server/include/util.h

/*
 * This file is part of Deliantra, the Roguelike Realtime MMORPG.
 * 
 * Copyright (©) 2005,2006,2007,2008,2009,2010 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__

#include <compiler.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

#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)

#if cplusplus_does_not_suck
// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm)
template<typename T, int N>
static inline int array_length (const T (&arr)[N])
{
  return N;
}
#else
#define array_length(name) (sizeof (name) / sizeof (name [0]))
#endif

// 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); }
template<>
inline sint16 sign (sint16 v) { return 1 - (sint16 (uint16 (v) >> 15) * 2); }
template<>
inline sint32 sign (sint32 v) { return 1 - (sint32 (uint32 (v) >> 31) * 2); }

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

template<typename T, typename U>
static inline T copysign (T a, U b) { return a > 0 ? b : -b; }

// 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;
}

template<> inline float  div (float  val, float  div) { return val / div; }
template<> inline double div (double val, double div) { return val / 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 crossfire's original algorithm
// on modern cpus
inline int
isqrt (int n)
{
  return (int)sqrtf ((float)n);
}

// this is kind of like the ^^ operator, if it would exist, without sequence point.
// more handy than it looks like, due to the implicit !! done on its arguments
inline bool
logical_xor (bool a, bool b)
{
  return a != b;
}

inline bool
logical_implies (bool a, bool b)
{
  return a <= b;
}

// 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;
}

// avoid ctz name because netbsd or freebsd spams it's namespace with it
#if GCC_VERSION(3,4)
static inline int least_significant_bit (uint32_t x)
{
  return __builtin_ctz (x);
}
#else
int least_significant_bit (uint32_t x);
#endif

#define for_all_bits_sparse_32(mask, idxvar)      \
  for (uint32_t idxvar, mask_ = mask;   \
       mask_ && ((idxvar = least_significant_bit (mask_)), mask_ &= ~(1 << idxvar), 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 ();
  }
};

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;
typedef refptr<region>    region_ptr;

#define STRHSH_NULL 2166136261

static inline uint32_t
strhsh (const char *s)
{
  // 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 concurrently with the looping logic.
  uint32_t hash = STRHSH_NULL;
  
  while (*s)
    hash = (hash ^ *s++) * 16777619U;

  return hash;
}

static inline uint32_t
memhsh (const char *s, size_t len)
{
  uint32_t hash = STRHSH_NULL;
  
  while (len--)
    hash = (hash ^ *s++) * 16777619U;

  return hash;
}

struct str_hash
{
  std::size_t operator ()(const char *s) const
  {
    return strhsh (s);
  }

  std::size_t operator ()(const shstr &s) const
  {
    return strhsh (s);
  }
};

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);
  }
};

/////////////////////////////////////////////////////////////////////////////

// something like a vector or stack, but without
// out of bounds checking
template<typename T>
struct fixed_stack
{
  T *data;
  int size;
  int max;

  fixed_stack ()
  : size (0), data (0)
  {
  }

  fixed_stack (int max)
  : size (0), max (max)
  {
    data = salloc<T> (max);
  }

  void reset (int new_max)
  {
    sfree (data, max);
    size = 0;
    max = new_max;
    data = salloc<T> (max);
  }

  void free ()
  {
    sfree (data, max);
    data = 0;
  }

  ~fixed_stack ()
  {
    sfree (data, max);
  }

  T &operator[](int idx)
  {
    return data [idx];
  }

  void push (T v)
  {
    data [size++] = v;
  }

  T &pop ()
  {
    return data [--size];
  }

  T remove (int idx)
  {
    T v = data [idx];

    data [idx] = data [--size];

    return v;
  }
};

/////////////////////////////////////////////////////////////////////////////

// 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 v?sprintf, but returns a "static" buffer
char *vformat (const char *format, va_list ap);
char *format (const char *format, ...) attribute ((format (printf, 1, 2)));

// 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:	1.111
Committed:	Tue Jul 6 20:00:46 2010 UTC (13 years, 10 months ago) by root
Content type:	text/plain
Branch:	MAIN
Changes since 1.110:	+68 -0 lines
Log Message:	move layout stuff to it's own include file
#	Content
1	/*
2	* This file is part of Deliantra, the Roguelike Realtime MMORPG.
3	*
4	* Copyright (©) 2005,2006,2007,2008,2009,2010 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	#include <compiler.h>
27
28	#define DEBUG_POISON 0x00 // poison memory before freeing it if != 0
29	#define DEBUG_SALLOC 0 // add a debug wrapper around all sallocs
30	#define PREFER_MALLOC 0 // use malloc and not the slice allocator
31
32	#include <pthread.h>
33
34	#include <cstddef>
35	#include <cmath>
36	#include <new>
37	#include <vector>
38
39	#include <glib.h>
40
41	#include <shstr.h>
42	#include <traits.h>
43
44	#if DEBUG_SALLOC
45	# define g_slice_alloc0(s) debug_slice_alloc0(s)
46	# define g_slice_alloc(s) debug_slice_alloc(s)
47	# define g_slice_free1(s,p) debug_slice_free1(s,p)
48	void *g_slice_alloc (unsigned long size);
49	void *g_slice_alloc0 (unsigned long size);
50	void g_slice_free1 (unsigned long size, void *ptr);
51	#elif PREFER_MALLOC
52	# define g_slice_alloc0(s) calloc (1, (s))
53	# define g_slice_alloc(s) malloc ((s))
54	# define g_slice_free1(s,p) free ((p))
55	#endif
56
57	// use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever)
58	#define auto(var,expr) decltype(expr) var = (expr)
59
60	#if cplusplus_does_not_suck
61	// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm)
62	template<typename T, int N>
63	static inline int array_length (const T (&arr)[N])
64	{
65	return N;
66	}
67	#else
68	#define array_length(name) (sizeof (name) / sizeof (name [0]))
69	#endif
70
71	// very ugly macro that basically declares and initialises a variable
72	// that is in scope for the next statement only
73	// works only for stuff that can be assigned 0 and converts to false
74	// (note: works great for pointers)
75	// most ugly macro I ever wrote
76	#define statementvar(type, name, value) if (type name = 0) { } else if (((name) = (value)), 1)
77
78	// in range including end
79	#define IN_RANGE_INC(val,beg,end) \
80	((unsigned int)(val) - (unsigned int)(beg) <= (unsigned int)(end) - (unsigned int)(beg))
81
82	// in range excluding end
83	#define IN_RANGE_EXC(val,beg,end) \
84	((unsigned int)(val) - (unsigned int)(beg) < (unsigned int)(end) - (unsigned int)(beg))
85
86	void cleanup (const char *cause, bool make_core = false);
87	void fork_abort (const char *msg);
88
89	// rationale for using (U) not (T) is to reduce signed/unsigned issues,
90	// as a is often a constant while b is the variable. it is still a bug, though.
91	template<typename T, typename U> static inline T min (T a, U b) { return (U)a < b ? (U)a : b; }
92	template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; }
93	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; }
94
95	template<typename T, typename U> static inline void min_it (T &v, U m) { v = min (v, (T)m); }
96	template<typename T, typename U> static inline void max_it (T &v, U m) { v = max (v, (T)m); }
97	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); }
98
99	template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; }
100
101	template<typename T, typename U, typename V> static inline T min (T a, U b, V c) { return min (a, min (b, c)); }
102	template<typename T, typename U, typename V> static inline T max (T a, U b, V c) { return max (a, max (b, c)); }
103
104	// sign returns -1 or +1
105	template<typename T>
106	static inline T sign (T v) { return v < 0 ? -1 : +1; }
107	// relies on 2c representation
108	template<>
109	inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); }
110	template<>
111	inline sint16 sign (sint16 v) { return 1 - (sint16 (uint16 (v) >> 15) * 2); }
112	template<>
113	inline sint32 sign (sint32 v) { return 1 - (sint32 (uint32 (v) >> 31) * 2); }
114
115	// sign0 returns -1, 0 or +1
116	template<typename T>
117	static inline T sign0 (T v) { return v ? sign (v) : 0; }
118
119	template<typename T, typename U>
120	static inline T copysign (T a, U b) { return a > 0 ? b : -b; }
121
122	// div* only work correctly for div > 0
123	// div, with correct rounding (< 0.5 downwards, >=0.5 upwards)
124	template<typename T> static inline T div (T val, T div)
125	{
126	return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div;
127	}
128
129	template<> inline float div (float val, float div) { return val / div; }
130	template<> inline double div (double val, double div) { return val / div; }
131
132	// div, round-up
133	template<typename T> static inline T div_ru (T val, T div)
134	{
135	return expect_false (val < 0) ? - ((-val ) / div) : (val + div - 1) / div;
136	}
137	// div, round-down
138	template<typename T> static inline T div_rd (T val, T div)
139	{
140	return expect_false (val < 0) ? - ((-val + (div - 1) ) / div) : (val ) / div;
141	}
142
143	// lerp* only work correctly for min_in < max_in
144	// Linear intERPolate, scales val from min_in..max_in to min_out..max_out
145	template<typename T>
146	static inline T
147	lerp (T val, T min_in, T max_in, T min_out, T max_out)
148	{
149	return min_out + div <T> ((val - min_in) * (max_out - min_out), max_in - min_in);
150	}
151
152	// lerp, round-down
153	template<typename T>
154	static inline T
155	lerp_rd (T val, T min_in, T max_in, T min_out, T max_out)
156	{
157	return min_out + div_rd<T> ((val - min_in) * (max_out - min_out), max_in - min_in);
158	}
159
160	// lerp, round-up
161	template<typename T>
162	static inline T
163	lerp_ru (T val, T min_in, T max_in, T min_out, T max_out)
164	{
165	return min_out + div_ru<T> ((val - min_in) * (max_out - min_out), max_in - min_in);
166	}
167
168	// lots of stuff taken from FXT
169
170	/* Rotate right. This is used in various places for checksumming */
171	//TODO: that sucks, use a better checksum algo
172	static inline uint32_t
173	rotate_right (uint32_t c, uint32_t count = 1)
174	{
175	return (c << (32 - count)) \| (c >> count);
176	}
177
178	static inline uint32_t
179	rotate_left (uint32_t c, uint32_t count = 1)
180	{
181	return (c >> (32 - count)) \| (c << count);
182	}
183
184	// Return abs(a-b)
185	// Both a and b must not have the most significant bit set
186	static inline uint32_t
187	upos_abs_diff (uint32_t a, uint32_t b)
188	{
189	long d1 = b - a;
190	long d2 = (d1 & (d1 >> 31)) << 1;
191
192	return d1 - d2; // == (b - d) - (a + d);
193	}
194
195	// Both a and b must not have the most significant bit set
196	static inline uint32_t
197	upos_min (uint32_t a, uint32_t b)
198	{
199	int32_t d = b - a;
200	d &= d >> 31;
201	return a + d;
202	}
203
204	// Both a and b must not have the most significant bit set
205	static inline uint32_t
206	upos_max (uint32_t a, uint32_t b)
207	{
208	int32_t d = b - a;
209	d &= d >> 31;
210	return b - d;
211	}
212
213	// this is much faster than crossfire's original algorithm
214	// on modern cpus
215	inline int
216	isqrt (int n)
217	{
218	return (int)sqrtf ((float)n);
219	}
220
221	// this is kind of like the ^^ operator, if it would exist, without sequence point.
222	// more handy than it looks like, due to the implicit !! done on its arguments
223	inline bool
224	logical_xor (bool a, bool b)
225	{
226	return a != b;
227	}
228
229	inline bool
230	logical_implies (bool a, bool b)
231	{
232	return a <= b;
233	}
234
235	// this is only twice as fast as naive sqrtf (dxdy+dydy)
236	#if 0
237	// and has a max. error of 6 in the range -100..+100.
238	#else
239	// and has a max. error of 9 in the range -100..+100.
240	#endif
241	inline int
242	idistance (int dx, int dy)
243	{
244	unsigned int dx_ = abs (dx);
245	unsigned int dy_ = abs (dy);
246
247	#if 0
248	return dx_ > dy_
249	? (dx_ * 61685 + dy_ * 26870) >> 16
250	: (dy_ * 61685 + dx_ * 26870) >> 16;
251	#else
252	return dx_ + dy_ - min (dx_, dy_) * 5 / 8;
253	#endif
254	}
255
256	/*
257	* absdir(int): Returns a number between 1 and 8, which represent
258	* the "absolute" direction of a number (it actually takes care of
259	* "overflow" in previous calculations of a direction).
260	*/
261	inline int
262	absdir (int d)
263	{
264	return ((d - 1) & 7) + 1;
265	}
266
267	// avoid ctz name because netbsd or freebsd spams it's namespace with it
268	#if GCC_VERSION(3,4)
269	static inline int least_significant_bit (uint32_t x)
270	{
271	return __builtin_ctz (x);
272	}
273	#else
274	int least_significant_bit (uint32_t x);
275	#endif
276
277	#define for_all_bits_sparse_32(mask, idxvar) \
278	for (uint32_t idxvar, mask_ = mask; \
279	mask_ && ((idxvar = least_significant_bit (mask_)), mask_ &= ~(1 << idxvar), 1);)
280
281	extern ssize_t slice_alloc; // statistics
282
283	void *salloc_ (int n) throw (std::bad_alloc);
284	void salloc_ (int n, void src) throw (std::bad_alloc);
285
286	// strictly the same as g_slice_alloc, but never returns 0
287	template<typename T>
288	inline T salloc (int n = 1) throw (std::bad_alloc) { return (T )salloc_ (n * sizeof (T)); }
289
290	// also copies src into the new area, like "memdup"
291	// if src is 0, clears the memory
292	template<typename T>
293	inline T salloc (int n, T src) throw (std::bad_alloc) { return (T )salloc_ (n sizeof (T), (void *)src); }
294
295	// clears the memory
296	template<typename T>
297	inline T salloc0(int n = 1) throw (std::bad_alloc) { return (T )salloc_ (n * sizeof (T), 0); }
298
299	// for symmetry
300	template<typename T>
301	inline void sfree (T *ptr, int n = 1) throw ()
302	{
303	if (expect_true (ptr))
304	{
305	slice_alloc -= n * sizeof (T);
306	if (DEBUG_POISON) memset (ptr, DEBUG_POISON, n * sizeof (T));
307	g_slice_free1 (n * sizeof (T), (void *)ptr);
308	assert (slice_alloc >= 0);//D
309	}
310	}
311
312	// nulls the pointer
313	template<typename T>
314	inline void sfree0 (T *&ptr, int n = 1) throw ()
315	{
316	sfree<T> (ptr, n);
317	ptr = 0;
318	}
319
320	// makes dynamically allocated objects zero-initialised
321	struct zero_initialised
322	{
323	void operator new (size_t s, void p)
324	{
325	memset (p, 0, s);
326	return p;
327	}
328
329	void *operator new (size_t s)
330	{
331	return salloc0<char> (s);
332	}
333
334	void *operator new[] (size_t s)
335	{
336	return salloc0<char> (s);
337	}
338
339	void operator delete (void *p, size_t s)
340	{
341	sfree ((char *)p, s);
342	}
343
344	void operator delete[] (void *p, size_t s)
345	{
346	sfree ((char *)p, s);
347	}
348	};
349
350	// makes dynamically allocated objects zero-initialised
351	struct slice_allocated
352	{
353	void operator new (size_t s, void p)
354	{
355	return p;
356	}
357
358	void *operator new (size_t s)
359	{
360	return salloc<char> (s);
361	}
362
363	void *operator new[] (size_t s)
364	{
365	return salloc<char> (s);
366	}
367
368	void operator delete (void *p, size_t s)
369	{
370	sfree ((char *)p, s);
371	}
372
373	void operator delete[] (void *p, size_t s)
374	{
375	sfree ((char *)p, s);
376	}
377	};
378
379	// a STL-compatible allocator that uses g_slice
380	// boy, this is verbose
381	template<typename Tp>
382	struct slice_allocator
383	{
384	typedef size_t size_type;
385	typedef ptrdiff_t difference_type;
386	typedef Tp *pointer;
387	typedef const Tp *const_pointer;
388	typedef Tp &reference;
389	typedef const Tp &const_reference;
390	typedef Tp value_type;
391
392	template <class U>
393	struct rebind
394	{
395	typedef slice_allocator<U> other;
396	};
397
398	slice_allocator () throw () { }
399	slice_allocator (const slice_allocator &) throw () { }
400	template<typename Tp2>
401	slice_allocator (const slice_allocator<Tp2> &) throw () { }
402
403	~slice_allocator () { }
404
405	pointer address (reference x) const { return &x; }
406	const_pointer address (const_reference x) const { return &x; }
407
408	pointer allocate (size_type n, const_pointer = 0)
409	{
410	return salloc<Tp> (n);
411	}
412
413	void deallocate (pointer p, size_type n)
414	{
415	sfree<Tp> (p, n);
416	}
417
418	size_type max_size () const throw ()
419	{
420	return size_t (-1) / sizeof (Tp);
421	}
422
423	void construct (pointer p, const Tp &val)
424	{
425	::new (p) Tp (val);
426	}
427
428	void destroy (pointer p)
429	{
430	p->~Tp ();
431	}
432	};
433
434	INTERFACE_CLASS (attachable)
435	struct refcnt_base
436	{
437	typedef int refcnt_t;
438	mutable refcnt_t ACC (RW, refcnt);
439
440	MTH void refcnt_inc () const { ++refcnt; }
441	MTH void refcnt_dec () const { --refcnt; }
442
443	refcnt_base () : refcnt (0) { }
444	};
445
446	// to avoid branches with more advanced compilers
447	extern refcnt_base::refcnt_t refcnt_dummy;
448
449	template<class T>
450	struct refptr
451	{
452	// p if not null
453	refcnt_base::refcnt_t *refcnt_ref () { return p ? &p->refcnt : &refcnt_dummy; }
454
455	void refcnt_dec ()
456	{
457	if (!is_constant (p))
458	--*refcnt_ref ();
459	else if (p)
460	--p->refcnt;
461	}
462
463	void refcnt_inc ()
464	{
465	if (!is_constant (p))
466	++*refcnt_ref ();
467	else if (p)
468	++p->refcnt;
469	}
470
471	T *p;
472
473	refptr () : p(0) { }
474	refptr (const refptr<T> &p) : p(p.p) { refcnt_inc (); }
475	refptr (T *p) : p(p) { refcnt_inc (); }
476	~refptr () { refcnt_dec (); }
477
478	const refptr<T> &operator =(T *o)
479	{
480	// if decrementing ever destroys we need to reverse the order here
481	refcnt_dec ();
482	p = o;
483	refcnt_inc ();
484	return *this;
485	}
486
487	const refptr<T> &operator =(const refptr<T> &o)
488	{
489	*this = o.p;
490	return *this;
491	}
492
493	T &operator * () const { return *p; }
494	T *operator ->() const { return p; }
495
496	operator T *() const { return p; }
497	};
498
499	typedef refptr<maptile> maptile_ptr;
500	typedef refptr<object> object_ptr;
501	typedef refptr<archetype> arch_ptr;
502	typedef refptr<client> client_ptr;
503	typedef refptr<player> player_ptr;
504	typedef refptr<region> region_ptr;
505
506	#define STRHSH_NULL 2166136261
507
508	static inline uint32_t
509	strhsh (const char *s)
510	{
511	// use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/)
512	// it is about twice as fast as the one-at-a-time one,
513	// with good distribution.
514	// FNV-1a is faster on many cpus because the multiplication
515	// runs concurrently with the looping logic.
516	uint32_t hash = STRHSH_NULL;
517
518	while (*s)
519	hash = (hash ^ s++) 16777619U;
520
521	return hash;
522	}
523
524	static inline uint32_t
525	memhsh (const char *s, size_t len)
526	{
527	uint32_t hash = STRHSH_NULL;
528
529	while (len--)
530	hash = (hash ^ s++) 16777619U;
531
532	return hash;
533	}
534
535	struct str_hash
536	{
537	std::size_t operator ()(const char *s) const
538	{
539	return strhsh (s);
540	}
541
542	std::size_t operator ()(const shstr &s) const
543	{
544	return strhsh (s);
545	}
546	};
547
548	struct str_equal
549	{
550	bool operator ()(const char a, const char b) const
551	{
552	return !strcmp (a, b);
553	}
554	};
555
556	// Mostly the same as std::vector, but insert/erase can reorder
557	// the elements, making append(=insert)/remove O(1) instead of O(n).
558	//
559	// NOTE: only some forms of erase are available
560	template<class T>
561	struct unordered_vector : std::vector<T, slice_allocator<T> >
562	{
563	typedef typename unordered_vector::iterator iterator;
564
565	void erase (unsigned int pos)
566	{
567	if (pos < this->size () - 1)
568	(this)[pos] = (this)[this->size () - 1];
569
570	this->pop_back ();
571	}
572
573	void erase (iterator i)
574	{
575	erase ((unsigned int )(i - this->begin ()));
576	}
577	};
578
579	// This container blends advantages of linked lists
580	// (efficiency) with vectors (random access) by
581	// by using an unordered vector and storing the vector
582	// index inside the object.
583	//
584	// + memory-efficient on most 64 bit archs
585	// + O(1) insert/remove
586	// + free unique (but varying) id for inserted objects
587	// + cache-friendly iteration
588	// - only works for pointers to structs
589	//
590	// NOTE: only some forms of erase/insert are available
591	typedef int object_vector_index;
592
593	template<class T, object_vector_index T::*indexmember>
594	struct object_vector : std::vector<T , slice_allocator<T > >
595	{
596	typedef typename object_vector::iterator iterator;
597
598	bool contains (const T *obj) const
599	{
600	return obj->*indexmember;
601	}
602
603	iterator find (const T *obj)
604	{
605	return obj->*indexmember
606	? this->begin () + obj->*indexmember - 1
607	: this->end ();
608	}
609
610	void push_back (T *obj)
611	{
612	std::vector<T , slice_allocator<T > >::push_back (obj);
613	obj->*indexmember = this->size ();
614	}
615
616	void insert (T *obj)
617	{
618	push_back (obj);
619	}
620
621	void insert (T &obj)
622	{
623	insert (&obj);
624	}
625
626	void erase (T *obj)
627	{
628	unsigned int pos = obj->*indexmember;
629	obj->*indexmember = 0;
630
631	if (pos < this->size ())
632	{
633	(this)[pos - 1] = (this)[this->size () - 1];
634	(this)[pos - 1]->indexmember = pos;
635	}
636
637	this->pop_back ();
638	}
639
640	void erase (T &obj)
641	{
642	erase (&obj);
643	}
644	};
645
646	/////////////////////////////////////////////////////////////////////////////
647
648	// something like a vector or stack, but without
649	// out of bounds checking
650	template<typename T>
651	struct fixed_stack
652	{
653	T *data;
654	int size;
655	int max;
656
657	fixed_stack ()
658	: size (0), data (0)
659	{
660	}
661
662	fixed_stack (int max)
663	: size (0), max (max)
664	{
665	data = salloc<T> (max);
666	}
667
668	void reset (int new_max)
669	{
670	sfree (data, max);
671	size = 0;
672	max = new_max;
673	data = salloc<T> (max);
674	}
675
676	void free ()
677	{
678	sfree (data, max);
679	data = 0;
680	}
681
682	~fixed_stack ()
683	{
684	sfree (data, max);
685	}
686
687	T &operator[](int idx)
688	{
689	return data [idx];
690	}
691
692	void push (T v)
693	{
694	data [size++] = v;
695	}
696
697	T &pop ()
698	{
699	return data [--size];
700	}
701
702	T remove (int idx)
703	{
704	T v = data [idx];
705
706	data [idx] = data [--size];
707
708	return v;
709	}
710	};
711
712	/////////////////////////////////////////////////////////////////////////////
713
714	// basically does what strncpy should do, but appends "..." to strings exceeding length
715	// returns the number of bytes actually used (including \0)
716	int assign (char dst, const char src, int maxsize);
717
718	// type-safe version of assign
719	template<int N>
720	inline int assign (char (&dst)[N], const char *src)
721	{
722	return assign ((char *)&dst, src, N);
723	}
724
725	typedef double tstamp;
726
727	// return current time as timestamp
728	tstamp now ();
729
730	int similar_direction (int a, int b);
731
732	// like v?sprintf, but returns a "static" buffer
733	char vformat (const char format, va_list ap);
734	char format (const char format, ...) attribute ((format (printf, 1, 2)));
735
736	// safety-check player input which will become object->msg
737	bool msg_is_safe (const char *msg);
738
739	/////////////////////////////////////////////////////////////////////////////
740	// threads, very very thin wrappers around pthreads
741
742	struct thread
743	{
744	pthread_t id;
745
746	void start (void (start_routine)(void ), void arg = 0);
747
748	void cancel ()
749	{
750	pthread_cancel (id);
751	}
752
753	void *join ()
754	{
755	void *ret;
756
757	if (pthread_join (id, &ret))
758	cleanup ("pthread_join failed", 1);
759
760	return ret;
761	}
762	};
763
764	// note that mutexes are not classes
765	typedef pthread_mutex_t smutex;
766
767	#if __linux && defined (PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP)
768	#define SMUTEX_INITIALISER PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP
769	#else
770	#define SMUTEX_INITIALISER PTHREAD_MUTEX_INITIALIZER
771	#endif
772
773	#define SMUTEX(name) smutex name = SMUTEX_INITIALISER
774	#define SMUTEX_LOCK(name) pthread_mutex_lock (&(name))
775	#define SMUTEX_UNLOCK(name) pthread_mutex_unlock (&(name))
776
777	typedef pthread_cond_t scond;
778
779	#define SCOND(name) scond name = PTHREAD_COND_INITIALIZER
780	#define SCOND_SIGNAL(name) pthread_cond_signal (&(name))
781	#define SCOND_BROADCAST(name) pthread_cond_broadcast (&(name))
782	#define SCOND_WAIT(name,mutex) pthread_cond_wait (&(name), &(mutex))
783
784	#endif
785