Permalink
Switch branches/tags
Nothing to show
Find file
Fetching contributors…
Cannot retrieve contributors at this time
603 lines (549 sloc) 20.4 KB
/*
* tools.h: Various tools
*
* See the main source file 'vdr.c' for copyright information and
* how to reach the author.
*
* $Id: tools.h 2.16 2012/02/29 10:41:00 kls Exp $
*/
#ifndef __TOOLS_H
#define __TOOLS_H
#include <dirent.h>
#include <errno.h>
#include <fcntl.h>
#include <float.h>
#include <iconv.h>
#include <math.h>
#include <poll.h>
#include <stdarg.h>
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <syslog.h>
#include <sys/stat.h>
#include <sys/types.h>
typedef unsigned char uchar;
extern int SysLogLevel;
#define esyslog(a...) void( (SysLogLevel > 0) ? syslog_with_tid(LOG_ERR, a) : void() )
#define isyslog(a...) void( (SysLogLevel > 1) ? syslog_with_tid(LOG_ERR, a) : void() )
#define dsyslog(a...) void( (SysLogLevel > 2) ? syslog_with_tid(LOG_ERR, a) : void() )
#define LOG_ERROR esyslog("ERROR (%s,%d): %m", __FILE__, __LINE__)
#define LOG_ERROR_STR(s) esyslog("ERROR (%s,%d): %s: %m", __FILE__, __LINE__, s)
#define SECSINDAY 86400
#define KILOBYTE(n) ((n) * 1024)
#define MEGABYTE(n) ((n) * 1024LL * 1024LL)
#define MALLOC(type, size) (type *)malloc(sizeof(type) * (size))
template<class T> inline void DELETENULL(T *&p) { T *q = p; p = NULL; delete q; }
#define CHECK(s) { if ((s) < 0) LOG_ERROR; } // used for 'ioctl()' calls
#define FATALERRNO (errno && errno != EAGAIN && errno != EINTR)
#ifndef __STL_CONFIG_H // in case some plugin needs to use the STL
template<class T> inline T min(T a, T b) { return a <= b ? a : b; }
template<class T> inline T max(T a, T b) { return a >= b ? a : b; }
template<class T> inline int sgn(T a) { return a < 0 ? -1 : a > 0 ? 1 : 0; }
template<class T> inline void swap(T &a, T &b) { T t = a; a = b; b = t; }
#endif
template<class T> inline T constrain(T v, T l, T h) { return v < l ? l : v > h ? h : v; }
void syslog_with_tid(int priority, const char *format, ...) __attribute__ ((format (printf, 2, 3)));
#define BCDCHARTOINT(x) (10 * ((x & 0xF0) >> 4) + (x & 0xF))
int BCD2INT(int x);
// Unfortunately there are no platform independent macros for unaligned
// access, so we do it this way:
template<class T> inline T get_unaligned(T *p)
{
struct s { T v; } __attribute__((packed));
return ((s *)p)->v;
}
template<class T> inline void put_unaligned(unsigned int v, T* p)
{
struct s { T v; } __attribute__((packed));
((s *)p)->v = v;
}
// Comparing doubles for equality is unsafe, but unfortunately we can't
// overwrite operator==(double, double), so this will have to do:
inline bool DoubleEqual(double a, double b)
{
return fabs(a - b) <= DBL_EPSILON;
}
// When handling strings that might contain UTF-8 characters, it may be necessary
// to process a "symbol" that consists of several actual character bytes. The
// following functions allow transparently accessing a "char *" string without
// having to worry about what character set is actually used.
int Utf8CharLen(const char *s);
///< Returns the number of character bytes at the beginning of the given
///< string that form a UTF-8 symbol.
uint Utf8CharGet(const char *s, int Length = 0);
///< Returns the UTF-8 symbol at the beginning of the given string.
///< Length can be given from a previous call to Utf8CharLen() to avoid calculating
///< it again. If no Length is given, Utf8CharLen() will be called.
int Utf8CharSet(uint c, char *s = NULL);
///< Converts the given UTF-8 symbol to a sequence of character bytes and copies
///< them to the given string. Returns the number of bytes written. If no string
///< is given, only the number of bytes is returned and nothing is copied.
int Utf8SymChars(const char *s, int Symbols);
///< Returns the number of character bytes at the beginning of the given
///< string that form at most the given number of UTF-8 symbols.
int Utf8StrLen(const char *s);
///< Returns the number of UTF-8 symbols formed by the given string of
///< character bytes.
char *Utf8Strn0Cpy(char *Dest, const char *Src, int n);
///< Copies at most n character bytes from Src to Dest, making sure that the
///< resulting copy ends with a complete UTF-8 symbol. The copy is guaranteed
///< to be zero terminated.
///< Returns a pointer to Dest.
int Utf8ToArray(const char *s, uint *a, int Size);
///< Converts the given character bytes (including the terminating 0) into an
///< array of UTF-8 symbols of the given Size. Returns the number of symbols
///< in the array (without the terminating 0).
int Utf8FromArray(const uint *a, char *s, int Size, int Max = -1);
///< Converts the given array of UTF-8 symbols (including the terminating 0)
///< into a sequence of character bytes of at most Size length. Returns the
///< number of character bytes written (without the terminating 0).
///< If Max is given, only that many symbols will be converted.
///< The resulting string is always zero-terminated if Size is big enough.
// When allocating buffer space, make sure we reserve enough space to hold
// a string in UTF-8 representation:
#define Utf8BufSize(s) ((s) * 4)
// The following macros automatically use the correct versions of the character
// class functions:
#define Utf8to(conv, c) (cCharSetConv::SystemCharacterTable() ? to##conv(c) : tow##conv(c))
#define Utf8is(ccls, c) (cCharSetConv::SystemCharacterTable() ? is##ccls(c) : isw##ccls(c))
class cCharSetConv {
private:
iconv_t cd;
char *result;
size_t length;
static char *systemCharacterTable;
public:
cCharSetConv(const char *FromCode = NULL, const char *ToCode = NULL);
///< Sets up a character set converter to convert from FromCode to ToCode.
///< If FromCode is NULL, the previously set systemCharacterTable is used
///< (or "UTF-8" if no systemCharacterTable has been set).
///< If ToCode is NULL, "UTF-8" is used.
~cCharSetConv();
const char *Convert(const char *From, char *To = NULL, size_t ToLength = 0);
///< Converts the given Text from FromCode to ToCode (as set in the constructor).
///< If To is given, it is used to copy at most ToLength bytes of the result
///< (including the terminating 0) into that buffer. If To is not given,
///< the result is copied into a dynamically allocated buffer and is valid as
///< long as this object lives, or until the next call to Convert(). The
///< return value always points to the result if the conversion was successful
///< (even if a fixed size To buffer was given and the result didn't fit into
///< it). If the string could not be converted, the result points to the
///< original From string.
static const char *SystemCharacterTable(void) { return systemCharacterTable; }
static void SetSystemCharacterTable(const char *CharacterTable);
};
class cString {
private:
char *s;
public:
cString(const char *S = NULL, bool TakePointer = false);
cString(const cString &String);
virtual ~cString();
operator const void * () const { return s; } // to catch cases where operator*() should be used
operator const char * () const { return s; } // for use in (const char *) context
const char * operator*() const { return s; } // for use in (const void *) context (printf() etc.)
cString &operator=(const cString &String);
cString &operator=(const char *String);
cString &Truncate(int Index); ///< Truncate the string at the given Index (if Index is < 0 it is counted from the end of the string).
static cString sprintf(const char *fmt, ...) __attribute__ ((format (printf, 1, 2)));
static cString sprintf(const char *fmt, va_list &ap);
};
ssize_t safe_read(int filedes, void *buffer, size_t size);
ssize_t safe_write(int filedes, const void *buffer, size_t size);
void writechar(int filedes, char c);
int WriteAllOrNothing(int fd, const uchar *Data, int Length, int TimeoutMs = 0, int RetryMs = 0);
///< Writes either all Data to the given file descriptor, or nothing at all.
///< If TimeoutMs is greater than 0, it will only retry for that long, otherwise
///< it will retry forever. RetryMs defines the time between two retries.
char *strcpyrealloc(char *dest, const char *src);
char *strn0cpy(char *dest, const char *src, size_t n);
char *strreplace(char *s, char c1, char c2);
char *strreplace(char *s, const char *s1, const char *s2); ///< re-allocates 's' and deletes the original string if necessary!
inline char *skipspace(const char *s)
{
if ((uchar)*s > ' ') // most strings don't have any leading space, so handle this case as fast as possible
return (char *)s;
while (*s && (uchar)*s <= ' ') // avoiding isspace() here, because it is much slower
s++;
return (char *)s;
}
char *stripspace(char *s);
char *compactspace(char *s);
cString strescape(const char *s, const char *chars);
bool startswith(const char *s, const char *p);
bool endswith(const char *s, const char *p);
bool isempty(const char *s);
int numdigits(int n);
bool isnumber(const char *s);
int64_t StrToNum(const char *s);
///< Converts the given string to a number.
///< The numerical part of the string may be followed by one of the letters
///< K, M, G or T to abbreviate Kilo-, Mega-, Giga- or Terabyte, respectively
///< (based on 1024). Everything after the first non-numeric character is
///< silently ignored, as are any characters other than the ones mentioned here.
cString itoa(int n);
cString AddDirectory(const char *DirName, const char *FileName);
bool EntriesOnSameFileSystem(const char *File1, const char *File2);
int FreeDiskSpaceMB(const char *Directory, int *UsedMB = NULL);
bool DirectoryOk(const char *DirName, bool LogErrors = false);
bool MakeDirs(const char *FileName, bool IsDirectory = false);
bool RemoveFileOrDir(const char *FileName, bool FollowSymlinks = false);
bool RemoveEmptyDirectories(const char *DirName, bool RemoveThis = false);
int DirSizeMB(const char *DirName); ///< returns the total size of the files in the given directory, or -1 in case of an error
char *ReadLink(const char *FileName); ///< returns a new string allocated on the heap, which the caller must delete (or NULL in case of an error)
bool SpinUpDisk(const char *FileName);
void TouchFile(const char *FileName);
time_t LastModifiedTime(const char *FileName);
off_t FileSize(const char *FileName); ///< returns the size of the given file, or -1 in case of an error (e.g. if the file doesn't exist)
cString WeekDayName(int WeekDay);
cString WeekDayName(time_t t);
cString WeekDayNameFull(int WeekDay);
cString WeekDayNameFull(time_t t);
cString DayDateTime(time_t t = 0);
cString TimeToString(time_t t);
cString DateString(time_t t);
cString TimeString(time_t t);
uchar *RgbToJpeg(uchar *Mem, int Width, int Height, int &Size, int Quality = 100);
///< Converts the given Memory to a JPEG image and returns a pointer
///< to the resulting image. Mem must point to a data block of exactly
///< (Width * Height) triplets of RGB image data bytes. Upon return, Size
///< will hold the number of bytes of the resulting JPEG data.
///< Quality can be in the range 0..100 and controls the quality of the
///< resulting image, where 100 is "best". The caller takes ownership of
///< the result and has to delete it once it is no longer needed.
///< The result may be NULL in case of an error.
class cBase64Encoder {
private:
const uchar *data;
int length;
int maxResult;
int i;
char *result;
static const char *b64;
public:
cBase64Encoder(const uchar *Data, int Length, int MaxResult = 64);
///< Sets up a new base 64 encoder for the given Data, with the given Length.
///< Data will not be copied and must be valid as long as NextLine() will be
///< called. MaxResult defines the maximum number of characters in any
///< result line. The resulting lines may be shorter than MaxResult in case
///< its value is not a multiple of 4.
~cBase64Encoder();
const char *NextLine(void);
///< Returns the next line of encoded data (terminated by '\0'), or NULL if
///< there is no more encoded data. The caller must call NextLine() and process
///< each returned line until NULL is returned, in order to get the entire
///< data encoded. The returned data is only valid until the next time NextLine()
///< is called, or until the object is destroyed.
};
class cBitStream {
private:
const uint8_t *data;
int length; // in bits
int index; // in bits
public:
cBitStream(const uint8_t *Data, int Length) : data(Data), length(Length), index(0) {}
~cBitStream() {}
int GetBit(void);
uint32_t GetBits(int n);
void ByteAlign(void);
void WordAlign(void);
bool SetLength(int Length);
void SkipBits(int n) { index += n; }
void SkipBit(void) { SkipBits(1); }
bool IsEOF(void) const { return index >= length; }
void Reset(void) { index = 0; }
int Length(void) const { return length; }
int Index(void) const { return (IsEOF() ? length : index); }
const uint8_t *GetData(void) const { return (IsEOF() ? NULL : data + (index / 8)); }
};
class cTimeMs {
private:
uint64_t begin;
public:
cTimeMs(int Ms = 0);
///< Creates a timer with ms resolution and an initial timeout of Ms.
///< If Ms is negative the timer is not initialized with the current
///< time.
static uint64_t Now(void);
void Set(int Ms = 0);
bool TimedOut(void);
uint64_t Elapsed(void);
};
class cReadLine {
private:
size_t size;
char *buffer;
public:
cReadLine(void);
~cReadLine();
char *Read(FILE *f);
};
class cPoller {
private:
enum { MaxPollFiles = 16 };
pollfd pfd[MaxPollFiles];
int numFileHandles;
public:
cPoller(int FileHandle = -1, bool Out = false);
bool Add(int FileHandle, bool Out);
bool Poll(int TimeoutMs = 0);
};
class cReadDir {
private:
DIR *directory;
struct dirent *result;
union { // according to "The GNU C Library Reference Manual"
struct dirent d;
char b[offsetof(struct dirent, d_name) + NAME_MAX + 1];
} u;
public:
cReadDir(const char *Directory);
~cReadDir();
bool Ok(void) { return directory != NULL; }
struct dirent *Next(void);
};
class cFile {
private:
static bool files[];
static int maxFiles;
int f;
public:
cFile(void);
~cFile();
operator int () { return f; }
bool Open(const char *FileName, int Flags, mode_t Mode = DEFFILEMODE);
bool Open(int FileDes);
void Close(void);
bool IsOpen(void) { return f >= 0; }
bool Ready(bool Wait = true);
static bool AnyFileReady(int FileDes = -1, int TimeoutMs = 1000);
static bool FileReady(int FileDes, int TimeoutMs = 1000);
static bool FileReadyForWriting(int FileDes, int TimeoutMs = 1000);
};
class cSafeFile {
private:
FILE *f;
char *fileName;
char *tempName;
public:
cSafeFile(const char *FileName);
~cSafeFile();
operator FILE* () { return f; }
bool Open(void);
bool Close(void);
};
/// cUnbufferedFile is used for large files that are mainly written or read
/// in a streaming manner, and thus should not be cached.
class cUnbufferedFile {
private:
int fd;
off_t curpos;
off_t cachedstart;
off_t cachedend;
off_t begin;
off_t lastpos;
off_t ahead;
size_t readahead;
size_t written;
size_t totwritten;
int FadviseDrop(off_t Offset, off_t Len);
public:
cUnbufferedFile(void);
~cUnbufferedFile();
int Open(const char *FileName, int Flags, mode_t Mode = DEFFILEMODE);
int Close(void);
void SetReadAhead(size_t ra);
off_t Seek(off_t Offset, int Whence);
ssize_t Read(void *Data, size_t Size);
ssize_t Write(const void *Data, size_t Size);
static cUnbufferedFile *Create(const char *FileName, int Flags, mode_t Mode = DEFFILEMODE);
};
class cLockFile {
private:
char *fileName;
int f;
public:
cLockFile(const char *Directory);
~cLockFile();
bool Lock(int WaitSeconds = 0);
void Unlock(void);
};
class cListObject {
private:
cListObject *prev, *next;
public:
cListObject(void);
virtual ~cListObject();
virtual int Compare(const cListObject &ListObject) const { return 0; }
///< Must return 0 if this object is equal to ListObject, a positive value
///< if it is "greater", and a negative value if it is "smaller".
void Append(cListObject *Object);
void Insert(cListObject *Object);
void Unlink(void);
int Index(void) const;
cListObject *Prev(void) const { return prev; }
cListObject *Next(void) const { return next; }
};
class cListBase {
protected:
cListObject *objects, *lastObject;
cListBase(void);
int count;
public:
virtual ~cListBase();
void Add(cListObject *Object, cListObject *After = NULL);
void Ins(cListObject *Object, cListObject *Before = NULL);
void Del(cListObject *Object, bool DeleteObject = true);
virtual void Move(int From, int To);
void Move(cListObject *From, cListObject *To);
virtual void Clear(void);
cListObject *Get(int Index) const;
int Count(void) const { return count; }
void Sort(void);
};
template<class T> class cList : public cListBase {
public:
T *Get(int Index) const { return (T *)cListBase::Get(Index); }
T *First(void) const { return (T *)objects; }
T *Last(void) const { return (T *)lastObject; }
T *Prev(const T *object) const { return (T *)object->cListObject::Prev(); } // need to call cListObject's members to
T *Next(const T *object) const { return (T *)object->cListObject::Next(); } // avoid ambiguities in case of a "list of lists"
};
template<class T> class cVector {
private:
mutable int allocated;
mutable int size;
mutable T *data;
cVector(const cVector &Vector) {} // don't copy...
cVector &operator=(const cVector &Vector) { return *this; } // ...or assign this!
void Realloc(int Index) const
{
if (++Index > allocated) {
data = (T *)realloc(data, Index * sizeof(T));
if (!data) {
esyslog("ERROR: out of memory - abort!");
abort();
}
for (int i = allocated; i < Index; i++)
data[i] = T(0);
allocated = Index;
}
}
public:
cVector(int Allocated = 10)
{
allocated = 0;
size = 0;
data = NULL;
Realloc(Allocated);
}
virtual ~cVector() { free(data); }
T& At(int Index) const
{
Realloc(Index);
if (Index >= size)
size = Index + 1;
return data[Index];
}
const T& operator[](int Index) const
{
return At(Index);
}
T& operator[](int Index)
{
return At(Index);
}
int Size(void) const { return size; }
virtual void Insert(T Data, int Before = 0)
{
if (Before < size) {
Realloc(size);
memmove(&data[Before + 1], &data[Before], (size - Before) * sizeof(T));
size++;
data[Before] = Data;
}
else
Append(Data);
}
virtual void Append(T Data)
{
if (size >= allocated)
Realloc(allocated * 4 / 2); // increase size by 50%
data[size++] = Data;
}
virtual void Remove(int Index)
{
if (Index < size - 1)
memmove(&data[Index], &data[Index + 1], (size - Index) * sizeof(T));
size--;
}
virtual void Clear(void)
{
size = 0;
}
void Sort(__compar_fn_t Compare)
{
qsort(data, size, sizeof(T), Compare);
}
};
inline int CompareStrings(const void *a, const void *b)
{
return strcmp(*(const char **)a, *(const char **)b);
}
inline int CompareStringsIgnoreCase(const void *a, const void *b)
{
return strcasecmp(*(const char **)a, *(const char **)b);
}
class cStringList : public cVector<char *> {
public:
cStringList(int Allocated = 10): cVector<char *>(Allocated) {}
virtual ~cStringList();
int Find(const char *s) const;
void Sort(bool IgnoreCase = false)
{
if (IgnoreCase)
cVector<char *>::Sort(CompareStringsIgnoreCase);
else
cVector<char *>::Sort(CompareStrings);
}
virtual void Clear(void);
};
class cFileNameList : public cStringList {
public:
cFileNameList(const char *Directory = NULL, bool DirsOnly = false);
bool Load(const char *Directory, bool DirsOnly = false);
};
class cHashObject : public cListObject {
friend class cHashBase;
private:
unsigned int id;
cListObject *object;
public:
cHashObject(cListObject *Object, unsigned int Id) { object = Object; id = Id; }
cListObject *Object(void) { return object; }
};
class cHashBase {
private:
cList<cHashObject> **hashTable;
int size;
unsigned int hashfn(unsigned int Id) const { return Id % size; }
protected:
cHashBase(int Size);
public:
virtual ~cHashBase();
void Add(cListObject *Object, unsigned int Id);
void Del(cListObject *Object, unsigned int Id);
void Clear(void);
cListObject *Get(unsigned int Id) const;
cList<cHashObject> *GetList(unsigned int Id) const;
};
#define HASHSIZE 512
template<class T> class cHash : public cHashBase {
public:
cHash(int Size = HASHSIZE) : cHashBase(Size) {}
T *Get(unsigned int Id) const { return (T *)cHashBase::Get(Id); }
};
#endif //__TOOLS_H