | Note that only one identifier or terminal token can take {} modifier.
|
---|
| 73 | '' terminal tokens are surrounded by single quotes. A terminal token is always one or more characters.
|
---|
| 74 | For example: 'Colour' defines a character sequence that must be matched in whole. Note that matching is case
|
---|
| 75 | sensitive.
|
---|
| 76 | @ turn on single character scanning and don't skip white space.
|
---|
| 77 | Mainly used for label processing that allow white space.
|
---|
| 78 | Example: '@ ' prevents the white space between the quotes from being skipped
|
---|
| 79 | -'' no terminal token is generated when a - precedes the first single quote but the text in between the quotes is still
|
---|
| 80 | tested against the characters in the source being parsed.
|
---|
| 81 | (?! ) negative lookahead (not test) inspired by Perl 5. Scans ahead for a non-terminal or terminal expression
|
---|
| 82 | that should fail in order to make the rule production pass.
|
---|
| 83 | Does not generate a token or advance the cursur. If the lookahead result fails ie token is found,
|
---|
| 84 | then the current rule fails and rollback occurs. Mainly used to solve multiple contexts of a token.
|
---|
| 85 | An Example of where not test is used to solve multiple contexts:
|
---|
| 86 |
|
---|
| 87 | <rule> ::= <identifier> "::=" <expression>
|
---|
| 88 | <expression> ::= <and_term> { <or_term> }
|
---|
| 89 | <or_term> ::= "|" <and_term>
|
---|
| 90 | <and_term> ::= <term> { <term> }
|
---|
| 91 | <term> ::= <identifier_right> | <terminal_symbol> | <repeat_expression> | <optional_expression>
|
---|
| 92 | <identifier_right> ::= <identifier> (?!"::=")
|
---|
| 93 |
|
---|
| 94 | <identiefier> appears on both sides of the ::= so (?!"::=") test to make sure that ::= is not on the
|
---|
| 95 | right which would indicate that a new rule was being formed.
|
---|
| 96 |
|
---|
| 97 | Works on both terminals and non-terminals.
|
---|
| 98 | Note: lookahead failure cause the whole rule to fail and rollback to occur
|
---|
| 99 |
|
---|
| 100 | <#name> # indicates that a numerical value is to be parsed to form a terminal token. Name is optional and is just a descriptor
|
---|
| 101 | to help with understanding what the value will be used for.
|
---|
| 102 | Example: <Colour> ::= <#red> <#green> <#blue>
|
---|
| 103 |
|
---|
| 104 | () parentheses enclose a set of characters that can be used to generate a user identifier. for example:
|
---|
| 105 | (0123456789) matches a single character found in that set.
|
---|
| 106 | An example of a user identifier:
|
---|
| 107 |
|
---|
| 108 | <Label> ::= <Character> {<Character>}
|
---|
| 109 | <Character> ::= (abcdefghijklmnopqrstuvwxyz)
|
---|
| 110 |
|
---|
| 111 | This will generate a rule that accepts one or more lowercase letters to make up the Label. The User identifier
|
---|
| 112 | stops collecting the characters into a string when a match cannot be found in the rule.
|
---|
| 113 | */
|
---|
| 114 | class _OgreExport Compiler2Pass
|
---|
| 115 | {
|
---|
| 116 |
|
---|
| 117 | protected:
|
---|
| 118 |
|
---|
| 119 | // BNF operation types
|
---|
| 120 | enum OperationType {otUNKNOWN, otRULE, otAND, otOR, otOPTIONAL, otREPEAT, otDATA, otNOT_TEST, otEND};
|
---|
| 121 |
|
---|
| 122 | /** structure used to build rule paths
|
---|
| 123 |
|
---|
| 124 | */
|
---|
| 125 | struct TokenRule
|
---|
| 126 | {
|
---|
| 127 | OperationType operation;
|
---|
| 128 | size_t tokenID;
|
---|
| 129 |
|
---|
| 130 | TokenRule(void) : operation(otUNKNOWN), tokenID(0) {}
|
---|
| 131 | TokenRule(const OperationType ot, const size_t token)
|
---|
| 132 | : operation(ot), tokenID(token) {}
|
---|
| 133 | };
|
---|
| 134 |
|
---|
| 135 | typedef std::vector<TokenRule> TokenRuleContainer;
|
---|
| 136 | typedef TokenRuleContainer::iterator TokenRuleIterator;
|
---|
| 137 |
|
---|
| 138 | static const size_t SystemTokenBase = 1000;
|
---|
| 139 | enum SystemRuleToken {
|
---|
| 140 | _no_token_ = SystemTokenBase,
|
---|
| 141 | _character_,
|
---|
| 142 | _value_,
|
---|
| 143 | _no_space_skip_
|
---|
| 144 | };
|
---|
| 145 |
|
---|
| 146 | enum BNF_ID {BNF_UNKOWN = 0,
|
---|
| 147 | BNF_SYNTAX, BNF_RULE, BNF_IDENTIFIER, BNF_IDENTIFIER_RIGHT, BNF_IDENTIFIER_CHARACTERS, BNF_ID_BEGIN, BNF_ID_END,
|
---|
| 148 | BNF_CONSTANT_BEGIN, BNF_SET_RULE, BNF_EXPRESSION,
|
---|
| 149 | BNF_AND_TERM, BNF_OR_TERM, BNF_TERM, BNF_TERM_ID, BNF_CONSTANT, BNF_OR, BNF_TERMINAL_SYMBOL, BNF_TERMINAL_START,
|
---|
| 150 | BNF_REPEAT_EXPRESSION, BNF_REPEAT_BEGIN, BNF_REPEAT_END, BNF_SET, BNF_SET_BEGIN, BNF_SET_END,
|
---|
| 151 | BNF_NOT_TEST, BNF_NOT_TEST_BEGIN, BNF_OPTIONAL_EXPRESSION, BNF_NOT_EXPRESSION, BNF_NOT_CHK,
|
---|
| 152 | BNF_OPTIONAL_BEGIN, BNF_OPTIONAL_END, BNF_NO_TOKEN_START, BNF_SINGLEQUOTE, BNF_SINGLE_QUOTE_EXC, BNF_SET_END_EXC,
|
---|
| 153 | BNF_ANY_CHARACTER, BNF_SPECIAL_CHARACTERS1,
|
---|
| 154 | BNF_SPECIAL_CHARACTERS2, BNF_WHITE_SPACE_CHK,
|
---|
| 155 |
|
---|
| 156 | BNF_LETTER, BNF_LETTER_DIGIT, BNF_DIGIT, BNF_WHITE_SPACE,
|
---|
| 157 | BNF_ALPHA_SET, BNF_NUMBER_SET, BNF_SPECIAL_CHARACTER_SET1,
|
---|
| 158 | BNF_SPECIAL_CHARACTER_SET2, BNF_SPECIAL_CHARACTER_SET3, BNF_NOT_CHARS
|
---|
| 159 | };
|
---|
| 160 |
|
---|
| 161 |
|
---|
| 162 | /** structure used to build lexeme Type library */
|
---|
| 163 | struct LexemeTokenDef
|
---|
| 164 | {
|
---|
| 165 | size_t ID; /// Token ID which is the index into the Lexeme Token Definition Container
|
---|
| 166 | bool hasAction; /// has an action associated with it. only applicable to terminal tokens
|
---|
| 167 | bool isNonTerminal; /// if true then token is non-terminal
|
---|
| 168 | size_t ruleID; /// index into Rule database for non-terminal token rulepath and lexeme
|
---|
| 169 | bool isCaseSensitive; /// if true use case sensitivity when comparing lexeme to source
|
---|
| 170 | String lexeme; /// text representation of token or valid characters for label parsing
|
---|
| 171 |
|
---|
| 172 | LexemeTokenDef(void) : ID(0), hasAction(false), isNonTerminal(false), ruleID(0), isCaseSensitive(false) {}
|
---|
| 173 | LexemeTokenDef( const size_t ID, const String& lexeme, const bool hasAction = false, const bool caseSensitive = false )
|
---|
| 174 | : ID(ID)
|
---|
| 175 | , hasAction(hasAction)
|
---|
| 176 | , isNonTerminal(false)
|
---|
| 177 | , ruleID(0)
|
---|
| 178 | , isCaseSensitive(caseSensitive)
|
---|
| 179 | , lexeme(lexeme)
|
---|
| 180 | {
|
---|
| 181 | }
|
---|
| 182 |
|
---|
| 183 | };
|
---|
| 184 |
|
---|
| 185 | typedef std::vector<LexemeTokenDef> LexemeTokenDefContainer;
|
---|
| 186 | typedef LexemeTokenDefContainer::iterator LexemeTokenDefIterator;
|
---|
| 187 |
|
---|
| 188 | typedef std::map<std::string, size_t> LexemeTokenMap;
|
---|
| 189 | typedef LexemeTokenMap::iterator TokenKeyIterator;
|
---|
| 190 | /// map used to lookup client token based on previously defined lexeme
|
---|
| 191 |
|
---|
| 192 |
|
---|
| 193 | /** structure for Token instructions that are constructed during first pass*/
|
---|
| 194 | struct TokenInst
|
---|
| 195 | {
|
---|
| 196 | size_t NTTRuleID; /// Non-Terminal Token Rule ID that generated Token
|
---|
| 197 | size_t tokenID; /// expected Token ID. Could be UNKNOWN if valid token was not found.
|
---|
| 198 | size_t line; /// line number in source code where Token was found
|
---|
| 199 | size_t pos; /// Character position in source where Token was found
|
---|
| 200 | bool found; /// is true if expected token was found
|
---|
| 201 | };
|
---|
| 202 |
|
---|
| 203 | typedef std::vector<TokenInst> TokenInstContainer;
|
---|
| 204 | typedef TokenInstContainer::iterator TokenInstIterator;
|
---|
| 205 |
|
---|
| 206 | // token que, definitions, rules
|
---|
| 207 | struct TokenState
|
---|
| 208 | {
|
---|
| 209 | TokenInstContainer tokenQue;
|
---|
| 210 | LexemeTokenDefContainer lexemeTokenDefinitions;
|
---|
| 211 | TokenRuleContainer rootRulePath;
|
---|
| 212 | LexemeTokenMap lexemeTokenMap;
|
---|
| 213 | };
|
---|
| 214 |
|
---|
| 215 | TokenState* mClientTokenState;
|
---|
| 216 |
|
---|
| 217 | /// Active token que, definitions, rules currntly being used by parser
|
---|
| 218 | TokenState* mActiveTokenState;
|
---|
| 219 | /// the location within the token instruction container where pass 2 is
|
---|
| 220 | size_t mPass2TokenQuePosition;
|
---|
| 221 | /** the que position of the previous token that had an action.
|
---|
| 222 | A token's action is fired on the next token having an action.
|
---|
| 223 | */
|
---|
| 224 | size_t mPreviousActionQuePosition;
|
---|
| 225 |
|
---|
| 226 | /// pointer to the source to be compiled
|
---|
| 227 | const String* mSource;
|
---|
| 228 | /// name of the source to be compiled
|
---|
| 229 | String mSourceName;
|
---|
| 230 | size_t mEndOfSource;
|
---|
| 231 |
|
---|
| 232 | size_t mCurrentLine; /// current line number in source being tokenized
|
---|
| 233 | size_t mCharPos; /// position in current line in source being tokenized
|
---|
| 234 |
|
---|
| 235 | /// storage container for constants defined in source
|
---|
| 236 | /// container uses Token index as a key associated with a float constant
|
---|
| 237 | std::map<size_t, float> mConstants;
|
---|
| 238 | /// storage container for string labels defined in source
|
---|
| 239 | /// container uses Token index as a key associated with a label
|
---|
| 240 | std::map<size_t, String> mLabels;
|
---|
| 241 | /// flag indicates when a label is being parsed.
|
---|
| 242 | /// It gets set false when a terminal token not of _character_ is encountered
|
---|
| 243 | bool mLabelIsActive;
|
---|
| 244 | /// the key of the active label being built during pass 1.
|
---|
| 245 | /// a new key is calculated when mLabelIsActive switches from false to true
|
---|
| 246 | size_t mActiveLabelKey;
|
---|
| 247 | /// flag being true indicates that spaces are not to be skipped
|
---|
| 248 | /// automatically gets set to false when mLabelIsActive goes to false
|
---|
| 249 | bool mNoSpaceSkip;
|
---|
| 250 | /// if flag is true then next terminal token is not added to token que if found
|
---|
| 251 | /// but does effect rule path flow
|
---|
| 252 | bool mNoTerminalToken;
|
---|
| 253 |
|
---|
| 254 | /// Active Contexts pattern used in pass 1 to determine which tokens are valid for a certain context
|
---|
| 255 | uint mActiveContexts;
|
---|
| 256 |
|
---|
| 257 | /** perform pass 1 of compile process
|
---|
| 258 | scans source for lexemes that can be tokenized and then
|
---|
| 259 | performs general semantic and context verification on each lexeme before it is tokenized.
|
---|
| 260 | A tokenized instruction list is built to be used by Pass 2.
|
---|
| 261 | A rule path can trigger Pass 2 execution if enough tokens have been generated in Pass 1.
|
---|
| 262 | Pass 1 will then pass control to pass 2 temporarily until the current tokens have been consumed.
|
---|
| 263 |
|
---|
| 264 | */
|
---|
| 265 | bool doPass1();
|
---|
| 266 |
|
---|
| 267 | /** performs Pass 2 of compile process which is execution of the tokens
|
---|
| 268 | @remark
|
---|
| 269 | Pass 2 takes the token instructions generated in Pass 1 and
|
---|
| 270 | builds the application specific instructions along with verifying
|
---|
| 271 | symantic and context rules that could not be checked in Pass 1.
|
---|
| 272 | @par
|
---|
| 273 | Pass 2 execution consumes tokens and moves the Pass 2 token instruction position towards the end
|
---|
| 274 | of the token container. Token execution can insert new tokens into the token container.
|
---|
| 275 | */
|
---|
| 276 | bool doPass2();
|
---|
| 277 |
|
---|
| 278 | /** execute the action associated with the token pointed to by the Pass 2 token instruction position.
|
---|
| 279 | Its upto the child class to implement how it will associate a token key with and action.
|
---|
| 280 | Actions should be placed at positions withing the BNF grammer (instruction que) that indicate
|
---|
| 281 | enough tokens exist for pass 2 processing to take place.
|
---|
| 282 | */
|
---|
| 283 | virtual void executeTokenAction(const size_t tokenID) = 0;
|
---|
| 284 | /** setup client token definitions. Gets called when BNF grammer is being setup.
|
---|
| 285 | */
|
---|
| 286 | virtual void setupTokenDefinitions(void) = 0;
|
---|
| 287 | /** Gets the next token from the instruction que. If an unkown token is found then an exception is raised but
|
---|
| 288 | the instruction pointer is still moved passed the unknown token. The subclass should catch the exception,
|
---|
| 289 | provide an error message, and attempt recovery.
|
---|
| 290 |
|
---|
| 291 | @param expectedTokenID if set then if tokenID does not match then an exception is raised
|
---|
| 292 | */
|
---|
| 293 | const TokenInst& getNextToken(const size_t expectedTokenID = 0);
|
---|
| 294 | /** Gets the current token from the instruction que.
|
---|
| 295 | */
|
---|
| 296 | const TokenInst& getCurrentToken(void);
|
---|
| 297 | /** If a next token instruction exist then test if its token ID matches.
|
---|
| 298 | This method is usefull for peeking ahead during pass 2 to see if a certain
|
---|
| 299 | token exists.
|
---|
| 300 | @param expectedTokenID is the ID of the token to match.
|
---|
| 301 | */
|
---|
| 302 | bool testNextTokenID(const size_t expectedTokenID);
|
---|
| 303 | /**
|
---|
| 304 | */
|
---|
| 305 | void replaceToken(void);
|
---|
| 306 | /** Gets the next token's associated floating point value in the instruction que that was parsed from the
|
---|
| 307 | text source. If an unkown token is found or no associated value was found then an exception is raised but
|
---|
| 308 | the instruction pointer is still moved passed the unknown token. The subclass should catch the exception,
|
---|
| 309 | provide an error message, and attempt recovery.
|
---|
| 310 | */
|
---|
| 311 | float getNextTokenValue(void);
|
---|
| 312 | /** Gets the next token's associated text label in the instruction que that was parsed from the
|
---|
| 313 | text source. If an unkown token is found or no associated label was found then an exception is raised but
|
---|
| 314 | the instruction pointer is still moved passed the unknown token. The subclass should catch the exception,
|
---|
| 315 | provide an error message, and attempt recovery.
|
---|
| 316 | */
|
---|
| 317 | const String& getNextTokenLabel(void);
|
---|
| 318 | /** Gets the number of tokens waiting in the instruction que that need to be processed by an token action in pass 2.
|
---|
| 319 | */
|
---|
| 320 | size_t getPass2TokenQueCount(void) const;
|
---|
| 321 | /** Get the number of tokens not processed by action token.
|
---|
| 322 | Client Actions should use this method to retreive the number of parameters(tokens)
|
---|
| 323 | remaining to be processed in the action.
|
---|
| 324 | */
|
---|
| 325 | size_t getRemainingTokensForAction(void) const;
|
---|
| 326 |
|
---|
| 327 | /** Add a lexeme token association. The backend compiler uses the associations between lexeme
|
---|
| 328 | and token when building the rule base from the BNF script so all associations must be done
|
---|
| 329 | prior to compiling a source.
|
---|
| 330 | @param lexeme is the name of the token and use when parsing the source to determin a match for a token.
|
---|
| 331 | @param token is the ID associated with the lexeme
|
---|
| 332 | @param hasAction must be set true if the client wants an action triggered when this token is generated
|
---|
| 333 | @param caseSensitive should be set true if lexeme match should use case sensitivity
|
---|
| 334 | */
|
---|
| 335 | void addLexemeToken(const String& lexeme, const size_t token, const bool hasAction = false, const bool caseSensitive = false);
|
---|
| 336 |
|
---|
| 337 | /** sets up the parser rules for the client based on the BNF Grammer text passed in.
|
---|
| 338 | Raises an exception if the grammer did not compile successfully. This method gets called
|
---|
| 339 | when a call to compile occurs and no compiled BNF grammer exists, otherwise nothing will happen since the compiler has no rules to work
|
---|
| 340 | with. The grammer only needs to be set once during the lifetime of the compiler unless the
|
---|
| 341 | grammer changes.
|
---|
| 342 | BNF Grammer rules are cached once the BNF grammer source is compiled.
|
---|
| 343 | The client should never have to call this method directly.
|
---|
| 344 | */
|
---|
| 345 | void setClientBNFGrammer(void);
|
---|
| 346 |
|
---|
| 347 |
|
---|
| 348 |
|
---|
| 349 | /// find the eol charater
|
---|
| 350 | void findEOL();
|
---|
| 351 |
|
---|
| 352 | /** check to see if the text at the present position in the source is a numerical constant
|
---|
| 353 | @param fvalue is a reference that will receive the float value that is in the source
|
---|
| 354 | @param charsize reference to receive number of characters that make of the value in the source
|
---|
| 355 | @return
|
---|
| 356 | true if characters form a valid float representation
|
---|
| 357 | false if a number value could not be extracted
|
---|
| 358 | */
|
---|
| 359 | bool isFloatValue(float& fvalue, size_t& charsize) const;
|
---|
| 360 |
|
---|
| 361 | /** Check if source at current position is supposed to be a user defined character label.
|
---|
| 362 | A new label is processed when previous operation was not _character_ otherwise the processed
|
---|
| 363 | character (if match was found) is added to the current label. This allows _character_ operations
|
---|
| 364 | to be chained together to form a crude regular expression to build a label.
|
---|
| 365 | @param rulepathIDX index into rule path database of token to validate.
|
---|
| 366 | @return
|
---|
| 367 | true if token was found for character label.
|
---|
| 368 | */
|
---|
| 369 | bool isCharacterLabel(const size_t rulepathIDX);
|
---|
| 370 | /** check to see if the text is in the lexeme text library
|
---|
| 371 | @param lexeme points to begining of text where a lexem token might exist
|
---|
| 372 | @param caseSensitive set to true if match should be case sensitive
|
---|
| 373 | @return
|
---|
| 374 | true if a matching token could be found in the token type library
|
---|
| 375 | false if could not be tokenized
|
---|
| 376 | */
|
---|
| 377 | bool isLexemeMatch(const String& lexeme, const bool caseSensitive) const;
|
---|
| 378 | /// position to the next possible valid sysmbol
|
---|
| 379 | bool positionToNextLexeme();
|
---|
| 380 | /** process input source text using rulepath to determine allowed tokens
|
---|
| 381 | @remarks
|
---|
| 382 | the method is reentrant and recursive
|
---|
| 383 | if a non-terminal token is encountered in the current rule path then the method is
|
---|
| 384 | called using the new rule path referenced by the non-terminal token
|
---|
| 385 | Tokens can have the following operation states which effects the flow path of the rule
|
---|
| 386 | RULE: defines a rule path for the non-terminal token
|
---|
| 387 | AND: the token is required for the rule to pass
|
---|
| 388 | OR: if the previous tokens failed then try these ones
|
---|
| 389 | OPTIONAL: the token is optional and does not cause the rule to fail if the token is not found
|
---|
| 390 | REPEAT: the token is required but there can be more than one in a sequence
|
---|
| 391 | DATA: Used by a previous token ie for character sets
|
---|
| 392 | NOTTEST: performs negative lookahead ie make sure the next token is not of a certain type
|
---|
| 393 | END: end of the rule path - the method returns the succuss of the rule
|
---|
| 394 |
|
---|
| 395 | @param rulepathIDX index into an array of Token Rules that define a rule path to be processed
|
---|
| 396 | @return
|
---|
| 397 | true if rule passed - all required tokens found
|
---|
| 398 | false if one or more tokens required to complete the rule were not found
|
---|
| 399 | */
|
---|
| 400 | bool processRulePath( size_t rulepathIDX);
|
---|
| 401 |
|
---|
| 402 |
|
---|
| 403 | /** setup ActiveContexts - should be called by subclass to setup initial language contexts
|
---|
| 404 | */
|
---|
| 405 | void setActiveContexts(const uint contexts){ mActiveContexts = contexts; }
|
---|
| 406 |
|
---|
| 407 | /// comment specifiers are hard coded
|
---|
| 408 | void skipComments();
|
---|
| 409 |
|
---|
| 410 | /// find end of line marker and move past it
|
---|
| 411 | void skipEOL();
|
---|
| 412 |
|
---|
| 413 | /// skip all the white space which includes spaces and tabs
|
---|
| 414 | void skipWhiteSpace();
|
---|
| 415 |
|
---|
| 416 |
|
---|
| 417 | /** check if current position in source has the lexeme text equivalent to the TokenID
|
---|
| 418 | @param rulepathIDX index into rule path database of token to validate
|
---|
| 419 | @param activeRuleID index of non-terminal rule that generated the token
|
---|
| 420 | @return
|
---|
| 421 | true if token was found
|
---|
| 422 | false if token lexeme text does not match the source text
|
---|
| 423 | if token is non-terminal then processRulePath is called
|
---|
| 424 | */
|
---|
| 425 | bool ValidateToken(const size_t rulepathIDX, const size_t activeRuleID);
|
---|
| 426 |
|
---|
| 427 | /** scan through all the rules and initialize token definition with index to rules for non-terminal tokens.
|
---|
| 428 | Gets called when internal grammer is being verified or after client grammer has been parsed.
|
---|
| 429 | @param grammerName is the name of the grammer the token rules are for
|
---|
| 430 | */
|
---|
| 431 | void verifyTokenRuleLinks(const String& grammerName);
|
---|
| 432 | /** Checks the last token instruction and if it has an action then it triggers the action of the previously
|
---|
| 433 | found token having an action.
|
---|
| 434 | */
|
---|
| 435 | void checkTokenActionTrigger(void);
|
---|
| 436 | /** Get the text representation of the rule path. This is a good way to way to visually verify
|
---|
| 437 | that the BNF grammer did compile correctly.
|
---|
| 438 | @param ruleID is the index into the rule path.
|
---|
| 439 | */
|
---|
| 440 | String getBNFGrammerTextFromRulePath(size_t ruleID);
|
---|
| 441 |
|
---|
| 442 |
|
---|
| 443 | private:
|
---|
| 444 | // used for interpreting BNF script
|
---|
| 445 | // keep it as static so that only one structure is created
|
---|
| 446 | // no matter how many times this class is instantiated.
|
---|
| 447 | static TokenState mBNFTokenState;
|
---|
| 448 | // maintain a map of BNF grammer
|
---|
| 449 | typedef std::map<String, TokenState> TokenStateContainer;
|
---|
| 450 | static TokenStateContainer mClientTokenStates;
|
---|
| 451 | /// if a previous token action was setup then activate it now
|
---|
| 452 | void activatePreviousTokenAction(void);
|
---|
| 453 | /// initialize token definitions and rule paths
|
---|
| 454 | void initBNFCompiler(void);
|
---|
| 455 | /// Convert BNF grammer token que created in pass 1 into a BNF rule path
|
---|
| 456 | void buildClientBNFRulePaths(void);
|
---|
| 457 | /// modify the last rule in the container. An end operation is added to the rule path.
|
---|
| 458 | void modifyLastRule(const OperationType pendingRuleOp, const size_t tokenID);
|
---|
| 459 | /** get the token ID for a lexeme in the client state. If the lexeme is not found then
|
---|
| 460 | it is added to the map and definition container and a new tokenID created.
|
---|
| 461 | @return the ID of the token.
|
---|
| 462 | */
|
---|
| 463 | size_t getClientLexemeTokenID(const String& lexeme, const bool isCaseSensitive = false);
|
---|
| 464 | /// Extract a Non Terminal identifier from the token que
|
---|
| 465 | void extractNonTerminal(const OperationType pendingRuleOp);
|
---|
| 466 | /// Extract a Terminal lexeme from the token que and add to current rule expression
|
---|
| 467 | void extractTerminal(const OperationType pendingRuleOp, const bool notoken = false);
|
---|
| 468 | /// Extract a set from the token que and add to current rule expression
|
---|
| 469 | void extractSet(const OperationType pendingRuleOp);
|
---|
| 470 | /// Extract a numeric constant from the token que and add it to the current rule expression
|
---|
| 471 | void extractNumericConstant(const OperationType pendingRuleOp);
|
---|
| 472 | String getLexemeText(size_t& ruleID);
|
---|
| 473 |
|
---|
| 474 | public:
|
---|
| 475 |
|
---|
| 476 | /// constructor
|
---|
| 477 | Compiler2Pass();
|
---|
| 478 | virtual ~Compiler2Pass() {}
|
---|
| 479 |
|
---|
| 480 | /** compile the source - performs 2 passes.
|
---|
| 481 | First pass is to tokinize, check semantics and context.
|
---|
| 482 | The second pass is performed by using tokens to look up function implementors and executing
|
---|
| 483 | them which convert tokens to application specific instructions.
|
---|
| 484 | @remark
|
---|
| 485 | Pass 2 only gets executed if Pass 1 has built enough tokens to complete a rule path and found no errors
|
---|
| 486 | @param source a pointer to the source text to be compiled
|
---|
| 487 | @return
|
---|
| 488 | true if Pass 1 and Pass 2 are successfull
|
---|
| 489 | false if any errors occur in Pass 1 or Pass 2
|
---|
| 490 | */
|
---|
| 491 | bool compile(const String& source, const String& sourceName);
|
---|
| 492 | /** gets BNF Grammer. Gets called when BNF grammer has to be compiled for the first time.
|
---|
| 493 | */
|
---|
| 494 | virtual const String& getClientBNFGrammer(void) = 0;
|
---|
| 495 |
|
---|
| 496 | /** get the name of the BNF grammer.
|
---|
| 497 | */
|
---|
| 498 | virtual const String& getClientGrammerName(void) = 0;
|
---|
| 499 |
|
---|
| 500 | };
|
---|
| 501 |
|
---|
| 502 | }
|
---|
| 503 |
|
---|
| 504 | #endif
|
---|
| 505 |
|
---|