[ViewVC] Diff of: cvs/deliantra/server/include/util.h

Comparing deliantra/server/include/util.h (file contents):
Revision 1.85 by root, Thu Jan 1 20:49:48 2009 UTC vs.
Revision 1.129 by root, Sat Dec 1 20:22:13 2018 UTC

 /*
  * This file is part of Deliantra, the Roguelike Realtime MMORPG.
  *
+ * Copyright (©) 2017,2018 Marc Alexander Lehmann / the Deliantra team
- * Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team
+ * Copyright (©) 2005,2006,2007,2008,2009,2010,2011,2012,2013,2014,2015,2016 Marc Alexander Lehmann / Robin Redeker / the Deliantra team
  *
- * Deliantra is free software: you can redistribute it and/or modify
+ * Deliantra is free software: you can redistribute it and/or modify it under
- * it under the terms of the GNU General Public License as published by
+ * the terms of the Affero GNU General Public License as published by the
- * the Free Software Foundation, either version 3 of the License, or
+ * Free Software Foundation, either version 3 of the License, or (at your
- * (at your option) any later version.
+ * 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 GNU General Public License
+ * You should have received a copy of the Affero GNU General Public License
- * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ * 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
-#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 <flat_hash_map.hpp>
 #include <shstr.h>
 #include <traits.h>
 #if DEBUG_SALLOC
 # 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
 // 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);
+ecb_cold void cleanup (const char *cause, bool make_core = false);
-void fork_abort (const char *msg);
+ecb_cold 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 min (T a, U b) { return a < (T)b ? a : (T)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> static inline T max (T a, U b) { return a > (T)b ? a : (T)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); }
 // 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); }
+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; }
+//clashes with C++0x
+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 (val + div / 2) / div; }
+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 (val + div - 1) / div; }
+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 (val          ) / div; }
+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);
   int32_t d = b - a;
   d &= d >> 31;
   return b - d;
 }
-// this is much faster than crossfires original algorithm
+// 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_
 #else
   return dx_ + dy_ - min (dx_, dy_) * 5 / 8;
 #endif
 }
+// can be substantially faster than floor, if your value range allows for it
+template<typename T>
+inline T
+fastfloor (T x)
+{
+   return std::floor (x);
+}
+inline float
+fastfloor (float x)
+{
+  return sint32(x) - (x < 0);
+}
+inline double
+fastfloor (double x)
+{
+  return sint64(x) - (x < 0);
+}
 /*
  * 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).
  */
 absdir (int d)
 {
   return ((d - 1) & 7) + 1;
 }
+#define for_all_bits_sparse_32(mask, idxvar)      \
+  for (uint32_t idxvar, mask_ = mask;   \
+       mask_ && ((idxvar = ecb_ctz32 (mask_)), mask_ &= ~(1 << idxvar), 1);)
 extern ssize_t slice_alloc; // statistics
-void *salloc_ (int n)            throw (std::bad_alloc);
+void *salloc_ (int n);
-void *salloc_ (int n, void *src) throw (std::bad_alloc);
+void *salloc_ (int n, void *src);
 // 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));              }
+inline T *salloc (int n = 1)     { 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); }
+inline T *salloc (int n, T *src) { 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);           }
+inline T *salloc0(int n = 1)     { return (T *)salloc_ (n * sizeof (T), 0);           }
 // for symmetry
 template<typename T>
-inline void sfree (T *ptr, int n = 1) throw ()
+inline void sfree (T *ptr, int n = 1) noexcept
 {
   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 ()
+inline void sfree0 (T *&ptr, int n = 1) noexcept
 {
   sfree<T> (ptr, n);
   ptr = 0;
 }
   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 () noexcept { }
-  slice_allocator (const slice_allocator &) throw () { }
+  slice_allocator (const slice_allocator &) noexcept { }
   template<typename Tp2>
-  slice_allocator (const slice_allocator<Tp2> &) throw () { }
+  slice_allocator (const slice_allocator<Tp2> &) noexcept { }
   ~slice_allocator () { }
   pointer address (reference x) const { return &x; }
   const_pointer address (const_reference x) const { return &x; }
   void deallocate (pointer p, size_type n)
   {
     sfree<Tp> (p, n);
   }
-  size_type max_size () const throw ()
+  size_type max_size () const noexcept
   {
     return size_t (-1) / sizeof (Tp);
   }
   void construct (pointer p, const Tp &val)
   {
     p->~Tp ();
   }
 };
-// P. L'Ecuyer, “Maximally Equidistributed Combined Tausworthe Generators”, Mathematics of Computation, 65, 213 (1996), 203–213.
+// basically a memory area, but refcounted
-// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps
+struct refcnt_buf
-// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
-struct tausworthe_random_generator
 {
-  uint32_t state [4];
+  char *data;
-  void operator =(const tausworthe_random_generator &src)
+  refcnt_buf (size_t size = 0);
-  {
+  refcnt_buf (void *data, size_t size);
-    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);
+  refcnt_buf (const refcnt_buf &src)
-  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.
-struct xorshift_random_generator
-{
-  uint32_t x, y;
-  void operator =(const xorshift_random_generator &src)
   {
-    x = src.x;
+    data = src.data;
-    y = src.y;
+    inc ();
   }
-  void seed (uint32_t seed)
+  ~refcnt_buf ();
-  {
-    x = seed;
-    y = seed * 69069U;
-  }
-  uint32_t next ()
+  refcnt_buf &operator =(const refcnt_buf &src);
+  operator char *()
   {
-    uint32_t t = x ^ (x << 10);
-    x = y;
-    y = y ^ (y >> 13) ^ t ^ (t >> 10);
-    return y;
+    return data;
   }
-};
-template<class generator>
+  size_t size () const
-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
+    return _size ();
-         : 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);
+  enum {
-  int get_range (int r_min, int r_max);
+    overhead = sizeof (uint32_t) * 2
-};
+  };
-typedef random_number_generator<tausworthe_random_generator> rand_gen;
+  uint32_t &_size () const
+  {
+    return ((unsigned int *)data)[-2];
+  }
-extern rand_gen rndm, rmg_rndm;
+  uint32_t &_refcnt () const
+  {
+    return ((unsigned int *)data)[-1];
+  }
+  void _alloc (uint32_t size)
+  {
+    data = ((char *)salloc<char> (size + overhead)) + overhead;
+    _size   () = size;
+    _refcnt () = 1;
+  }
+  void _dealloc ();
+  void inc ()
+  {
+    ++_refcnt ();
+  }
+  void dec ()
+  {
+    if (!--_refcnt ())
+      _dealloc ();
+  }
+};
 INTERFACE_CLASS (attachable)
 struct refcnt_base
 {
   typedef int refcnt_t;
   // p if not null
   refcnt_base::refcnt_t *refcnt_ref () { return p ? &p->refcnt : &refcnt_dummy; }
   void refcnt_dec ()
   {
-    if (!is_constant (p))
+    if (!ecb_is_constant (p))
       --*refcnt_ref ();
     else if (p)
       --p->refcnt;
   }
   void refcnt_inc ()
   {
-    if (!is_constant (p))
+    if (!ecb_is_constant (p))
       ++*refcnt_ref ();
     else if (p)
       ++p->refcnt;
   }
 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.
+  // we modify the hash a bit to improve its distribution
+  uint32_t hash = STRHSH_NULL;
+  while (*s)
+    hash = (hash ^ *s++) * 16777619U;
+  return hash ^ (hash >> 16);
+}
+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
   {
-#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;
+    return strhsh (s);
   }
+  std::size_t operator ()(const shstr &s) const
+  {
+    return strhsh (s);
+  }
+  typedef ska::power_of_two_hash_policy hash_policy;
 };
 struct str_equal
 {
   bool operator ()(const char *a, const char *b) const
   }
 };
 // This container blends advantages of linked lists
 // (efficiency) with vectors (random access) by
-// by using an unordered vector and storing the vector
+// 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
     insert (&obj);
   }
   void erase (T *obj)
   {
-    unsigned int pos = obj->*indexmember;
+    object_vector_index pos = obj->*indexmember;
     obj->*indexmember = 0;
     if (pos < this->size ())
       {
         (*this)[pos - 1] = (*this)[this->size () - 1];
   {
     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)
-void assign (char *dst, const char *src, int maxlen);
+int assign (char *dst, const char *src, int maxsize);
 // type-safe version of assign
 template<int N>
-inline void assign (char (&dst)[N], const char *src)
+inline int assign (char (&dst)[N], const char *src)
 {
-  assign ((char *)&dst, src, N);
+  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
+// like v?sprintf, but returns a "static" buffer
-const char *format (const char *format, ...);
+char *vformat (const char *format, va_list ap);
+char *format (const char *format, ...) ecb_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

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing deliantra/server/include/util.h (file contents): Revision 1.85 by root, Thu Jan 1 20:49:48 2009 UTC vs. Revision 1.129 by root, Sat Dec 1 20:22:13 2018 UTC

Diff Legend

Comparing deliantra/server/include/util.h (file contents):
Revision 1.85 by root, Thu Jan 1 20:49:48 2009 UTC vs.
Revision 1.129 by root, Sat Dec 1 20:22:13 2018 UTC