X-Git-Url: https://git.saurik.com/apt.git/blobdiff_plain/37adedc9d0b66eeae7efb88aebd08dfbc6e06f77..1e3f4083db29bba600b9725e9456b0e140975c99:/apt-pkg/contrib/strutl.cc?ds=sidebyside diff --git a/apt-pkg/contrib/strutl.cc b/apt-pkg/contrib/strutl.cc index 861cdcbeb..d4f53ea3a 100644 --- a/apt-pkg/contrib/strutl.cc +++ b/apt-pkg/contrib/strutl.cc @@ -23,6 +23,7 @@ #include #include +#include #include #include #include @@ -35,7 +36,30 @@ using namespace std; /*}}}*/ +// Strip - Remove white space from the front and back of a string /*{{{*/ +// --------------------------------------------------------------------- +namespace APT { + namespace String { +std::string Strip(const std::string &s) +{ + size_t start = s.find_first_not_of(" \t\n"); + // only whitespace + if (start == string::npos) + return ""; + size_t end = s.find_last_not_of(" \t\n"); + return s.substr(start, end-start+1); +} + +bool Endswith(const std::string &s, const std::string &end) +{ + if (end.size() > s.size()) + return false; + return (s.substr(s.size() - end.size(), s.size()) == end); +} +} +} + /*}}}*/ // UTF8ToCodeset - Convert some UTF-8 string for some codeset /*{{{*/ // --------------------------------------------------------------------- /* This is handy to use before display some information for enduser */ @@ -116,7 +140,13 @@ char *_strstrip(char *String) if (*String == 0) return String; - + return _strrstrip(String); +} + /*}}}*/ +// strrstrip - Remove white space from the back of a string /*{{{*/ +// --------------------------------------------------------------------- +char *_strrstrip(char *String) +{ char *End = String + strlen(String) - 1; for (;End != String - 1 && (*End == ' ' || *End == '\t' || *End == '\n' || *End == '\r'); End--); @@ -396,7 +426,7 @@ string TimeToStr(unsigned long Sec) /*}}}*/ // SubstVar - Substitute a string for another string /*{{{*/ // --------------------------------------------------------------------- -/* This replaces all occurances of Subst with Contents in Str. */ +/* This replaces all occurrences of Subst with Contents in Str. */ string SubstVar(const string &Str,const string &Subst,const string &Contents) { string::size_type Pos = 0; @@ -751,7 +781,8 @@ bool ReadMessages(int Fd, vector &List) // Look for the end of the message for (char *I = Buffer; I + 1 < End; I++) { - if (I[0] != '\n' || I[1] != '\n') + if (I[1] != '\n' || + (I[0] != '\n' && strncmp(I, "\r\n\r\n", 4) != 0)) continue; // Pull the message out @@ -759,7 +790,7 @@ bool ReadMessages(int Fd, vector &List) PartialMessage += Message; // Fix up the buffer - for (; I < End && *I == '\n'; I++); + for (; I < End && (*I == '\n' || *I == '\r'); ++I); End -= I-Buffer; memmove(Buffer,I,End-Buffer); I = Buffer; @@ -895,7 +926,7 @@ bool FTPMDTMStrToTime(const char* const str,time_t &time) /*}}}*/ // StrToTime - Converts a string into a time_t /*{{{*/ // --------------------------------------------------------------------- -/* This handles all 3 populare time formats including RFC 1123, RFC 1036 +/* This handles all 3 popular time formats including RFC 1123, RFC 1036 and the C library asctime format. It requires the GNU library function 'timegm' to convert a struct tm in UTC to a time_t. For some bizzar reason the C library does not provide any such function :< This also @@ -910,17 +941,17 @@ bool StrToTime(const string &Val,time_t &Result) // Handle RFC 1123 time Month[0] = 0; - if (sscanf(I," %d %3s %d %d:%d:%d GMT",&Tm.tm_mday,Month,&Tm.tm_year, + if (sscanf(I," %2d %3s %4d %2d:%2d:%2d GMT",&Tm.tm_mday,Month,&Tm.tm_year, &Tm.tm_hour,&Tm.tm_min,&Tm.tm_sec) != 6) { // Handle RFC 1036 time - if (sscanf(I," %d-%3s-%d %d:%d:%d GMT",&Tm.tm_mday,Month, + if (sscanf(I," %2d-%3s-%3d %2d:%2d:%2d GMT",&Tm.tm_mday,Month, &Tm.tm_year,&Tm.tm_hour,&Tm.tm_min,&Tm.tm_sec) == 6) Tm.tm_year += 1900; else { // asctime format - if (sscanf(I," %3s %d %d:%d:%d %d",Month,&Tm.tm_mday, + if (sscanf(I," %3s %2d %2d:%2d:%2d %4d",Month,&Tm.tm_mday, &Tm.tm_hour,&Tm.tm_min,&Tm.tm_sec,&Tm.tm_year) != 6) { // 'ftp' time @@ -935,6 +966,8 @@ bool StrToTime(const string &Val,time_t &Result) Tm.tm_isdst = 0; if (Month[0] != 0) Tm.tm_mon = MonthConv(Month); + else + Tm.tm_mon = 0; // we don't have a month, so pick something Tm.tm_year -= 1900; // Convert to local time and then to GMT @@ -1108,6 +1141,37 @@ vector VectorizeString(string const &haystack, char const &split) return exploded; } /*}}}*/ +// StringSplit - split a string into a string vector by token /*{{{*/ +// --------------------------------------------------------------------- +/* See header for details. + */ +vector StringSplit(std::string const &s, std::string const &sep, + unsigned int maxsplit) +{ + vector split; + size_t start, pos; + + // no seperator given, this is bogus + if(sep.size() == 0) + return split; + + start = pos = 0; + while (pos != string::npos) + { + pos = s.find(sep, start); + split.push_back(s.substr(start, pos-start)); + + // if maxsplit is reached, the remaining string is the last item + if(split.size() >= maxsplit) + { + split[split.size()-1] = s.substr(start); + break; + } + start = pos+sep.size(); + } + return split; +} + /*}}}*/ // RegexChoice - Simple regex list/list matcher /*{{{*/ // --------------------------------------------------------------------- /* */ @@ -1168,34 +1232,50 @@ unsigned long RegexChoice(RxChoiceList *Rxs,const char **ListBegin, return Hits; } /*}}}*/ -// ioprintf - C format string outputter to C++ iostreams /*{{{*/ +// {str,io}printf - C format string outputter to C++ strings/iostreams /*{{{*/ // --------------------------------------------------------------------- /* This is used to make the internationalization strings easier to translate and to allow reordering of parameters */ -void ioprintf(ostream &out,const char *format,...) +static bool iovprintf(ostream &out, const char *format, + va_list &args, ssize_t &size) { + char *S = (char*)malloc(size); + ssize_t const n = vsnprintf(S, size, format, args); + if (n > -1 && n < size) { + out << S; + free(S); + return true; + } else { + if (n > -1) + size = n + 1; + else + size *= 2; + } + free(S); + return false; +} +void ioprintf(ostream &out,const char *format,...) { va_list args; - va_start(args,format); - - // sprintf the description - char S[4096]; - vsnprintf(S,sizeof(S),format,args); - out << S; + ssize_t size = 400; + while (true) { + va_start(args,format); + if (iovprintf(out, format, args, size) == true) + return; + va_end(args); + } } - /*}}}*/ -// strprintf - C format string outputter to C++ strings /*{{{*/ -// --------------------------------------------------------------------- -/* This is used to make the internationalization strings easier to translate - and to allow reordering of parameters */ -void strprintf(string &out,const char *format,...) +void strprintf(string &out,const char *format,...) { va_list args; - va_start(args,format); - - // sprintf the description - char S[4096]; - vsnprintf(S,sizeof(S),format,args); - out = string(S); + ssize_t size = 400; + std::ostringstream outstr; + while (true) { + va_start(args,format); + if (iovprintf(outstr, format, args, size) == true) + break; + va_end(args); + } + out = outstr.str(); } /*}}}*/ // safe_snprintf - Safer snprintf /*{{{*/ @@ -1209,12 +1289,12 @@ char *safe_snprintf(char *Buffer,char *End,const char *Format,...) va_list args; int Did; - va_start(args,Format); - if (End <= Buffer) return End; - + va_start(args,Format); Did = vsnprintf(Buffer,End - Buffer,Format,args); + va_end(args); + if (Did < 0 || Buffer + Did > End) return End; return Buffer + Did; @@ -1229,11 +1309,11 @@ string StripEpoch(const string &VerStr) return VerStr; return VerStr.substr(i+1); } - + /*}}}*/ // tolower_ascii - tolower() function that ignores the locale /*{{{*/ // --------------------------------------------------------------------- /* This little function is the most called method we have and tries - therefore to do the absolut minimum - and is noteable faster than + therefore to do the absolut minimum - and is notable faster than standard tolower/toupper and as a bonus avoids problems with different locales - we only operate on ascii chars anyway. */ int tolower_ascii(int const c) @@ -1244,9 +1324,9 @@ int tolower_ascii(int const c) } /*}}}*/ -// CheckDomainList - See if Host is in a , seperate list /*{{{*/ +// CheckDomainList - See if Host is in a , separate list /*{{{*/ // --------------------------------------------------------------------- -/* The domain list is a comma seperate list of domains that are suffix +/* The domain list is a comma separate list of domains that are suffix matched against the argument */ bool CheckDomainList(const string &Host,const string &List) { @@ -1267,14 +1347,26 @@ bool CheckDomainList(const string &Host,const string &List) return false; } /*}}}*/ -// DeEscapeString - unescape (\0XX and \xXX) from a string /*{{{*/ +// strv_length - Return the length of a NULL-terminated string array /*{{{*/ +// --------------------------------------------------------------------- +/* */ +size_t strv_length(const char **str_array) +{ + size_t i; + for (i=0; str_array[i] != NULL; i++) + /* nothing */ + ; + return i; +} + +// DeEscapeString - unescape (\0XX and \xXX) from a string /*{{{*/ // --------------------------------------------------------------------- /* */ string DeEscapeString(const string &input) { char tmp[3]; - string::const_iterator it, escape_start; - string output, octal, hex; + string::const_iterator it; + string output; for (it = input.begin(); it != input.end(); ++it) { // just copy non-escape chars @@ -1460,9 +1552,12 @@ URI::operator string() if (User.empty() == false) { - Res += User; + // FIXME: Technically userinfo is permitted even less + // characters than these, but this is not conveniently + // expressed with a blacklist. + Res += QuoteString(User, ":/?#[]@"); if (Password.empty() == false) - Res += ":" + Password; + Res += ":" + QuoteString(Password, ":/?#[]@"); Res += "@"; } @@ -1501,7 +1596,6 @@ string URI::SiteOnly(const string &URI) U.User.clear(); U.Password.clear(); U.Path.clear(); - U.Port = 0; return U; } /*}}}*/ @@ -1513,7 +1607,6 @@ string URI::NoUserPassword(const string &URI) ::URI U(URI); U.User.clear(); U.Password.clear(); - U.Port = 0; return U; } /*}}}*/