From 343d0d81769aeaf1ba3f505681549e8ee51c644d Mon Sep 17 00:00:00 2001 From: rubidium Date: Sat, 17 Dec 2011 14:41:10 +0000 Subject: (svn r23569) -Codechange: put more logic in some of the strgen structs, e.g. allocating and freeing its memory, and don't use a global variable for the string data --- src/strgen/strgen.cpp | 246 ++++++++++++++++++++++++++++++++------------------ 1 file changed, 156 insertions(+), 90 deletions(-) (limited to 'src/strgen') diff --git a/src/strgen/strgen.cpp b/src/strgen/strgen.cpp index 52bee9eb3..397639783 100644 --- a/src/strgen/strgen.cpp +++ b/src/strgen/strgen.cpp @@ -40,12 +40,6 @@ /* Compiles a list of strings into a compiled string list */ -struct Case { - int caseidx; - char *string; - Case *next; -}; - static bool _translated; ///< Whether the current language is not the master language static bool _translation; ///< Is the current file actually a translation or not static const char *_file = "(unknown file)"; ///< The filename of the input, so we can refer to it in errors/warnings @@ -54,25 +48,136 @@ static int _errors, _warnings, _show_todo; static const ptrdiff_t MAX_COMMAND_PARAM_SIZE = 100; ///< Maximum size of every command block, not counting the name of the command itself +/** Container for the different cases of a string. */ +struct Case { + int caseidx; ///< The index of the case. + char *string; ///< The translation of the case. + Case *next; ///< The next, chained, case. + + /** + * Create a new case. + * @param caseidx The index of the case. + * @param string The translation of the case. + * @param next The next chained case. + */ + Case(int caseidx, const char *string, Case *next) : + caseidx(caseidx), string(strdup(string)), next(next) + { + } + + /** Free everything we allocated. */ + ~Case() + { + free(this->string); + delete this->next; + } +}; + +/** Information about a single string. */ struct LangString { - char *name; // Name of the string - char *english; // English text - char *translated; // Translated text - uint16 hash_next; // next hash entry - uint16 index; - int line; // line of string in source-file - Case *translated_case; // cases for foreign + char *name; ///< Name of the string. + char *english; ///< English text. + char *translated; ///< Translated text. + uint16 hash_next; ///< Next hash entry. + uint16 index; ///< The index in the language file. + int line; ///< Line of string in source-file. + Case *translated_case; ///< Cases of the translation. + + /** + * Create a new string. + * @param name The name of the string. + * @param english The english "translation" of the string. + * @param index The index in the string table. + * @param line The line this string was found on. + */ + LangString(const char *name, const char *english, int index, int line) : + name(strdup(name)), english(strdup(english)), translated(NULL), + hash_next(0), index(index), line(line), translated_case(NULL) + { + } + + /** Free everything we allocated. */ + ~LangString() + { + free(this->name); + free(this->english); + free(this->translated); + delete this->translated_case; + } }; -static LangString *_strings[65536]; -static LanguagePackHeader _lang; ///< Header information about a language. +/** Information about the currently known strings. */ +struct StringData { + LangString **strings; ///< Array of all known strings. + uint16 *hash_heads; ///< Hash table for the strings. + size_t max_strings; ///< The maxmimum number of strings. + int next_string_id; ///< The next string ID to allocate. + /** + * Create a new string data container. + * @param max_strings The maximum number of strings. + */ + StringData(size_t max_strings = 65536) : max_strings(max_strings) + { + this->strings = CallocT(max_strings); + this->hash_heads = CallocT(max_strings); + this->next_string_id = 0; + } -#define HASH_SIZE 32767 -static uint16 _hash_head[HASH_SIZE]; + /** Free everything we allocated. */ + ~StringData() + { + for (size_t i = 0; i < this->max_strings; i++) delete this->strings[i]; + free(this->strings); + free(this->hash_heads); + } -static int _next_string_id; + /** + * Create a hash of the string for finding them back quickly. + * @param s The string to hash. + * @return The hashed string. + */ + uint HashStr(const char *s) + { + uint hash = 0; + for (; *s != '\0'; s++) hash = ROL(hash, 3) ^ *s; + return hash % this->max_strings; + } + /** + * Add a newly created LangString. + * @param s The name of the string. + * @param ls The string to add. + */ + void Add(const char *s, LangString *ls) + { + uint hash = this->HashStr(s); + ls->hash_next = this->hash_heads[hash]; + /* Off-by-one for hash find. */ + this->hash_heads[hash] = ls->index + 1; + this->strings[ls->index] = ls; + } + + /** + * Find a LangString based on the string name. + * @param s The string name to search on. + * @return The LangString or NULL if it is not known. + */ + LangString *Find(const char *s) + { + int idx = this->hash_heads[this->HashStr(s)]; + + while (--idx >= 0) { + LangString *ls = this->strings[idx]; + + if (strcmp(ls->name, s) == 0) return ls; + idx = ls->hash_next; + } + return NULL; + } +}; + +static LanguagePackHeader _lang; ///< Header information about a language. static uint32 _hash; static const char *_cur_ident; @@ -92,33 +197,6 @@ struct ParsedCommandStruct { static ParsedCommandStruct _cur_pcs; static int _cur_argidx; -static uint HashStr(const char *s) -{ - uint hash = 0; - for (; *s != '\0'; s++) hash = ROL(hash, 3) ^ *s; - return hash % HASH_SIZE; -} - -static void HashAdd(const char *s, LangString *ls) -{ - uint hash = HashStr(s); - ls->hash_next = _hash_head[hash]; - _hash_head[hash] = ls->index + 1; -} - -static LangString *HashFind(const char *s) -{ - int idx = _hash_head[HashStr(s)]; - - while (--idx >= 0) { - LangString *ls = _strings[idx]; - - if (strcmp(ls->name, s) == 0) return ls; - idx = ls->hash_next; - } - return NULL; -} - #ifdef _MSC_VER # define LINE_NUM_FMT(s) "%s (%d): warning: %s (" s ")\n" #else @@ -500,10 +578,10 @@ static const CmdStruct *ParseCommandString(const char **str, char *param, int *a } -static void HandlePragma(char *str, bool master) +static void HandlePragma(StringData &data, char *str, bool master) { if (!memcmp(str, "id ", 3)) { - _next_string_id = strtoul(str + 3, NULL, 0); + data.next_string_id = strtoul(str + 3, NULL, 0); } else if (!memcmp(str, "name ", 5)) { strecpy(_lang.name, str + 5, lastof(_lang.name)); } else if (!memcmp(str, "ownname ", 8)) { @@ -680,10 +758,10 @@ static bool CheckCommandsMatch(char *a, char *b, const char *name) return result; } -static void HandleString(char *str, bool master) +static void HandleString(StringData &data, char *str, bool master) { if (*str == '#') { - if (str[1] == '#' && str[2] != '#') HandlePragma(str + 2, master); + if (str[1] == '#' && str[2] != '#') HandlePragma(data, str + 2, master); return; } @@ -726,7 +804,7 @@ static void HandleString(char *str, bool master) if (casep != NULL) *casep++ = '\0'; /* Check if this string already exists.. */ - LangString *ent = HashFind(str); + LangString *ent = data.Find(str); if (master) { if (casep != NULL) { @@ -739,21 +817,13 @@ static void HandleString(char *str, bool master) return; } - if (_strings[_next_string_id]) { - strgen_error("String ID 0x%X for '%s' already in use by '%s'", _next_string_id, str, _strings[_next_string_id]->name); + if (data.strings[data.next_string_id] != NULL) { + strgen_error("String ID 0x%X for '%s' already in use by '%s'", data.next_string_id, str, data.strings[data.next_string_id]->name); return; } /* Allocate a new LangString */ - ent = CallocT(1); - _strings[_next_string_id] = ent; - ent->index = _next_string_id++; - ent->name = strdup(str); - ent->line = _cur_line; - - HashAdd(str, ent); - - ent->english = strdup(s); + data.Add(str, new LangString(str, s, data.next_string_id++, _cur_line)); } else { if (ent == NULL) { strgen_warning("String name '%s' does not exist in master file", str); @@ -769,12 +839,7 @@ static void HandleString(char *str, bool master) if (!CheckCommandsMatch(s, ent->english, str)) return; if (casep != NULL) { - Case *c = MallocT(1); - - c->caseidx = ResolveCaseName(casep, strlen(casep)); - c->string = strdup(s); - c->next = ent->translated_case; - ent->translated_case = c; + ent->translated_case = new Case(ResolveCaseName(casep, strlen(casep)), s, ent->translated_case); } else { ent->translated = strdup(s); /* If the string was translated, use the line from the @@ -794,7 +859,7 @@ static void rstrip(char *buf) } -static void ParseFile(const char *file, bool english) +static void ParseFile(StringData &data, const char *file, bool english) { FILE *in; char buf[2048]; @@ -816,7 +881,7 @@ static void ParseFile(const char *file, bool english) _cur_line = 1; while (fgets(buf, sizeof(buf), in) != NULL) { rstrip(buf); - HandleString(buf, english); + HandleString(data, buf, english); _cur_line++; } fclose(in); @@ -838,13 +903,12 @@ static uint32 MyHashStr(uint32 hash, const char *s) /* make a hash of the file to get a unique "version number" */ -static void MakeHashOfStrings() +static void MakeHashOfStrings(const StringData &data) { uint32 hash = 0; - uint i; - for (i = 0; i != lengthof(_strings); i++) { - const LangString *ls = _strings[i]; + for (size_t i = 0; i < data.max_strings; i++) { + const LangString *ls = data.strings[i]; if (ls != NULL) { const CmdStruct *cs; @@ -871,11 +935,11 @@ static void MakeHashOfStrings() } -static uint CountInUse(uint grp) +static uint CountInUse(const StringData &data, uint grp) { int i; - for (i = 0x800; --i >= 0;) if (_strings[(grp << 11) + i] != NULL) break; + for (i = 0x800; --i >= 0;) if (data.strings[(grp << 11) + i] != NULL) break; return i + 1; } @@ -963,12 +1027,12 @@ struct HeaderWriter { virtual ~HeaderWriter() {}; /** Write the header information. */ - void WriteHeader() + void WriteHeader(const StringData &data) { int last = 0; - for (int i = 0; i != lengthof(_strings); i++) { - if (_strings[i] != NULL) { - this->WriteStringID(_strings[i]->name, i); + for (size_t i = 0; i < data.max_strings; i++) { + if (data.strings[i] != NULL) { + this->WriteStringID(data.strings[i]->name, i); last = i; } } @@ -1154,17 +1218,17 @@ struct LanguageWriter { /** * Actually write the language. */ - void WriteLang() + void WriteLang(const StringData &data) { uint in_use[32]; for (int i = 0; i != 32; i++) { - uint n = CountInUse(i); + uint n = CountInUse(data, i); in_use[i] = n; _lang.offsets[i] = TO_LE16(n); for (uint j = 0; j != in_use[i]; j++) { - const LangString *ls = _strings[(i << 11) + j]; + const LangString *ls = data.strings[(i << 11) + j]; if (ls != NULL && ls->translated == NULL) _lang.missing++; } } @@ -1179,7 +1243,7 @@ struct LanguageWriter { for (int i = 0; i != 32; i++) { for (uint j = 0; j != in_use[i]; j++) { - const LangString *ls = _strings[(i << 11) + j]; + const LangString *ls = data.strings[(i << 11) + j]; const Case *casep; const char *cmdp; @@ -1448,9 +1512,10 @@ int CDECL main(int argc, char *argv[]) if (mgo.numleft == 0) { mkpath(pathbuf, lengthof(pathbuf), src_dir, "english.txt"); + StringData data; /* parse master file */ - ParseFile(pathbuf, true); - MakeHashOfStrings(); + ParseFile(data, pathbuf, true); + MakeHashOfStrings(data); if (_errors != 0) return 1; /* write strings.h */ @@ -1458,17 +1523,18 @@ int CDECL main(int argc, char *argv[]) mkpath(pathbuf, lengthof(pathbuf), dest_dir, "strings.h"); HeaderFileWriter writer(pathbuf); - writer.WriteHeader(); + writer.WriteHeader(data); writer.Finalise(); } else if (mgo.numleft == 1) { char *r; mkpath(pathbuf, lengthof(pathbuf), src_dir, "english.txt"); + StringData data; /* parse master file and check if target file is correct */ - ParseFile(pathbuf, true); - MakeHashOfStrings(); - ParseFile(replace_pathsep(mgo.argv[0]), false); // target file + ParseFile(data, pathbuf, true); + MakeHashOfStrings(data); + ParseFile(data, replace_pathsep(mgo.argv[0]), false); // target file if (_errors != 0) return 1; /* get the targetfile, strip any directories and append to destination path */ @@ -1481,7 +1547,7 @@ int CDECL main(int argc, char *argv[]) ttd_strlcpy(r, ".lng", (size_t)(r - pathbuf)); LanguageFileWriter writer(pathbuf); - writer.WriteLang(); + writer.WriteLang(data); writer.Finalise(); /* if showing warnings, print a summary of the language */ -- cgit v1.2.3-70-g09d2