diff options
author | glx <glx@openttd.org> | 2009-05-14 21:59:24 +0000 |
---|---|---|
committer | glx <glx@openttd.org> | 2009-05-14 21:59:24 +0000 |
commit | b3a539b0168c4058ffe19930a455f987778a5fbf (patch) | |
tree | fcecf03292e29df6fde60c7292f0e4c761bae755 /src/depend | |
parent | 203cd6c2ddf4fe6853d0a72d280640e10579918c (diff) | |
download | openttd-b3a539b0168c4058ffe19930a455f987778a5fbf.tar.xz |
(svn r16307) -Fix: makedepend can't handle the amount of files we have and it also miss some dependencies. That's why we introduce our custom implementation of makedepend.
Diffstat (limited to 'src/depend')
-rw-r--r-- | src/depend/depend.cpp | 941 |
1 files changed, 941 insertions, 0 deletions
diff --git a/src/depend/depend.cpp b/src/depend/depend.cpp new file mode 100644 index 000000000..79c70f7bf --- /dev/null +++ b/src/depend/depend.cpp @@ -0,0 +1,941 @@ +/* $Id$ */ + +/** + * @file depend/depend.cpp Custom implementation of Makedepend. + * + * We previously used makedepend, but that could not handle the amount of + * files we have and does not handle conditional includes in a sane manner. + * This caused many link problems because not enough files were recompiled. + * This has lead to the development of our own dependency generator. It is + * meant to be a substitute to the (relatively slow) dependency generation + * via gcc. It thus helps speeding up compilation. It will also ignore + * system headers making it less error prone when system headers are moved + * or renamed. + */ + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <ctype.h> +#include <limits.h> +#include <unistd.h> +#include <map> +#include <set> +#include <stack> + +/** Simple string comparator using strcmp as implementation */ +struct StringCompare { + /** + * Compare a to b using strcmp. + * @param a string to compare. + * @param b string to compare. + * @return whether a is less than b. + */ + bool operator () (const char *a, const char *b) const + { + return strcmp(a, b) < 0; + } +}; +/** Set of C-style strings. */ +typedef std::set<const char*, StringCompare> StringSet; +/** Mapping of C-style string to a set of C-style strings. */ +typedef std::map<const char*, StringSet*, StringCompare> StringMap; +/** Pair of C-style string and a set of C-style strings. */ +typedef std::pair<const char*, StringSet*> StringMapItem; + +/** Include directory to search in. */ +static StringSet _include_dirs; +/** Files that have been parsed/handled with their dependencies. */ +static StringMap _files; +/** Dependencies of headers. */ +static StringMap _headers; +/** The current 'active' defines. */ +static StringSet _defines; + +/** + * Helper class to read a file. + */ +class File { +public: + /** + * Create the helper by opening the given file. + * @param filename the file to open + * @post the file is open; otherwise the application is killed. + */ + File(const char *filename) : filename(filename) + { + this->fp = fopen(filename, "r"); + if (this->fp == NULL) { + fprintf(stdout, "Could not open %s for reading\n", filename); + exit(1); + } + this->dirname = strdup(filename); + char *last = strrchr(this->dirname, '/'); + if (last != NULL) { + *last = '\0'; + } else { + *this->dirname = '\0'; + } + } + + /** Free everything we have allocated. */ + ~File() + { + fclose(this->fp); + free(this->dirname); + } + + /** + * Get a single character from the file. + * If we are reading beyond the end of the file '\0' is returned. + * @return the read character. + */ + char GetChar() const + { + char c = fgetc(this->fp); + return (c == EOF) ? '\0' : c; + } + + /** + * Get the directory name of the file. + * @return the directory name. + */ + const char *GetDirname() const + { + return this->dirname; + } + +private: + FILE *fp; ///< The currently opened file. + char *dirname; ///< The directory of the file. + const char *filename; ///< The name of the file. +}; + +/** A token returned by the tokenizer. */ +enum Token { + TOKEN_UNKNOWN, ///< Unknown token + TOKEN_END, ///< End of document + TOKEN_EOL, ///< End of line + TOKEN_SHARP, ///< # character, usually telling something important comes. + TOKEN_LOCAL, ///< Read a local include + TOKEN_GLOBAL, ///< Read a global include + TOKEN_IDENTIFIER, ///< Identifier within the data. + TOKEN_DEFINE, ///< (#)define in code + TOKEN_IF, ///< (#)if in code + TOKEN_IFDEF, ///< (#)ifdef in code + TOKEN_IFNDEF, ///< (#)ifndef in code + TOKEN_ELIF, ///< (#)elif in code + TOKEN_ELSE, ///< (#)else in code + TOKEN_ENDIF, ///< (#)endif in code + TOKEN_UNDEF, ///< (#)undef in code + TOKEN_OR, ///< '||' within #if expression + TOKEN_AND, ///< '&&' within #if expression + TOKEN_DEFINED, ///< 'defined' within #if expression + TOKEN_OPEN, ///< '(' within #if expression + TOKEN_CLOSE, ///< ')' within #if expression + TOKEN_NOT, ///< '!' within #if expression + TOKEN_ZERO, ///< '0' within #if expression + TOKEN_INCLUDE, ///< (#)include in code +}; + +/** Mapping from a C-style keyword representation to a Token. */ +typedef std::map<const char*, Token, StringCompare> KeywordList; + +/** + * Lexer of a file. + */ +class Lexer { +public: + /** + * Create the lexer and fill the keywords table. + * @param file the file to read from. + */ + Lexer(const File *file) : file(file), current_char('\0'), string(NULL), token(TOKEN_UNKNOWN) + { + this->keywords["define"] = TOKEN_DEFINE; + this->keywords["defined"] = TOKEN_DEFINED; + this->keywords["if"] = TOKEN_IF; + this->keywords["ifdef"] = TOKEN_IFDEF; + this->keywords["ifndef"] = TOKEN_IFNDEF; + this->keywords["include"] = TOKEN_INCLUDE; + this->keywords["elif"] = TOKEN_ELIF; + this->keywords["else"] = TOKEN_ELSE; + this->keywords["endif"] = TOKEN_ENDIF; + this->keywords["undef"] = TOKEN_UNDEF; + + /* Initialise currently read character. */ + this->Next(); + + /* Allocate the buffer. */ + this->buf_len = 32; + this->buf = (char*)malloc(sizeof(*this->buf) * this->buf_len); + } + + /** Free everything */ + ~Lexer() + { + free(this->buf); + } + + /** + * Read the next character into 'current_char'. + */ + void Next() + { + this->current_char = this->file->GetChar(); + } + + /** + * Get the current token. + * @return the token. + */ + Token GetToken() const + { + return this->token; + } + + /** + * Read the currenty processed string. + * @return the string, can be NULL. + */ + const char *GetString() const + { + return this->string; + } + + /** + * Perform the lexing/tokenizing of the file till we can return something + * that must be parsed. + */ + void Lex() + { + for (;;) { + free(this->string); + this->string = NULL; + this->token = TOKEN_UNKNOWN; + + switch (this->current_char) { + /* '\0' means End-Of-File */ + case '\0': this->token = TOKEN_END; return; + + /* Skip some chars, as they don't do anything */ + case '\t': this->Next(); break; + case '\r': this->Next(); break; + case ' ': this->Next(); break; + + case '\\': + this->Next(); + if (this->current_char == '\n') this->Next(); + break; + + case '\n': + this->token = TOKEN_EOL; + this->Next(); + return; + + case '#': + this->token = TOKEN_SHARP; + this->Next(); + return; + + case '"': + this->ReadString('"', TOKEN_LOCAL); + this->Next(); + return; + + case '<': + this->ReadString('>', TOKEN_GLOBAL); + this->Next(); + return; + + case '&': + this->Next(); + if (this->current_char == '&') { + this->Next(); + this->token = TOKEN_AND; + return; + } + break; + + case '|': + this->Next(); + if (this->current_char == '|') { + this->Next(); + this->token = TOKEN_OR; + return; + } + break; + + case '(': + this->Next(); + this->token = TOKEN_OPEN; + return; + + case ')': + this->Next(); + this->token = TOKEN_CLOSE; + return; + + case '!': + this->Next(); + if (this->current_char != '=') { + this->token = TOKEN_NOT; + return; + } + break; + + /* Possible begin of comment */ + case '/': + this->Next(); + switch (this->current_char) { + case '*': { + this->Next(); + char previous_char = '\0'; + while ((this->current_char != '/' || previous_char != '*') && this->current_char != '\0') { + previous_char = this->current_char; + this->Next(); + } + this->Next(); + break; + } + case '/': while (this->current_char != '\n' && this->current_char != '\0') this->Next(); break; + default: break; + } + break; + + default: + if (isalpha(this->current_char) || this->current_char == '_') { + /* If the name starts with a letter, it is an identifier */ + this->ReadIdentifier(); + return; + } + if (isdigit(this->current_char)) { + bool zero = this->current_char == '0'; + this->Next(); + if (this->current_char == 'x' || this->current_char == 'X') Next(); + while (isdigit(this->current_char) || this->current_char == '.' || (this->current_char >= 'a' && this->current_char <= 'f') || (this->current_char >= 'A' && this->current_char <= 'F')) { + zero &= this->current_char == '0'; + this->Next(); + } + if (zero) this->token = TOKEN_ZERO; + return; + } + this->Next(); + break; + } + } + } + +private: + /** + * The token based on keyword with a given name. + * @param name the actual keyword. + * @return the token of the keyword. + */ + Token FindKeyword(const char *name) const + { + KeywordList::const_iterator it = this->keywords.find(name); + if (it == this->keywords.end()) return TOKEN_IDENTIFIER; + return (*it).second; + } + + /** + * Read an identifier. + */ + void ReadIdentifier() + { + size_t count = 0; + + /* Read the rest of the identifier */ + do { + this->buf[count++] = this->current_char; + this->Next(); + + if (count >= buf_len) { + /* Scale the buffer if required */ + this->buf_len *= 2; + this->buf = (char *)realloc(this->buf, sizeof(*this->buf) * this->buf_len); + } + } while ((isalpha(this->current_char) || this->current_char == '_' || isdigit(this->current_char))); + this->buf[count] = '\0'; + + free(this->string); + this->string = strdup(this->buf); + this->token = FindKeyword(this->string); + } + + /** + * Read a string up to a given character, then set the given token. + * @param end the 'marker' for the end of the string. + * @param token the token to set after returning. + */ + void ReadString(char end, Token token) + { + size_t count = 0; + this->Next(); + while (this->current_char != end && this->current_char != ')' && this->current_char != '\n' && this->current_char != '\0') { + this->buf[count++] = this->current_char; + this->Next(); + + if (count >= this->buf_len) { + /* Scale the buffer if required */ + this->buf_len *= 2; + this->buf = (char *)realloc(this->buf, sizeof(*this->buf) * this->buf_len); + } + } + this->buf[count] = '\0'; + free(this->string); + this->string = strdup(this->buf); + this->token = token; + } + + const File *file; ///< The file to read from. + char current_char; ///< The current character to process. + char *string; ///< Currently processed string. + Token token; ///< The current token to process. + char *buf; ///< Temporary buffer. + size_t buf_len; ///< Length of the temporary buffer. + KeywordList keywords; ///< All keywords we know of. +}; + +/** + * Generate a path from a directory name and a relative filename. + * If the file is not local the include directory names will be used instead + * of the passed parameter with directory name. If the file is local both will + * be queried where the parameter takes precedence. + * @param dirname the directory to look in. + * @param filename the file to look for. + * @param local whether to look locally (in dirname) for the file. + * @return the absolute path, or NULL if the file doesn't exist. + */ +const char *GeneratePath(const char *dirname, const char *filename, bool local) +{ + if (local) { + if (access(filename, R_OK) == 0) return strdup(filename); + + char path[PATH_MAX]; + strcpy(path, dirname); + const char *p = filename; + /* Remove '..' from the begin of the filename. */ + while (*p == '.') { + if (*(++p) == '.') { + *(strrchr(path, '/')) = '\0'; + p += 2; + } + } + strcat(path, "/"); + strcat(path, p); + + if (access(path, R_OK) == 0) return strdup(path); + } + + for (StringSet::iterator it = _include_dirs.begin(); it != _include_dirs.end(); it++) { + char path[PATH_MAX]; + strcpy(path, *it); + const char *p = filename; + /* Remove '..' from the begin of the filename. */ + while (*p == '.') { + if (*(++p) == '.') { + *(strrchr(path, '/')) = '\0'; + p += 2; + } + } + strcat(path, "/"); + strcat(path, p); + + if (access(path, R_OK) == 0) return strdup(path); + } + + return NULL; +} + +/** + * Try to parse a 'defined(expr)' expression. + * @param lexer the lexer to get tokens from. + * @param defines the set of known defines. + * @param verbose whether to give verbose debugging information. + * @return the value of the expression. + */ +bool ExpressionDefined(Lexer *lexer, StringSet *defines, bool verbose); + +/** + * Try to parse a 'expr || expr' expression. + * @param lexer the lexer to get tokens from. + * @param defines the set of known defines. + * @param verbose whether to give verbose debugging information. + * @return the value of the expression. + */ +bool ExpressionOr(Lexer *lexer, StringSet *defines, bool verbose); + +/** + * Try to parse a '!expr' expression. Also parses the '(expr)', '0' and + * identifiers. Finally it also consumes any unknown tokens. + * @param lexer the lexer to get tokens from. + * @param defines the set of known defines. + * @param verbose whether to give verbose debugging information. + * @return the value of the expression. + */ +bool ExpressionNot(Lexer *lexer, StringSet *defines, bool verbose) +{ + if (lexer->GetToken() == TOKEN_NOT) { + if (verbose) fprintf(stderr, "!"); + lexer->Lex(); + bool value = !ExpressionDefined(lexer, defines, verbose); + if (verbose) fprintf(stderr, "[%d]", value); + return value; + } + + if (lexer->GetToken() == TOKEN_OPEN) { + if (verbose) fprintf(stderr, "("); + lexer->Lex(); + bool value = ExpressionOr(lexer, defines, verbose); + if (verbose) fprintf(stderr, ")[%d]", value); + lexer->Lex(); + return value; + } + + if (lexer->GetToken() == TOKEN_ZERO) { + if (verbose) fprintf(stderr, "0"); + lexer->Lex(); + if (verbose) fprintf(stderr, "[0]"); + return false; + } + + bool first = true; + while (lexer->GetToken() == TOKEN_UNKNOWN || lexer->GetToken() == TOKEN_IDENTIFIER) { + if (verbose && first) fprintf(stderr, "<assumed true>"); + first = false; + lexer->Lex(); + } + + return true; +} + +/** + * Try to parse a 'defined(expr)' expression. + * @param lexer the lexer to get tokens from. + * @param defines the set of known defines. + * @param verbose whether to give verbose debugging information. + * @return the value of the expression. + */ +bool ExpressionDefined(Lexer *lexer, StringSet *defines, bool verbose) +{ + bool value = ExpressionNot(lexer, defines, verbose); + + if (lexer->GetToken() != TOKEN_DEFINED) return value; + lexer->Lex(); + if (verbose) fprintf(stderr, "defined"); + bool open = (lexer->GetToken() == TOKEN_OPEN); + if (open) lexer->Lex(); + if (verbose) fprintf(stderr, open ? "(" : " "); + if (lexer->GetToken() == TOKEN_IDENTIFIER) { + if (verbose) fprintf(stderr, "%s", lexer->GetString()); + value = defines->find(lexer->GetString()) != defines->end(); + } + if (open) { + if (verbose) fprintf(stderr, ")"); + lexer->Lex(); + } + lexer->Lex(); + if (verbose) fprintf(stderr, "[%d]", value); + return value; +} + +/** + * Try to parse a 'expr && expr' expression. + * @param lexer the lexer to get tokens from. + * @param defines the set of known defines. + * @param verbose whether to give verbose debugging information. + * @return the value of the expression. + */ +bool ExpressionAnd(Lexer *lexer, StringSet *defines, bool verbose) +{ + bool value = ExpressionDefined(lexer, defines, verbose); + + while (true) { + if (lexer->GetToken() != TOKEN_AND) return value; + if (verbose) fprintf(stderr, " && "); + lexer->Lex(); + value = value && ExpressionDefined(lexer, defines, verbose); + } +} + +/** + * Try to parse a 'expr || expr' expression. + * @param lexer the lexer to get tokens from. + * @param defines the set of known defines. + * @param verbose whether to give verbose debugging information. + * @return the value of the expression. + */ +bool ExpressionOr(Lexer *lexer, StringSet *defines, bool verbose) +{ + bool value = ExpressionAnd(lexer, defines, verbose); + + while (true) { + if (lexer->GetToken() != TOKEN_OR) return value; + if (verbose) fprintf(stderr, " || "); + lexer->Lex(); + value = value || ExpressionAnd(lexer, defines, verbose); + } +} + +/** Enumerator to tell how long to ignore 'stuff'. */ +enum Ignore { + NOT_IGNORE, ///< No ignoring. + IGNORE_UNTIL_ELSE, ///< Ignore till a #else is reached. + IGNORE_UNTIL_ENDIF, ///< Ignore till a #endif is reached. +}; + +/** + * Scan a file for includes, defines and the lot. + * @param filename the name of the file to scan. + * @param ext the extension of the filename. + * @param header whether the file is a header or not. + * @param verbose whether to give verbose debugging information. + */ +void ScanFile(const char *filename, const char *ext, bool header, bool verbose) +{ + static StringSet defines; + static std::stack<Ignore> ignore; + /* Copy in the default defines (parameters of depend) */ + if (!header) { + for (StringSet::iterator it = _defines.begin(); it != _defines.end(); it++) { + defines.insert(strdup(*it)); + } + } + + File file(filename); + Lexer lexer(&file); + + /* Start the lexing! */ + lexer.Lex(); + + while (lexer.GetToken() != TOKEN_END) { + switch (lexer.GetToken()) { + /* We reached the end of the file... yay, we're done! */ + case TOKEN_END: break; + + /* The line started with a # (minus whitespace) */ + case TOKEN_SHARP: + lexer.Lex(); + switch (lexer.GetToken()) { + case TOKEN_INCLUDE: + if (verbose) fprintf(stderr, "%s #include ", filename); + lexer.Lex(); + switch (lexer.GetToken()) { + case TOKEN_LOCAL: + case TOKEN_GLOBAL: { + if (verbose) fprintf(stderr, "%s", lexer.GetString()); + if (!ignore.empty() && ignore.top() != NOT_IGNORE) { + if (verbose) fprintf(stderr, " (ignored)"); + break; + } + const char *h = GeneratePath(file.GetDirname(), lexer.GetString(), lexer.GetToken() == TOKEN_LOCAL); + if (h != NULL) { + StringMap::iterator it = _headers.find(h); + if (it == _headers.end()) { + it = (_headers.insert(StringMapItem(strdup(h), new StringSet()))).first; + if (verbose) fprintf(stderr, "\n"); + ScanFile(h, ext, true, verbose); + } + StringMap::iterator curfile; + if (header) { + curfile = _headers.find(filename); + } else { + /* Replace the extension with the provided extension of '.o'. */ + char path[PATH_MAX]; + strcpy(path, filename); + *(strchr(path, '.')) = '\0'; + strcat(path, ext != NULL ? ext : ".o"); + curfile = _files.find(path); + if (curfile == _files.end()) { + curfile = (_files.insert(StringMapItem(strdup(path), new StringSet()))).first; + } + } + if (it != _headers.end()) { + for (StringSet::iterator header = it->second->begin(); header != it->second->end(); header++) { + if (curfile->second->find(*header) == curfile->second->end()) curfile->second->insert(strdup(*header)); + } + } + if (curfile->second->find(h) == curfile->second->end()) curfile->second->insert(strdup(h)); + free((void*)h); + } + } + /* Fall through */ + default: break; + } + break; + + case TOKEN_DEFINE: + if (verbose) fprintf(stderr, "%s #define ", filename); + lexer.Lex(); + if (lexer.GetToken() == TOKEN_IDENTIFIER) { + if (verbose) fprintf(stderr, "%s", lexer.GetString()); + if (!ignore.empty() && ignore.top() != NOT_IGNORE) { + if (verbose) fprintf(stderr, " (ignored)"); + break; + } + if (defines.find(lexer.GetString()) == defines.end()) defines.insert(strdup(lexer.GetString())); + lexer.Lex(); + } + break; + + case TOKEN_UNDEF: + if (verbose) fprintf(stderr, "%s #undef ", filename); + lexer.Lex(); + if (lexer.GetToken() == TOKEN_IDENTIFIER) { + if (verbose) fprintf(stderr, "%s", lexer.GetString()); + if (!ignore.empty() && ignore.top() != NOT_IGNORE) { + if (verbose) fprintf(stderr, " (ignored)"); + break; + } + StringSet::iterator it = defines.find(lexer.GetString()); + if (it != defines.end()) { + free((void*)*it); + defines.erase(it); + } + lexer.Lex(); + } + break; + + case TOKEN_ENDIF: + if (verbose) fprintf(stderr, "%s #endif", filename); + lexer.Lex(); + if (!ignore.empty()) ignore.pop(); + if (verbose) fprintf(stderr, " -> %signore", (!ignore.empty() && ignore.top() != NOT_IGNORE) ? "" : "not "); + break; + + case TOKEN_ELSE: { + if (verbose) fprintf(stderr, "%s #else", filename); + lexer.Lex(); + Ignore last = ignore.empty() ? NOT_IGNORE : ignore.top(); + if (!ignore.empty()) ignore.pop(); + if (ignore.empty() || ignore.top() == NOT_IGNORE) { + ignore.push(last == IGNORE_UNTIL_ELSE ? NOT_IGNORE : IGNORE_UNTIL_ENDIF); + } else { + ignore.push(IGNORE_UNTIL_ENDIF); + } + if (verbose) fprintf(stderr, " -> %signore", (!ignore.empty() && ignore.top() != NOT_IGNORE) ? "" : "not "); + break; + } + + case TOKEN_ELIF: { + if (verbose) fprintf(stderr, "%s #elif ", filename); + lexer.Lex(); + Ignore last = ignore.empty() ? NOT_IGNORE : ignore.top(); + if (!ignore.empty()) ignore.pop(); + if (ignore.empty() || ignore.top() == NOT_IGNORE) { + bool value = ExpressionOr(&lexer, &defines, verbose); + ignore.push(last == IGNORE_UNTIL_ELSE ? (value ? NOT_IGNORE : IGNORE_UNTIL_ELSE) : IGNORE_UNTIL_ENDIF); + } else { + ignore.push(IGNORE_UNTIL_ENDIF); + } + if (verbose) fprintf(stderr, " -> %signore", (!ignore.empty() && ignore.top() != NOT_IGNORE) ? "" : "not "); + break; + } + + case TOKEN_IF: { + if (verbose) fprintf(stderr, "%s #if ", filename); + lexer.Lex(); + if (ignore.empty() || ignore.top() == NOT_IGNORE) { + bool value = ExpressionOr(&lexer, &defines, verbose); + ignore.push(value ? NOT_IGNORE : IGNORE_UNTIL_ELSE); + } else { + ignore.push(IGNORE_UNTIL_ENDIF); + } + if (verbose) fprintf(stderr, " -> %signore", (!ignore.empty() && ignore.top() != NOT_IGNORE) ? "" : "not "); + break; + } + + case TOKEN_IFDEF: + if (verbose) fprintf(stderr, "%s #ifdef ", filename); + lexer.Lex(); + if (lexer.GetToken() == TOKEN_IDENTIFIER) { + bool value = defines.find(lexer.GetString()) != defines.end(); + if (verbose) fprintf(stderr, "%s[%d]", lexer.GetString(), value); + if (ignore.empty() || ignore.top() == NOT_IGNORE) { + ignore.push(value ? NOT_IGNORE : IGNORE_UNTIL_ELSE); + } else { + ignore.push(IGNORE_UNTIL_ENDIF); + } + } + if (verbose) fprintf(stderr, " -> %signore", (!ignore.empty() && ignore.top() != NOT_IGNORE) ? "" : "not "); + break; + + case TOKEN_IFNDEF: + if (verbose) fprintf(stderr, "%s #ifndef ", filename); + lexer.Lex(); + if (lexer.GetToken() == TOKEN_IDENTIFIER) { + bool value = defines.find(lexer.GetString()) != defines.end(); + if (verbose) fprintf(stderr, "%s[%d]", lexer.GetString(), value); + if (ignore.empty() || ignore.top() == NOT_IGNORE) { + ignore.push(!value ? NOT_IGNORE : IGNORE_UNTIL_ELSE); + } else { + ignore.push(IGNORE_UNTIL_ENDIF); + } + } + if (verbose) fprintf(stderr, " -> %signore", (!ignore.empty() && ignore.top() != NOT_IGNORE) ? "" : "not "); + break; + + default: + if (verbose) fprintf(stderr, "%s #<unknown>", filename); + lexer.Lex(); + break; + } + if (verbose) fprintf(stderr, "\n"); + /* Fall through */ + default: + /* Ignore the rest of the garbage on this line */ + while (lexer.GetToken() != TOKEN_EOL && lexer.GetToken() != TOKEN_END) lexer.Lex(); + lexer.Lex(); + break; + } + } + + if (!header) { + for (StringSet::iterator it = defines.begin(); it != defines.end(); it++) { + free((void*)*it); + } + defines.clear(); + while (!ignore.empty()) ignore.pop(); + } +} + +/** + * Entry point. Arguably the most common function in all applications. + * @param argc the number of arguments. + * @param argv the actual arguments. + * @return return value for the caller to tell we succeed or not. + */ +int main(int argc, char *argv[]) +{ + bool ignorenext = true; + char *filename = NULL; + char *ext = NULL; + bool append = false; + bool verbose = false; + + for (int i = 0; i < argc; i++) { + if (ignorenext) { + ignorenext = false; + continue; + } + if (argv[i][0] == '-') { + /* Append */ + if (strncmp(argv[i], "-a", 2) == 0) append = true; + /* Include dir */ + if (strncmp(argv[i], "-I", 2) == 0) { + if (argv[i][2] == '\0') { + i++; + _include_dirs.insert(strdup(argv[i])); + } else { + _include_dirs.insert(strdup(&argv[i][2])); + } + continue; + } + /* Define */ + if (strncmp(argv[i], "-D", 2) == 0) { + char *p = strchr(argv[i], '='); + if (p != NULL) *p = '\0'; + _defines.insert(strdup(&argv[i][2])); + continue; + } + /* Output file */ + if (strncmp(argv[i], "-f", 2) == 0) { + if (filename != NULL) continue; + filename = strdup(&argv[i][2]); + continue; + } + /* Object file extension */ + if (strncmp(argv[i], "-o", 2) == 0) { + if (ext != NULL) continue; + ext = strdup(&argv[i][2]); + continue; + } + /* Verbose */ + if (strncmp(argv[i], "-v", 2) == 0) verbose = true; + continue; + } + ScanFile(argv[i], ext, false, verbose); + } + + /* Default output file is Makefile */ + if (filename == NULL) filename = strdup("Makefile"); + + char backup[PATH_MAX]; + strcpy(backup, filename); + strcat(backup, ".bak"); + + char *content = NULL; + long size = 0; + + /* Read in the current file; so we can overwrite everything from the + * end of non-depend data marker down till the end. */ + FILE *src = fopen(filename, "rb"); + if (src != NULL) { + fseek(src, 0, SEEK_END); + size = ftell(src); + rewind(src); + content = (char*)malloc(size * sizeof(*content)); + fread(content, 1, size, src); + fclose(src); + } + + FILE *dst = fopen(backup, "wb"); + if (content != NULL) fwrite(content, 1, size, dst); + fclose(dst); + + /* Then append it to the real file. */ + src = fopen(backup, "r"); + dst = fopen(filename, "w"); + while (fgets(content, size, src) != NULL) { + fputs(content, dst); + if (!append && !strncmp(content, "# DO NOT DELETE", 15)) break; + } + if (ftell(src) == 0) fprintf(dst, "\n# DO NOT DELETE\n"); + for (StringMap::iterator it = _files.begin(); it != _files.end(); it++) { + for (StringSet::iterator h = it->second->begin(); h != it->second->end(); h++) { + fprintf(dst, "%s: %s\n", it->first, *h); + } + } + + /* Clean up our mess. */ + fclose(dst); + fclose(src); + + free(filename); + free(ext); + free(content); + + for (StringMap::iterator it = _files.begin(); it != _files.end(); it++) { + for (StringSet::iterator h = it->second->begin(); h != it->second->end(); h++) { + free((void*)*h); + } + it->second->clear(); + delete it->second; + free((void*)it->first); + } + _files.clear(); + + for (StringMap::iterator it = _headers.begin(); it != _headers.end(); it++) { + for (StringSet::iterator h = it->second->begin(); h != it->second->end(); h++) { + free((void*)*h); + } + it->second->clear(); + delete it->second; + free((void*)it->first); + } + _headers.clear(); + + for (StringSet::iterator it = _defines.begin(); it != _defines.end(); it++) { + free((void*)*it); + } + _defines.clear(); + + for (StringSet::iterator it = _include_dirs.begin(); it != _include_dirs.end(); it++) { + free((void*)*it); + } + _include_dirs.clear(); + + return 0; +} |