server/include/util.h

/*
 * This file is part of Deliantra, the Roguelike Realtime MMORPG.
 * 
 * Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team
 * 
 * Deliantra is free software: you can redistribute it and/or modify it under
 * the terms of the Affero GNU General Public License as published by the
 * Free Software Foundation, either version 3 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 Affero GNU General Public License
 * and the GNU General Public License along with this program. If not, see
 * <http://www.gnu.org/licenses/>.
 * 
 * The authors can be reached via e-mail to <support@deliantra.net>
 */

#ifndef UTIL_H__
#define UTIL_H__

#define DEBUG_POISON 0x00 // poison memory before freeing it if != 0
#define DEBUG_SALLOC  0   // add a debug wrapper around all sallocs
#define PREFER_MALLOC 0   // use malloc and not the slice allocator

#if __GNUC__ >= 3
# define is_constant(c)             __builtin_constant_p (c)
# define expect(expr,value)         __builtin_expect ((expr),(value))
# define prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
# define noinline                   __attribute__((__noinline__))
#else
# define is_constant(c)             0
# define expect(expr,value)         (expr)
# define prefetch(addr,rw,locality)
# define noinline
#endif

#if __GNUC__ < 4 || (__GNUC__ == 4 || __GNUC_MINOR__ < 4)
# define decltype(x) typeof(x)
#endif

// put into ifs if you are very sure that the expression
// is mostly true or mosty false. note that these return
// booleans, not the expression.
#define expect_false(expr) expect ((expr) ? 1 : 0, 0)
#define expect_true(expr)  expect ((expr) ? 1 : 0, 1)

#include <pthread.h>

#include <cstddef>
#include <cmath>
#include <new>
#include <vector>

#include <glib.h>

#include <shstr.h>
#include <traits.h>

#if DEBUG_SALLOC
# define g_slice_alloc0(s) debug_slice_alloc0(s)
# define g_slice_alloc(s) debug_slice_alloc(s)
# define g_slice_free1(s,p) debug_slice_free1(s,p)
void *g_slice_alloc (unsigned long size);
void *g_slice_alloc0 (unsigned long size);
void g_slice_free1 (unsigned long size, void *ptr);
#elif PREFER_MALLOC
# define g_slice_alloc0(s) calloc (1, (s))
# define g_slice_alloc(s) malloc ((s))
# define g_slice_free1(s,p) free ((p))
#endif

// use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever)
#define auto(var,expr) decltype(expr) var = (expr)

// very ugly macro that basically declares and initialises a variable
// that is in scope for the next statement only
// works only for stuff that can be assigned 0 and converts to false
// (note: works great for pointers)
// most ugly macro I ever wrote
#define statementvar(type, name, value) if (type name = 0) { } else if (((name) = (value)), 1)

// in range including end
#define IN_RANGE_INC(val,beg,end) \
  ((unsigned int)(val) - (unsigned int)(beg) <= (unsigned int)(end) - (unsigned int)(beg))

// in range excluding end
#define IN_RANGE_EXC(val,beg,end) \
  ((unsigned int)(val) - (unsigned int)(beg) <  (unsigned int)(end) - (unsigned int)(beg))

void cleanup (const char *cause, bool make_core = false);
void fork_abort (const char *msg);

// rationale for using (U) not (T) is to reduce signed/unsigned issues,
// as a is often a constant while b is the variable. it is still a bug, though.
template<typename T, typename U> static inline T min (T a, U b) { return (U)a < b ? (U)a : b; }
template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; }
template<typename T, typename U, typename V> static inline T clamp (T v, U a, V b) { return v < (T)a ? (T)a : v >(T)b ? (T)b : v; }

template<typename T, typename U> static inline void min_it (T &v, U m) { v = min (v, (T)m); }
template<typename T, typename U> static inline void max_it (T &v, U m) { v = max (v, (T)m); }
template<typename T, typename U, typename V> static inline void clamp_it (T &v, U a, V b) { v = clamp (v, (T)a, (T)b); }

template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; }

template<typename T, typename U, typename V> static inline T min (T a, U b, V c) { return min (a, min (b, c)); }
template<typename T, typename U, typename V> static inline T max (T a, U b, V c) { return max (a, max (b, c)); }

// sign returns -1 or +1
template<typename T>
static inline T sign (T v) { return v < 0 ? -1 : +1; }
// relies on 2c representation
template<>
inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); }

// sign0 returns -1, 0 or +1
template<typename T>
static inline T sign0 (T v) { return v ? sign (v) : 0; }

// div* only work correctly for div > 0
// div, with correct rounding (< 0.5 downwards, >=0.5 upwards)
template<typename T> static inline T div    (T val, T div)
{
  return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div;
}
// div, round-up
template<typename T> static inline T div_ru (T val, T div)
{
  return expect_false (val < 0) ? - ((-val                ) / div) : (val + div - 1) / div;
}
// div, round-down
template<typename T> static inline T div_rd (T val, T div)
{
  return expect_false (val < 0) ? - ((-val + (div - 1)    ) / div) : (val          ) / div;
}

// lerp* only work correctly for min_in < max_in
// Linear intERPolate, scales val from min_in..max_in to min_out..max_out
template<typename T>
static inline T
lerp (T val, T min_in, T max_in, T min_out, T max_out)
{
  return min_out + div   <T> ((val - min_in) * (max_out - min_out), max_in  - min_in);
}

// lerp, round-down
template<typename T>
static inline T
lerp_rd (T val, T min_in, T max_in, T min_out, T max_out)
{
  return min_out + div_rd<T> ((val - min_in) * (max_out - min_out), max_in  - min_in);
}

// lerp, round-up
template<typename T>
static inline T
lerp_ru (T val, T min_in, T max_in, T min_out, T max_out)
{
  return min_out + div_ru<T> ((val - min_in) * (max_out - min_out), max_in  - min_in);
}

// lots of stuff taken from FXT

/* Rotate right. This is used in various places for checksumming */
//TODO: that sucks, use a better checksum algo
static inline uint32_t
rotate_right (uint32_t c, uint32_t count = 1)
{
  return (c << (32 - count)) | (c >> count);
}

static inline uint32_t
rotate_left (uint32_t c, uint32_t count = 1)
{
  return (c >> (32 - count)) | (c << count);
}

// Return abs(a-b)
// Both a and b must not have the most significant bit set
static inline uint32_t
upos_abs_diff (uint32_t a, uint32_t b)
{
  long d1 = b - a;
  long d2 = (d1 & (d1 >> 31)) << 1;

  return d1 - d2;               // == (b - d) - (a + d);
}

// Both a and b must not have the most significant bit set
static inline uint32_t
upos_min (uint32_t a, uint32_t b)
{
  int32_t d = b - a;
  d &= d >> 31;
  return a + d;
}

// Both a and b must not have the most significant bit set
static inline uint32_t
upos_max (uint32_t a, uint32_t b)
{
  int32_t d = b - a;
  d &= d >> 31;
  return b - d;
}

// this is much faster than crossfires original algorithm
// on modern cpus
inline int
isqrt (int n)
{
  return (int)sqrtf ((float)n);
}

// this is only twice as fast as naive sqrtf (dx*dy+dy*dy)
#if 0
// and has a max. error of 6 in the range -100..+100.
#else
// and has a max. error of 9 in the range -100..+100.
#endif
inline int 
idistance (int dx, int dy)
{ 
  unsigned int dx_ = abs (dx);
  unsigned int dy_ = abs (dy);

#if 0
  return dx_ > dy_
      ? (dx_ * 61685 + dy_ * 26870) >> 16
      : (dy_ * 61685 + dx_ * 26870) >> 16;
#else
  return dx_ + dy_ - min (dx_, dy_) * 5 / 8;
#endif
}

/*
 * absdir(int): Returns a number between 1 and 8, which represent
 * the "absolute" direction of a number (it actually takes care of
 * "overflow" in previous calculations of a direction).
 */
inline int
absdir (int d)
{
  return ((d - 1) & 7) + 1;
}

extern ssize_t slice_alloc; // statistics

void *salloc_ (int n)            throw (std::bad_alloc);
void *salloc_ (int n, void *src) throw (std::bad_alloc);

// strictly the same as g_slice_alloc, but never returns 0
template<typename T>
inline T *salloc (int n = 1)     throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T));              }

// also copies src into the new area, like "memdup"
// if src is 0, clears the memory
template<typename T>
inline T *salloc (int n, T *src) throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T), (void *)src); }

// clears the memory
template<typename T>
inline T *salloc0(int n = 1)     throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T), 0);           }

// for symmetry
template<typename T>
inline void sfree (T *ptr, int n = 1) throw ()
{
  if (expect_true (ptr))
    {
      slice_alloc -= n * sizeof (T);
      if (DEBUG_POISON) memset (ptr, DEBUG_POISON, n * sizeof (T));
      g_slice_free1 (n * sizeof (T), (void *)ptr);
      assert (slice_alloc >= 0);//D
    }
}

// nulls the pointer
template<typename T>
inline void sfree0 (T *&ptr, int n = 1) throw ()
{
  sfree<T> (ptr, n);
  ptr = 0;
}

// makes dynamically allocated objects zero-initialised
struct zero_initialised
{
  void *operator new (size_t s, void *p)
  {
    memset (p, 0, s);
    return p;
  }

  void *operator new (size_t s)
  {
    return salloc0<char> (s);
  }

  void *operator new[] (size_t s)
  {
    return salloc0<char> (s);
  }

  void operator delete (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }

  void operator delete[] (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }
};

// makes dynamically allocated objects zero-initialised
struct slice_allocated
{
  void *operator new (size_t s, void *p)
  {
    return p;
  }

  void *operator new (size_t s)
  {
    return salloc<char> (s);
  }

  void *operator new[] (size_t s)
  {
    return salloc<char> (s);
  }

  void operator delete (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }

  void operator delete[] (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }
};

// a STL-compatible allocator that uses g_slice
// boy, this is verbose
template<typename Tp>
struct slice_allocator
{
  typedef size_t size_type;
  typedef ptrdiff_t difference_type;
  typedef Tp *pointer;
  typedef const Tp *const_pointer;
  typedef Tp &reference;
  typedef const Tp &const_reference;
  typedef Tp value_type;

  template <class U> 
  struct rebind
  {
    typedef slice_allocator<U> other;
  };

  slice_allocator () throw () { }
  slice_allocator (const slice_allocator &) throw () { }
  template<typename Tp2>
  slice_allocator (const slice_allocator<Tp2> &) throw () { }

  ~slice_allocator () { }

  pointer address (reference x) const { return &x; }
  const_pointer address (const_reference x) const { return &x; }

  pointer allocate (size_type n, const_pointer = 0)
  {
    return salloc<Tp> (n);
  }

  void deallocate (pointer p, size_type n)
  {
    sfree<Tp> (p, n);
  }

  size_type max_size () const throw ()
  {
    return size_t (-1) / sizeof (Tp);
  }

  void construct (pointer p, const Tp &val)
  {
    ::new (p) Tp (val);
  }

  void destroy (pointer p)
  {
    p->~Tp ();
  }
};

// P. L'Ecuyer, “Maximally Equidistributed Combined Tausworthe Generators”, Mathematics of Computation, 65, 213 (1996), 203–213.
// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps
// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
struct tausworthe_random_generator
{
  uint32_t state [4];

  void operator =(const tausworthe_random_generator &src)
  {
    state [0] = src.state [0];
    state [1] = src.state [1];
    state [2] = src.state [2];
    state [3] = src.state [3];
  }

  void seed (uint32_t seed);
  uint32_t next ();
};

// Xorshift RNGs, George Marsaglia
// http://www.jstatsoft.org/v08/i14/paper
// this one is about 40% faster than the tausworthe one above (i.e. not much),
// despite the inlining, and has the issue of only creating 2**32-1 numbers.
// see also http://www.iro.umontreal.ca/~lecuyer/myftp/papers/xorshift.pdf
struct xorshift_random_generator
{
  uint32_t x, y;

  void operator =(const xorshift_random_generator &src)
  {
    x = src.x;
    y = src.y;
  }

  void seed (uint32_t seed)
  {
    x = seed;
    y = seed * 69069U;
  }

  uint32_t next ()
  {
    uint32_t t = x ^ (x << 10);
    x = y;
    y = y ^ (y >> 13) ^ t ^ (t >> 10);
    return y;
  }
};

template<class generator>
struct random_number_generator : generator
{
  // uniform distribution, 0 .. max (0, num - 1)
  uint32_t operator ()(uint32_t num)
  {
    return !is_constant (num) ? get_range (num)                        // non-constant
         : num & (num - 1)    ? (this->next () * (uint64_t)num) >> 32U // constant, non-power-of-two
         :                      this->next () & (num - 1);             // constant, power-of-two
  }

  // return a number within (min .. max)
  int operator () (int r_min, int r_max)
  {
    return is_constant (r_min) && is_constant (r_max) && r_min <= r_max
              ? r_min + operator ()(r_max - r_min + 1)
              : get_range (r_min, r_max);
  }

  double operator ()()
  {
    return this->next () / (double)0xFFFFFFFFU;
  }

protected:
  uint32_t get_range (uint32_t r_max);
  int get_range (int r_min, int r_max);
};

typedef random_number_generator<tausworthe_random_generator> rand_gen;

extern rand_gen rndm, rmg_rndm;

INTERFACE_CLASS (attachable)
struct refcnt_base
{
  typedef int refcnt_t;
  mutable refcnt_t ACC (RW, refcnt);

  MTH void refcnt_inc () const { ++refcnt; }
  MTH void refcnt_dec () const { --refcnt; }

  refcnt_base () : refcnt (0) { }
};

// to avoid branches with more advanced compilers
extern refcnt_base::refcnt_t refcnt_dummy;

template<class T>
struct refptr
{
  // p if not null
  refcnt_base::refcnt_t *refcnt_ref () { return p ? &p->refcnt : &refcnt_dummy; }

  void refcnt_dec ()
  {
    if (!is_constant (p))
      --*refcnt_ref ();
    else if (p)
      --p->refcnt;
  }

  void refcnt_inc ()
  {
    if (!is_constant (p))
      ++*refcnt_ref ();
    else if (p)
      ++p->refcnt;
  }

  T *p;

  refptr () : p(0) { }
  refptr (const refptr<T> &p) : p(p.p) { refcnt_inc (); }
  refptr (T *p) : p(p) { refcnt_inc (); }
  ~refptr () { refcnt_dec (); }

  const refptr<T> &operator =(T *o)
  {
    // if decrementing ever destroys we need to reverse the order here
    refcnt_dec ();
    p = o;
    refcnt_inc ();
    return *this;
  }

  const refptr<T> &operator =(const refptr<T> &o)
  {
    *this = o.p;
    return *this;
  }

  T &operator * () const { return *p; }
  T *operator ->() const { return  p; }

  operator T *() const { return p; }
};

typedef refptr<maptile>   maptile_ptr;
typedef refptr<object>    object_ptr;
typedef refptr<archetype> arch_ptr;
typedef refptr<client>    client_ptr;
typedef refptr<player>    player_ptr;

struct str_hash
{
  std::size_t operator ()(const char *s) const
  {
#if 0
    uint32_t hash = 0;

    /* use the one-at-a-time hash function, which supposedly is
     * better than the djb2-like one used by perl5.005, but
     * certainly is better then the bug used here before.
     * see http://burtleburtle.net/bob/hash/doobs.html
     */
    while (*s)
      {
        hash += *s++;
        hash += hash << 10;
        hash ^= hash >>  6;
      }

    hash += hash <<  3;
    hash ^= hash >> 11;
    hash += hash << 15;
#else
    // use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/)
    // it is about twice as fast as the one-at-a-time one,
    // with good distribution.
    // FNV-1a is faster on many cpus because the multiplication
    // runs concurrent with the looping logic.
    uint32_t hash = 2166136261;
    
    while (*s)
      hash = (hash ^ *s++) * 16777619;
#endif

    return hash;
  }
};

struct str_equal
{
  bool operator ()(const char *a, const char *b) const
  {
    return !strcmp (a, b);
  }
};

// Mostly the same as std::vector, but insert/erase can reorder
// the elements, making append(=insert)/remove O(1) instead of O(n).
//
// NOTE: only some forms of erase are available
template<class T>
struct unordered_vector : std::vector<T, slice_allocator<T> >
{
  typedef typename unordered_vector::iterator iterator;

  void erase (unsigned int pos)
  {
    if (pos < this->size () - 1)
      (*this)[pos] = (*this)[this->size () - 1];

    this->pop_back ();
  }

  void erase (iterator i)
  {
    erase ((unsigned int )(i - this->begin ()));
  }
};

// This container blends advantages of linked lists
// (efficiency) with vectors (random access) by
// by using an unordered vector and storing the vector
// index inside the object.
//
// + memory-efficient on most 64 bit archs
// + O(1) insert/remove
// + free unique (but varying) id for inserted objects
// + cache-friendly iteration
// - only works for pointers to structs
//
// NOTE: only some forms of erase/insert are available
typedef int object_vector_index;

template<class T, object_vector_index T::*indexmember>
struct object_vector : std::vector<T *, slice_allocator<T *> >
{
  typedef typename object_vector::iterator iterator;

  bool contains (const T *obj) const
  {
    return obj->*indexmember;
  }

  iterator find (const T *obj)
  {
    return obj->*indexmember
      ? this->begin () + obj->*indexmember - 1
      : this->end ();
  }

  void push_back (T *obj)
  {
    std::vector<T *, slice_allocator<T *> >::push_back (obj);
    obj->*indexmember = this->size ();
  }

  void insert (T *obj)
  {
    push_back (obj);
  }

  void insert (T &obj)
  {
    insert (&obj);
  }

  void erase (T *obj)
  {
    unsigned int pos = obj->*indexmember;
    obj->*indexmember = 0;

    if (pos < this->size ())
      {
        (*this)[pos - 1] = (*this)[this->size () - 1];
        (*this)[pos - 1]->*indexmember = pos;
      }

    this->pop_back ();
  }

  void erase (T &obj)
  {
    erase (&obj);
  }
};

// basically does what strncpy should do, but appends "..." to strings exceeding length
// returns the number of bytes actually used (including \0)
int assign (char *dst, const char *src, int maxsize);

// type-safe version of assign
template<int N>
inline int assign (char (&dst)[N], const char *src)
{
  return assign ((char *)&dst, src, N);
}

typedef double tstamp;

// return current time as timestamp
tstamp now ();

int similar_direction (int a, int b);

// like sprintf, but returns a "static" buffer
const char *format (const char *format, ...);

// safety-check player input which will become object->msg
bool msg_is_safe (const char *msg);

/////////////////////////////////////////////////////////////////////////////
// threads, very very thin wrappers around pthreads

struct thread
{
  pthread_t id;

  void start (void *(*start_routine)(void *), void *arg = 0);

  void cancel ()
  {
    pthread_cancel (id);
  }

  void *join ()
  {
    void *ret;

    if (pthread_join (id, &ret))
      cleanup ("pthread_join failed", 1);

    return ret;
  }
};

// note that mutexes are not classes
typedef pthread_mutex_t smutex;

#if __linux && defined (PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP)
 #define SMUTEX_INITIALISER PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP
#else
 #define SMUTEX_INITIALISER PTHREAD_MUTEX_INITIALIZER
#endif

#define SMUTEX(name) smutex name = SMUTEX_INITIALISER
#define SMUTEX_LOCK(name)   pthread_mutex_lock   (&(name))
#define SMUTEX_UNLOCK(name) pthread_mutex_unlock (&(name))

typedef pthread_cond_t scond;

#define SCOND(name) scond name = PTHREAD_COND_INITIALIZER
#define SCOND_SIGNAL(name)     pthread_cond_signal    (&(name))
#define SCOND_BROADCAST(name)  pthread_cond_broadcast (&(name))
#define SCOND_WAIT(name,mutex) pthread_cond_wait      (&(name), &(mutex))

#endif

Revision:	1.90
Committed:	Mon Oct 12 14:00:58 2009 UTC (14 years, 7 months ago) by root
Content type:	text/plain
Branch:	MAIN
CVS Tags:	rel-2_81
Changes since 1.89:	+7 -6 lines
Log Message:	clarify license
#	Content
1	/*
2	* This file is part of Deliantra, the Roguelike Realtime MMORPG.
3	*
4	* Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team
5	*
6	* Deliantra is free software: you can redistribute it and/or modify it under
7	* the terms of the Affero GNU General Public License as published by the
8	* Free Software Foundation, either version 3 of the License, or (at your
9	* option) any later version.
10	*
11	* This program is distributed in the hope that it will be useful,
12	* but WITHOUT ANY WARRANTY; without even the implied warranty of
13	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14	* GNU General Public License for more details.
15	*
16	* You should have received a copy of the Affero GNU General Public License
17	* and the GNU General Public License along with this program. If not, see
18	* <http://www.gnu.org/licenses/>.
19	*
20	* The authors can be reached via e-mail to <support@deliantra.net>
21	*/
22
23	#ifndef UTIL_H__
24	#define UTIL_H__
25
26	#define DEBUG_POISON 0x00 // poison memory before freeing it if != 0
27	#define DEBUG_SALLOC 0 // add a debug wrapper around all sallocs
28	#define PREFER_MALLOC 0 // use malloc and not the slice allocator
29
30	#if __GNUC__ >= 3
31	# define is_constant(c) __builtin_constant_p (c)
32	# define expect(expr,value) __builtin_expect ((expr),(value))
33	# define prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
34	# define noinline __attribute__((__noinline__))
35	#else
36	# define is_constant(c) 0
37	# define expect(expr,value) (expr)
38	# define prefetch(addr,rw,locality)
39	# define noinline
40	#endif
41
42	#if __GNUC__ < 4 \|\| (__GNUC__ == 4 \|\| __GNUC_MINOR__ < 4)
43	# define decltype(x) typeof(x)
44	#endif
45
46	// put into ifs if you are very sure that the expression
47	// is mostly true or mosty false. note that these return
48	// booleans, not the expression.
49	#define expect_false(expr) expect ((expr) ? 1 : 0, 0)
50	#define expect_true(expr) expect ((expr) ? 1 : 0, 1)
51
52	#include <pthread.h>
53
54	#include <cstddef>
55	#include <cmath>
56	#include <new>
57	#include <vector>
58
59	#include <glib.h>
60
61	#include <shstr.h>
62	#include <traits.h>
63
64	#if DEBUG_SALLOC
65	# define g_slice_alloc0(s) debug_slice_alloc0(s)
66	# define g_slice_alloc(s) debug_slice_alloc(s)
67	# define g_slice_free1(s,p) debug_slice_free1(s,p)
68	void *g_slice_alloc (unsigned long size);
69	void *g_slice_alloc0 (unsigned long size);
70	void g_slice_free1 (unsigned long size, void *ptr);
71	#elif PREFER_MALLOC
72	# define g_slice_alloc0(s) calloc (1, (s))
73	# define g_slice_alloc(s) malloc ((s))
74	# define g_slice_free1(s,p) free ((p))
75	#endif
76
77	// use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever)
78	#define auto(var,expr) decltype(expr) var = (expr)
79
80	// very ugly macro that basically declares and initialises a variable
81	// that is in scope for the next statement only
82	// works only for stuff that can be assigned 0 and converts to false
83	// (note: works great for pointers)
84	// most ugly macro I ever wrote
85	#define statementvar(type, name, value) if (type name = 0) { } else if (((name) = (value)), 1)
86
87	// in range including end
88	#define IN_RANGE_INC(val,beg,end) \
89	((unsigned int)(val) - (unsigned int)(beg) <= (unsigned int)(end) - (unsigned int)(beg))
90
91	// in range excluding end
92	#define IN_RANGE_EXC(val,beg,end) \
93	((unsigned int)(val) - (unsigned int)(beg) < (unsigned int)(end) - (unsigned int)(beg))
94
95	void cleanup (const char *cause, bool make_core = false);
96	void fork_abort (const char *msg);
97
98	// rationale for using (U) not (T) is to reduce signed/unsigned issues,
99	// as a is often a constant while b is the variable. it is still a bug, though.
100	template<typename T, typename U> static inline T min (T a, U b) { return (U)a < b ? (U)a : b; }
101	template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; }
102	template<typename T, typename U, typename V> static inline T clamp (T v, U a, V b) { return v < (T)a ? (T)a : v >(T)b ? (T)b : v; }
103
104	template<typename T, typename U> static inline void min_it (T &v, U m) { v = min (v, (T)m); }
105	template<typename T, typename U> static inline void max_it (T &v, U m) { v = max (v, (T)m); }
106	template<typename T, typename U, typename V> static inline void clamp_it (T &v, U a, V b) { v = clamp (v, (T)a, (T)b); }
107
108	template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; }
109
110	template<typename T, typename U, typename V> static inline T min (T a, U b, V c) { return min (a, min (b, c)); }
111	template<typename T, typename U, typename V> static inline T max (T a, U b, V c) { return max (a, max (b, c)); }
112
113	// sign returns -1 or +1
114	template<typename T>
115	static inline T sign (T v) { return v < 0 ? -1 : +1; }
116	// relies on 2c representation
117	template<>
118	inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); }
119
120	// sign0 returns -1, 0 or +1
121	template<typename T>
122	static inline T sign0 (T v) { return v ? sign (v) : 0; }
123
124	// div* only work correctly for div > 0
125	// div, with correct rounding (< 0.5 downwards, >=0.5 upwards)
126	template<typename T> static inline T div (T val, T div)
127	{
128	return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div;
129	}
130	// div, round-up
131	template<typename T> static inline T div_ru (T val, T div)
132	{
133	return expect_false (val < 0) ? - ((-val ) / div) : (val + div - 1) / div;
134	}
135	// div, round-down
136	template<typename T> static inline T div_rd (T val, T div)
137	{
138	return expect_false (val < 0) ? - ((-val + (div - 1) ) / div) : (val ) / div;
139	}
140
141	// lerp* only work correctly for min_in < max_in
142	// Linear intERPolate, scales val from min_in..max_in to min_out..max_out
143	template<typename T>
144	static inline T
145	lerp (T val, T min_in, T max_in, T min_out, T max_out)
146	{
147	return min_out + div <T> ((val - min_in) * (max_out - min_out), max_in - min_in);
148	}
149
150	// lerp, round-down
151	template<typename T>
152	static inline T
153	lerp_rd (T val, T min_in, T max_in, T min_out, T max_out)
154	{
155	return min_out + div_rd<T> ((val - min_in) * (max_out - min_out), max_in - min_in);
156	}
157
158	// lerp, round-up
159	template<typename T>
160	static inline T
161	lerp_ru (T val, T min_in, T max_in, T min_out, T max_out)
162	{
163	return min_out + div_ru<T> ((val - min_in) * (max_out - min_out), max_in - min_in);
164	}
165
166	// lots of stuff taken from FXT
167
168	/* Rotate right. This is used in various places for checksumming */
169	//TODO: that sucks, use a better checksum algo
170	static inline uint32_t
171	rotate_right (uint32_t c, uint32_t count = 1)
172	{
173	return (c << (32 - count)) \| (c >> count);
174	}
175
176	static inline uint32_t
177	rotate_left (uint32_t c, uint32_t count = 1)
178	{
179	return (c >> (32 - count)) \| (c << count);
180	}
181
182	// Return abs(a-b)
183	// Both a and b must not have the most significant bit set
184	static inline uint32_t
185	upos_abs_diff (uint32_t a, uint32_t b)
186	{
187	long d1 = b - a;
188	long d2 = (d1 & (d1 >> 31)) << 1;
189
190	return d1 - d2; // == (b - d) - (a + d);
191	}
192
193	// Both a and b must not have the most significant bit set
194	static inline uint32_t
195	upos_min (uint32_t a, uint32_t b)
196	{
197	int32_t d = b - a;
198	d &= d >> 31;
199	return a + d;
200	}
201
202	// Both a and b must not have the most significant bit set
203	static inline uint32_t
204	upos_max (uint32_t a, uint32_t b)
205	{
206	int32_t d = b - a;
207	d &= d >> 31;
208	return b - d;
209	}
210
211	// this is much faster than crossfires original algorithm
212	// on modern cpus
213	inline int
214	isqrt (int n)
215	{
216	return (int)sqrtf ((float)n);
217	}
218
219	// this is only twice as fast as naive sqrtf (dxdy+dydy)
220	#if 0
221	// and has a max. error of 6 in the range -100..+100.
222	#else
223	// and has a max. error of 9 in the range -100..+100.
224	#endif
225	inline int
226	idistance (int dx, int dy)
227	{
228	unsigned int dx_ = abs (dx);
229	unsigned int dy_ = abs (dy);
230
231	#if 0
232	return dx_ > dy_
233	? (dx_ * 61685 + dy_ * 26870) >> 16
234	: (dy_ * 61685 + dx_ * 26870) >> 16;
235	#else
236	return dx_ + dy_ - min (dx_, dy_) * 5 / 8;
237	#endif
238	}
239
240	/*
241	* absdir(int): Returns a number between 1 and 8, which represent
242	* the "absolute" direction of a number (it actually takes care of
243	* "overflow" in previous calculations of a direction).
244	*/
245	inline int
246	absdir (int d)
247	{
248	return ((d - 1) & 7) + 1;
249	}
250
251	extern ssize_t slice_alloc; // statistics
252
253	void *salloc_ (int n) throw (std::bad_alloc);
254	void salloc_ (int n, void src) throw (std::bad_alloc);
255
256	// strictly the same as g_slice_alloc, but never returns 0
257	template<typename T>
258	inline T salloc (int n = 1) throw (std::bad_alloc) { return (T )salloc_ (n * sizeof (T)); }
259
260	// also copies src into the new area, like "memdup"
261	// if src is 0, clears the memory
262	template<typename T>
263	inline T salloc (int n, T src) throw (std::bad_alloc) { return (T )salloc_ (n sizeof (T), (void *)src); }
264
265	// clears the memory
266	template<typename T>
267	inline T salloc0(int n = 1) throw (std::bad_alloc) { return (T )salloc_ (n * sizeof (T), 0); }
268
269	// for symmetry
270	template<typename T>
271	inline void sfree (T *ptr, int n = 1) throw ()
272	{
273	if (expect_true (ptr))
274	{
275	slice_alloc -= n * sizeof (T);
276	if (DEBUG_POISON) memset (ptr, DEBUG_POISON, n * sizeof (T));
277	g_slice_free1 (n * sizeof (T), (void *)ptr);
278	assert (slice_alloc >= 0);//D
279	}
280	}
281
282	// nulls the pointer
283	template<typename T>
284	inline void sfree0 (T *&ptr, int n = 1) throw ()
285	{
286	sfree<T> (ptr, n);
287	ptr = 0;
288	}
289
290	// makes dynamically allocated objects zero-initialised
291	struct zero_initialised
292	{
293	void operator new (size_t s, void p)
294	{
295	memset (p, 0, s);
296	return p;
297	}
298
299	void *operator new (size_t s)
300	{
301	return salloc0<char> (s);
302	}
303
304	void *operator new[] (size_t s)
305	{
306	return salloc0<char> (s);
307	}
308
309	void operator delete (void *p, size_t s)
310	{
311	sfree ((char *)p, s);
312	}
313
314	void operator delete[] (void *p, size_t s)
315	{
316	sfree ((char *)p, s);
317	}
318	};
319
320	// makes dynamically allocated objects zero-initialised
321	struct slice_allocated
322	{
323	void operator new (size_t s, void p)
324	{
325	return p;
326	}
327
328	void *operator new (size_t s)
329	{
330	return salloc<char> (s);
331	}
332
333	void *operator new[] (size_t s)
334	{
335	return salloc<char> (s);
336	}
337
338	void operator delete (void *p, size_t s)
339	{
340	sfree ((char *)p, s);
341	}
342
343	void operator delete[] (void *p, size_t s)
344	{
345	sfree ((char *)p, s);
346	}
347	};
348
349	// a STL-compatible allocator that uses g_slice
350	// boy, this is verbose
351	template<typename Tp>
352	struct slice_allocator
353	{
354	typedef size_t size_type;
355	typedef ptrdiff_t difference_type;
356	typedef Tp *pointer;
357	typedef const Tp *const_pointer;
358	typedef Tp &reference;
359	typedef const Tp &const_reference;
360	typedef Tp value_type;
361
362	template <class U>
363	struct rebind
364	{
365	typedef slice_allocator<U> other;
366	};
367
368	slice_allocator () throw () { }
369	slice_allocator (const slice_allocator &) throw () { }
370	template<typename Tp2>
371	slice_allocator (const slice_allocator<Tp2> &) throw () { }
372
373	~slice_allocator () { }
374
375	pointer address (reference x) const { return &x; }
376	const_pointer address (const_reference x) const { return &x; }
377
378	pointer allocate (size_type n, const_pointer = 0)
379	{
380	return salloc<Tp> (n);
381	}
382
383	void deallocate (pointer p, size_type n)
384	{
385	sfree<Tp> (p, n);
386	}
387
388	size_type max_size () const throw ()
389	{
390	return size_t (-1) / sizeof (Tp);
391	}
392
393	void construct (pointer p, const Tp &val)
394	{
395	::new (p) Tp (val);
396	}
397
398	void destroy (pointer p)
399	{
400	p->~Tp ();
401	}
402	};
403
404	// P. L'Ecuyer, “Maximally Equidistributed Combined Tausworthe Generators”, Mathematics of Computation, 65, 213 (1996), 203–213.
405	// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps
406	// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
407	struct tausworthe_random_generator
408	{
409	uint32_t state [4];
410
411	void operator =(const tausworthe_random_generator &src)
412	{
413	state [0] = src.state [0];
414	state [1] = src.state [1];
415	state [2] = src.state [2];
416	state [3] = src.state [3];
417	}
418
419	void seed (uint32_t seed);
420	uint32_t next ();
421	};
422
423	// Xorshift RNGs, George Marsaglia
424	// http://www.jstatsoft.org/v08/i14/paper
425	// this one is about 40% faster than the tausworthe one above (i.e. not much),
426	// despite the inlining, and has the issue of only creating 2**32-1 numbers.
427	// see also http://www.iro.umontreal.ca/~lecuyer/myftp/papers/xorshift.pdf
428	struct xorshift_random_generator
429	{
430	uint32_t x, y;
431
432	void operator =(const xorshift_random_generator &src)
433	{
434	x = src.x;
435	y = src.y;
436	}
437
438	void seed (uint32_t seed)
439	{
440	x = seed;
441	y = seed * 69069U;
442	}
443
444	uint32_t next ()
445	{
446	uint32_t t = x ^ (x << 10);
447	x = y;
448	y = y ^ (y >> 13) ^ t ^ (t >> 10);
449	return y;
450	}
451	};
452
453	template<class generator>
454	struct random_number_generator : generator
455	{
456	// uniform distribution, 0 .. max (0, num - 1)
457	uint32_t operator ()(uint32_t num)
458	{
459	return !is_constant (num) ? get_range (num) // non-constant
460	: num & (num - 1) ? (this->next () * (uint64_t)num) >> 32U // constant, non-power-of-two
461	: this->next () & (num - 1); // constant, power-of-two
462	}
463
464	// return a number within (min .. max)
465	int operator () (int r_min, int r_max)
466	{
467	return is_constant (r_min) && is_constant (r_max) && r_min <= r_max
468	? r_min + operator ()(r_max - r_min + 1)
469	: get_range (r_min, r_max);
470	}
471
472	double operator ()()
473	{
474	return this->next () / (double)0xFFFFFFFFU;
475	}
476
477	protected:
478	uint32_t get_range (uint32_t r_max);
479	int get_range (int r_min, int r_max);
480	};
481
482	typedef random_number_generator<tausworthe_random_generator> rand_gen;
483
484	extern rand_gen rndm, rmg_rndm;
485
486	INTERFACE_CLASS (attachable)
487	struct refcnt_base
488	{
489	typedef int refcnt_t;
490	mutable refcnt_t ACC (RW, refcnt);
491
492	MTH void refcnt_inc () const { ++refcnt; }
493	MTH void refcnt_dec () const { --refcnt; }
494
495	refcnt_base () : refcnt (0) { }
496	};
497
498	// to avoid branches with more advanced compilers
499	extern refcnt_base::refcnt_t refcnt_dummy;
500
501	template<class T>
502	struct refptr
503	{
504	// p if not null
505	refcnt_base::refcnt_t *refcnt_ref () { return p ? &p->refcnt : &refcnt_dummy; }
506
507	void refcnt_dec ()
508	{
509	if (!is_constant (p))
510	--*refcnt_ref ();
511	else if (p)
512	--p->refcnt;
513	}
514
515	void refcnt_inc ()
516	{
517	if (!is_constant (p))
518	++*refcnt_ref ();
519	else if (p)
520	++p->refcnt;
521	}
522
523	T *p;
524
525	refptr () : p(0) { }
526	refptr (const refptr<T> &p) : p(p.p) { refcnt_inc (); }
527	refptr (T *p) : p(p) { refcnt_inc (); }
528	~refptr () { refcnt_dec (); }
529
530	const refptr<T> &operator =(T *o)
531	{
532	// if decrementing ever destroys we need to reverse the order here
533	refcnt_dec ();
534	p = o;
535	refcnt_inc ();
536	return *this;
537	}
538
539	const refptr<T> &operator =(const refptr<T> &o)
540	{
541	*this = o.p;
542	return *this;
543	}
544
545	T &operator * () const { return *p; }
546	T *operator ->() const { return p; }
547
548	operator T *() const { return p; }
549	};
550
551	typedef refptr<maptile> maptile_ptr;
552	typedef refptr<object> object_ptr;
553	typedef refptr<archetype> arch_ptr;
554	typedef refptr<client> client_ptr;
555	typedef refptr<player> player_ptr;
556
557	struct str_hash
558	{
559	std::size_t operator ()(const char *s) const
560	{
561	#if 0
562	uint32_t hash = 0;
563
564	/* use the one-at-a-time hash function, which supposedly is
565	* better than the djb2-like one used by perl5.005, but
566	* certainly is better then the bug used here before.
567	* see http://burtleburtle.net/bob/hash/doobs.html
568	*/
569	while (*s)
570	{
571	hash += *s++;
572	hash += hash << 10;
573	hash ^= hash >> 6;
574	}
575
576	hash += hash << 3;
577	hash ^= hash >> 11;
578	hash += hash << 15;
579	#else
580	// use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/)
581	// it is about twice as fast as the one-at-a-time one,
582	// with good distribution.
583	// FNV-1a is faster on many cpus because the multiplication
584	// runs concurrent with the looping logic.
585	uint32_t hash = 2166136261;
586
587	while (*s)
588	hash = (hash ^ s++) 16777619;
589	#endif
590
591	return hash;
592	}
593	};
594
595	struct str_equal
596	{
597	bool operator ()(const char a, const char b) const
598	{
599	return !strcmp (a, b);
600	}
601	};
602
603	// Mostly the same as std::vector, but insert/erase can reorder
604	// the elements, making append(=insert)/remove O(1) instead of O(n).
605	//
606	// NOTE: only some forms of erase are available
607	template<class T>
608	struct unordered_vector : std::vector<T, slice_allocator<T> >
609	{
610	typedef typename unordered_vector::iterator iterator;
611
612	void erase (unsigned int pos)
613	{
614	if (pos < this->size () - 1)
615	(this)[pos] = (this)[this->size () - 1];
616
617	this->pop_back ();
618	}
619
620	void erase (iterator i)
621	{
622	erase ((unsigned int )(i - this->begin ()));
623	}
624	};
625
626	// This container blends advantages of linked lists
627	// (efficiency) with vectors (random access) by
628	// by using an unordered vector and storing the vector
629	// index inside the object.
630	//
631	// + memory-efficient on most 64 bit archs
632	// + O(1) insert/remove
633	// + free unique (but varying) id for inserted objects
634	// + cache-friendly iteration
635	// - only works for pointers to structs
636	//
637	// NOTE: only some forms of erase/insert are available
638	typedef int object_vector_index;
639
640	template<class T, object_vector_index T::*indexmember>
641	struct object_vector : std::vector<T , slice_allocator<T > >
642	{
643	typedef typename object_vector::iterator iterator;
644
645	bool contains (const T *obj) const
646	{
647	return obj->*indexmember;
648	}
649
650	iterator find (const T *obj)
651	{
652	return obj->*indexmember
653	? this->begin () + obj->*indexmember - 1
654	: this->end ();
655	}
656
657	void push_back (T *obj)
658	{
659	std::vector<T , slice_allocator<T > >::push_back (obj);
660	obj->*indexmember = this->size ();
661	}
662
663	void insert (T *obj)
664	{
665	push_back (obj);
666	}
667
668	void insert (T &obj)
669	{
670	insert (&obj);
671	}
672
673	void erase (T *obj)
674	{
675	unsigned int pos = obj->*indexmember;
676	obj->*indexmember = 0;
677
678	if (pos < this->size ())
679	{
680	(this)[pos - 1] = (this)[this->size () - 1];
681	(this)[pos - 1]->indexmember = pos;
682	}
683
684	this->pop_back ();
685	}
686
687	void erase (T &obj)
688	{
689	erase (&obj);
690	}
691	};
692
693	// basically does what strncpy should do, but appends "..." to strings exceeding length
694	// returns the number of bytes actually used (including \0)
695	int assign (char dst, const char src, int maxsize);
696
697	// type-safe version of assign
698	template<int N>
699	inline int assign (char (&dst)[N], const char *src)
700	{
701	return assign ((char *)&dst, src, N);
702	}
703
704	typedef double tstamp;
705
706	// return current time as timestamp
707	tstamp now ();
708
709	int similar_direction (int a, int b);
710
711	// like sprintf, but returns a "static" buffer
712	const char format (const char format, ...);
713
714	// safety-check player input which will become object->msg
715	bool msg_is_safe (const char *msg);
716
717	/////////////////////////////////////////////////////////////////////////////
718	// threads, very very thin wrappers around pthreads
719
720	struct thread
721	{
722	pthread_t id;
723
724	void start (void (start_routine)(void ), void arg = 0);
725
726	void cancel ()
727	{
728	pthread_cancel (id);
729	}
730
731	void *join ()
732	{
733	void *ret;
734
735	if (pthread_join (id, &ret))
736	cleanup ("pthread_join failed", 1);
737
738	return ret;
739	}
740	};
741
742	// note that mutexes are not classes
743	typedef pthread_mutex_t smutex;
744
745	#if __linux && defined (PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP)
746	#define SMUTEX_INITIALISER PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP
747	#else
748	#define SMUTEX_INITIALISER PTHREAD_MUTEX_INITIALIZER
749	#endif
750
751	#define SMUTEX(name) smutex name = SMUTEX_INITIALISER
752	#define SMUTEX_LOCK(name) pthread_mutex_lock (&(name))
753	#define SMUTEX_UNLOCK(name) pthread_mutex_unlock (&(name))
754
755	typedef pthread_cond_t scond;
756
757	#define SCOND(name) scond name = PTHREAD_COND_INITIALIZER
758	#define SCOND_SIGNAL(name) pthread_cond_signal (&(name))
759	#define SCOND_BROADCAST(name) pthread_cond_broadcast (&(name))
760	#define SCOND_WAIT(name,mutex) pthread_cond_wait (&(name), &(mutex))
761
762	#endif
763