/* ScummVM - Graphic Adventure Engine * * ScummVM is the legal property of its developers, whose names * are too numerous to list here. Please refer to the COPYRIGHT * file distributed with this source distribution. * * This program 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. * This program 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 this program; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * * $URL$ * $Id$ * */ #ifndef SCI_SCICORE_VOCABULARY_H #define SCI_SCICORE_VOCABULARY_H #include "common/str.h" #include "sci/scicore/versions.h" namespace Sci { class ResourceManager; /*#define VOCABULARY_DEBUG */ /*#define SCI_SIMPLE_SAID_CODE */ /* Whether the simplified Said() matching should be used */ /*#define SCI_SIMPLE_SAID_DEBUG */ /* uncomment to enable simple said debugging */ #define SCRIPT_UNKNOWN_FUNCTION_STRING "[Unknown]" /* The string used to identify the "unknown" SCI0 function for each game */ #define PARSE_HEAP_SIZE 64 /* Number of bytes allocated on the heap to store bad words if parsing fails */ struct opcode { int type; int number; char* name; }; #define VOCAB_RESOURCE_OPCODES 998 #define VOCAB_RESOURCE_KNAMES 999 #define VOCAB_RESOURCE_SCI0_MAIN_VOCAB 0 #define VOCAB_RESOURCE_SCI0_PARSE_TREE_BRANCHES 900 #define VOCAB_RESOURCE_SCI0_SUFFIX_VOCAB 901 #define VOCAB_RESOURCE_SCI1_MAIN_VOCAB 900 #define VOCAB_RESOURCE_SCI1_PARSE_TREE_BRANCHES 901 #define VOCAB_RESOURCE_SCI1_SUFFIX_VOCAB 902 #define VOCAB_RESOURCE_SCI1_CHAR_TRANSFORMS 913 #define VOCAB_CLASS_PREPOSITION 0x01 #define VOCAB_CLASS_ARTICLE 0x02 #define VOCAB_CLASS_ADJECTIVE 0x04 #define VOCAB_CLASS_PRONOUN 0x08 #define VOCAB_CLASS_NOUN 0x10 #define VOCAB_CLASS_INDICATIVE_VERB 0x20 #define VOCAB_CLASS_ADVERB 0x40 #define VOCAB_CLASS_IMPERATIVE_VERB 0x80 #define VOCAB_CLASS_NUMBER 0x001 extern const char *class_names[]; /* Vocabulary class names */ #define VOCAB_CLASS_ANYWORD 0xff /* Anywords are ignored by the parser */ #define VOCAB_MAGIC_NUMBER_GROUP 0xffd /* 0xffe ? */ /* This word class is used for numbers */ #define VOCAB_TREE_NODES 500 /* Number of nodes for each parse_tree_node structure */ #define VOCAB_TREE_NODE_LAST_WORD_STORAGE 0x140 #define VOCAB_TREE_NODE_COMPARE_TYPE 0x146 #define VOCAB_TREE_NODE_COMPARE_GROUP 0x14d #define VOCAB_TREE_NODE_FORCE_STORAGE 0x154 #define SAID_COMMA 0xf0 #define SAID_AMP 0xf1 #define SAID_SLASH 0xf2 #define SAID_PARENO 0xf3 #define SAID_PARENC 0xf4 #define SAID_BRACKO 0xf5 #define SAID_BRACKC 0xf6 #define SAID_HASH 0xf7 #define SAID_LT 0xf8 #define SAID_GT 0xf9 #define SAID_TERM 0xff #define SAID_FIRST SAID_COMMA /* There was no 'last matching word': */ #define SAID_FULL_MATCH 0xffff #define SAID_NO_MATCH 0xfffe #define SAID_PARTIAL_MATCH 0xfffd #define SAID_LONG(x) ((x) << 8) struct word_t { int w_class; /* Word class */ int group; /* Word group */ char word[1]; /* The actual word */ }; struct parse_rule_t { int id; /* non-terminal ID */ int first_special; /* first terminal or non-terminal */ int specials_nr; /* number of terminals and non-terminals */ int length; int data[1]; /* actual data (size 1 to avoid compiler warnings) */ }; struct parse_rule_list_t { int terminal; /* Terminal character this rule matches against or 0 for a non-terminal rule */ parse_rule_t *rule; parse_rule_list_t *next; }; struct suffix_t { int class_mask; /* the word class this suffix applies to */ int result_class; /* the word class a word is morphed to if it doesn't fail this check */ int alt_suffix_length; /* String length of the suffix */ int word_suffix_length; /* String length of the other suffix */ char *alt_suffix; /* The alternative suffix */ char *word_suffix; /* The suffix as used in the word vocabulary */ }; struct result_word_t { int w_class; /* Word class */ int group; /* Word group */ }; struct synonym_t { int replaceant; /* The word group to replace */ int replacement; /* The replacement word group for this one */ }; struct parse_tree_branch_t { int id; int data[10]; }; #define PARSE_TREE_NODE_LEAF 0 #define PARSE_TREE_NODE_BRANCH 1 struct parse_tree_node_t { short type; /* leaf or branch */ union { int value; /* For leaves */ short branches[2]; /* For branches */ } content; }; /*FIXME: These need freeing functions...*/ int *vocabulary_get_classes(ResourceManager *resmgr, int *count); int vocabulary_get_class_count(ResourceManager *resmgr); /** * Fills the given StringList with selector names. * Returns true upon success, false oterwise. */ bool vocabulary_get_snames(ResourceManager *resmgr, sci_version_t version, Common::StringList &selectorNames); /* Look up a selector name in an array, return the index */ int vocabulary_lookup_sname(const Common::StringList &selectorNames, const char *sname); /** * Returns a null terminated array of opcodes. */ opcode *vocabulary_get_opcodes(ResourceManager *resmgr); void vocabulary_free_opcodes(opcode *opcodes); /* Frees a previously allocated list of opcodes ** Parameters: (opcode *) opcodes: Opcodes to free ** Returns : (void) */ /** * Returns a null terminated array of kernel function names. * * This function reads the kernel function name table from resource_map, * and returns a null terminated array of deep copies of them. * The returned array has the same format regardless of the format of the * name table of the resource (the format changed between version 0 and 1). */ char **vocabulary_get_knames(ResourceManager *resmgr, int* count); void vocabulary_free_knames(char** names); word_t **vocab_get_words(ResourceManager *resmgr, int *word_counter); /* Gets all words from the main vocabulary ** Parameters: (ResourceManager *) resmr: The resource manager to read from ** (int *) word_counter: The int which the number of words is stored in ** Returns : (word_t **): A list of all words, dynamically allocated */ void vocab_free_words(word_t **words, int words_nr); /* Frees memory allocated by vocab_get_words ** Parameters: (word_t **) words: The words to free ** (int) words_nr: Number of words in the structure ** Returns : (void) */ suffix_t **vocab_get_suffices(ResourceManager *resmgr, int *suffices_nr); /* Gets all suffixes from the suffix vocabulary ** Parameters: (ResourceManager*) resmgr: Resource manager the resources are ** read from ** (int *) suffices_nr: The variable to store the number of suffices in ** Returns : (suffix_t **): A list of suffixes */ void vocab_free_suffices(ResourceManager *resmgr, suffix_t **suffices, int suffices_nr); /* Frees suffices_nr suffices ** Parameters: (ResourceManager *) resmgr: The resource manager to free from ** (suffix_t **) suffices: The suffixes to free ** (int) suffices_nr: Number of entrie sin suffices ** Returns : (void) */ parse_tree_branch_t *vocab_get_branches(ResourceManager *resmgr, int *branches_nr); /* Retrieves all grammar rules from the resource data ** Parameters: (ResourceManager*) resmgr: Resource manager the rules are ** read from ** (int *) branches_nr: Pointer to the variable which the number of entries is to be ** stored in ** Returns : (parse_tree_branch_t *): The rules, or NULL on error */ void vocab_free_branches(parse_tree_branch_t *parser_branches); /* Frees all branches ** Parameters: (parse_tree_branch_t *) parser_branches: The branches to free ** Returns : (null) */ result_word_t *vocab_lookup_word(char *word, int word_len, word_t **words, int words_nr, suffix_t **suffices, int suffices_nr); /* Looks up a single word in the words and suffixes list ** Parameters: (char *) word: Pointer to the word to look up ** (int) word_len: Length of the word to look up ** (word_t **) words: List of words ** (int) words_nr: Number of elements in 'words' ** (suffix_t **) suffices: List of suffices ** (int) suffices_nr: Number of entries in 'suffices' ** Returns : (result_word_t *) A malloc'd result_word_t, or NULL if the word ** could not be found. */ result_word_t *vocab_tokenize_string(char *sentence, int *result_nr, word_t **words, int words_nr, suffix_t **suffices, int suffices_nr, char **error); /* Tokenizes a string and compiles it into word_ts. ** Parameters: (char *) sentence: The sentence to examine ** (int *) result_nr: The variable to store the resulting number of words in ** (word_t **) words: The words to scan for ** (int) words_nr: Number of words to scan for ** (suffix_t **) suffices: suffixes to scan for ** (int) suffices_nr: Number of suffices to scan for ** (char **) error: Points to a malloc'd copy of the offending text or to NULL on error ** Returns : (word_t *): A list of word_ts containing the result, or NULL. ** On error, NULL is returned. If *error is NULL, the sentence did not contain any useful words; ** if not, *error points to a malloc'd copy of the offending word. ** The returned list may contain anywords. */ parse_rule_list_t *vocab_build_gnf(parse_tree_branch_t *branches, int branches_nr); /* Constructs the Greibach Normal Form of the grammar supplied in 'branches' ** Parameters: (parse_tree_branch_t *) branches: The parser's branches ** (int) branches_nr: Number of parser branches ** Returns : (parse_rule_list_t *): Pointer to a list of singly linked ** GNF rules describing the same language ** that was described by 'branches' ** The original SCI rules are in almost-CNF (Chomsky Normal Form). Note that ** branch[0] is used only for a few magical incantations, as it is treated ** specially by the SCI parser. */ void vocab_free_rule_list(parse_rule_list_t *rule_list); /* Frees a parser rule list as returned by vocab_build_gnf() ** Parameters: (parse_rule_list_t *) rule_list: The rule list to free ** Returns : (void) */ int vocab_build_parse_tree(parse_tree_node_t *nodes, result_word_t *words, int words_nr, parse_tree_branch_t *branch0, parse_rule_list_t *rules); /* Builds a parse tree from a list of words ** Parameters: (parse_tree_node_t *) nodes: A node list to store the tree in (must have ** at least VOCAB_TREE_NODES entries) ** (result_word_t *) words: The words to build the tree from ** (int) words_nr: The number of words ** (parse_tree_branch_t *) branche0: The zeroeth original branch of the ** original CNF parser grammar ** (parse_rule_list *) rules: The GNF ruleset to parse with ** Returns : 0 on success, 1 if the tree couldn't be built in VOCAB_TREE_NODES nodes ** or if the sentence structure in 'words' is not part of the language ** described by the grammar passed in 'rules'. */ void vocab_dump_parse_tree(const char *tree_name, parse_tree_node_t *nodes); /* Prints a parse tree ** Parameters: (const char *) tree_name: Name of the tree to dump (free-form) ** (parse_tree_node_t *) nodes: The nodes containing the parse tree ** Returns : (void) */ int said(EngineState *s, byte *spec, int verbose); /* Builds a parse tree from a spec and compares it to a parse tree ** Parameters: (EngineState *) s: The affected state ** (byte *) spec: Pointer to the spec to build ** (int) verbose: Whether to display the parse tree after building it ** Returns : (int) 1 on a match, 0 otherwise */ const char *vocab_get_any_group_word(int group, word_t **words, int words_nr); /* Gets any word from the specified group. ** Parameters: (int) group: Group number. ** (word_t **) words: List of words ** (int) words_nr: Count of words in the list. ** For debugging only. */ void vocab_decypher_said_block(EngineState *s, byte *pos); /* Decyphers a said block and dumps its content via sciprintf. ** Parameters: (EngineState *) s: The state to use ** (byte *) pos: Pointer to the data to dump ** For debugging only. */ void vocab_synonymize_tokens(result_word_t *words, int words_nr, synonym_t *synonyms, int synonyms_nr); /* Synonymizes a token list ** Parameters: (result_wort_t *) words: The word list to synonymize ** (int) words_nr: Number of word_ts in the list ** (synonym_t *) synonyms: Synonym list ** (int) synonyms_nr: Number of synonyms in the list */ int vocab_gnf_parse(parse_tree_node_t *nodes, result_word_t *words, int words_nr, parse_tree_branch_t *branch0, parse_rule_list_t *tlist, int verbose); void vocab_gnf_dump(parse_tree_branch_t *branches, int branches_nr); } // End of namespace Sci #endif // SCI_SCICORE_VOCABULARY_H