gccxml_gem 0.9-x86-darwin-9
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- data/Rakefile +75 -0
- data/bin/gccxml +0 -0
- data/bin/gccxml_cc1plus +0 -0
- data/gccxml.rb +44 -0
- data/share/doc/gccxml-0.9/Copyright.txt +55 -0
- data/share/doc/gccxml-0.9/gccxml.html +168 -0
- data/share/doc/gccxml-0.9/gccxml.txt +293 -0
- data/share/gccxml-0.9/GCC/2.95/algorithm +76 -0
- data/share/gccxml-0.9/GCC/2.95/bitset +17 -0
- data/share/gccxml-0.9/GCC/2.95/cctype +24 -0
- data/share/gccxml-0.9/GCC/2.95/clocale +14 -0
- data/share/gccxml-0.9/GCC/2.95/cmath +33 -0
- data/share/gccxml-0.9/GCC/2.95/complex +38 -0
- data/share/gccxml-0.9/GCC/2.95/csetjmp +13 -0
- data/share/gccxml-0.9/GCC/2.95/csignal +14 -0
- data/share/gccxml-0.9/GCC/2.95/cstdarg +12 -0
- data/share/gccxml-0.9/GCC/2.95/cstddef +13 -0
- data/share/gccxml-0.9/GCC/2.95/cstdio +55 -0
- data/share/gccxml-0.9/GCC/2.95/cstdlib +66 -0
- data/share/gccxml-0.9/GCC/2.95/cstring +34 -0
- data/share/gccxml-0.9/GCC/2.95/ctime +24 -0
- data/share/gccxml-0.9/GCC/2.95/cwchar +65 -0
- data/share/gccxml-0.9/GCC/2.95/cwctype +31 -0
- data/share/gccxml-0.9/GCC/2.95/deque +19 -0
- data/share/gccxml-0.9/GCC/2.95/exception +20 -0
- data/share/gccxml-0.9/GCC/2.95/fstream +23 -0
- data/share/gccxml-0.9/GCC/2.95/functional +64 -0
- data/share/gccxml-0.9/GCC/2.95/gccxml_bitset +1066 -0
- data/share/gccxml-0.9/GCC/2.95/iomanip +20 -0
- data/share/gccxml-0.9/GCC/2.95/iosfwd +20 -0
- data/share/gccxml-0.9/GCC/2.95/iostream +27 -0
- data/share/gccxml-0.9/GCC/2.95/iterator +39 -0
- data/share/gccxml-0.9/GCC/2.95/list +19 -0
- data/share/gccxml-0.9/GCC/2.95/map +20 -0
- data/share/gccxml-0.9/GCC/2.95/memory +21 -0
- data/share/gccxml-0.9/GCC/2.95/new +13 -0
- data/share/gccxml-0.9/GCC/2.95/numeric +15 -0
- data/share/gccxml-0.9/GCC/2.95/pthread.h +16 -0
- data/share/gccxml-0.9/GCC/2.95/queue +20 -0
- data/share/gccxml-0.9/GCC/2.95/set +20 -0
- data/share/gccxml-0.9/GCC/2.95/sstream +24 -0
- data/share/gccxml-0.9/GCC/2.95/stack +19 -0
- data/share/gccxml-0.9/GCC/2.95/std/bastring.cc +524 -0
- data/share/gccxml-0.9/GCC/2.95/std/complext.h +397 -0
- data/share/gccxml-0.9/GCC/2.95/std/dcomplex.h +92 -0
- data/share/gccxml-0.9/GCC/2.95/std/fcomplex.h +88 -0
- data/share/gccxml-0.9/GCC/2.95/std/gslice_array.h +170 -0
- data/share/gccxml-0.9/GCC/2.95/std/indirect_array.h +157 -0
- data/share/gccxml-0.9/GCC/2.95/std/ldcomplex.h +96 -0
- data/share/gccxml-0.9/GCC/2.95/std/mask_array.h +154 -0
- data/share/gccxml-0.9/GCC/2.95/std/slice_array.h +156 -0
- data/share/gccxml-0.9/GCC/2.95/std/std_valarray.h +728 -0
- data/share/gccxml-0.9/GCC/2.95/std/straits.h +162 -0
- data/share/gccxml-0.9/GCC/2.95/std/valarray_meta.h +1035 -0
- data/share/gccxml-0.9/GCC/2.95/stdexcept +17 -0
- data/share/gccxml-0.9/GCC/2.95/stl_alloc.h +1057 -0
- data/share/gccxml-0.9/GCC/2.95/stl_bvector.h +836 -0
- data/share/gccxml-0.9/GCC/2.95/stl_deque.h +1699 -0
- data/share/gccxml-0.9/GCC/2.95/stl_list.h +843 -0
- data/share/gccxml-0.9/GCC/2.95/stl_tree.h +1331 -0
- data/share/gccxml-0.9/GCC/2.95/stl_vector.h +828 -0
- data/share/gccxml-0.9/GCC/2.95/string +26 -0
- data/share/gccxml-0.9/GCC/2.95/strstream +23 -0
- data/share/gccxml-0.9/GCC/2.95/typeinfo +11 -0
- data/share/gccxml-0.9/GCC/2.95/utility +25 -0
- data/share/gccxml-0.9/GCC/2.95/valarray +52 -0
- data/share/gccxml-0.9/GCC/2.95/vector +19 -0
- data/share/gccxml-0.9/GCC/2.96/sstream +305 -0
- data/share/gccxml-0.9/GCC/3.0/pthread.h +16 -0
- data/share/gccxml-0.9/GCC/3.1/pthread.h +16 -0
- data/share/gccxml-0.9/GCC/3.2/bits/fstream.tcc +500 -0
- data/share/gccxml-0.9/GCC/3.2/bits/istream.tcc +1207 -0
- data/share/gccxml-0.9/GCC/3.2/bits/locale_facets.h +1810 -0
- data/share/gccxml-0.9/GCC/3.2/bits/locale_facets.tcc +2397 -0
- data/share/gccxml-0.9/GCC/3.2/bits/messages_members.h +108 -0
- data/share/gccxml-0.9/GCC/3.2/bits/ostream.tcc +713 -0
- data/share/gccxml-0.9/GCC/3.2/bits/sstream.tcc +241 -0
- data/share/gccxml-0.9/GCC/3.2/bits/stl_deque.h +1682 -0
- data/share/gccxml-0.9/GCC/3.2/bits/stl_list.h +989 -0
- data/share/gccxml-0.9/GCC/3.2/bits/stl_tree.h +1462 -0
- data/share/gccxml-0.9/GCC/3.2/bits/stl_vector.h +1085 -0
- data/share/gccxml-0.9/GCC/3.2/bits/valarray_meta.h +1063 -0
- data/share/gccxml-0.9/GCC/3.2/fstream +579 -0
- data/share/gccxml-0.9/GCC/3.2/pthread.h +16 -0
- data/share/gccxml-0.9/GCC/3.2/sstream +384 -0
- data/share/gccxml-0.9/GCC/3.3/bits/fstream.tcc +530 -0
- data/share/gccxml-0.9/GCC/3.3/bits/list.tcc +378 -0
- data/share/gccxml-0.9/GCC/3.3/bits/locale_facets.h +2050 -0
- data/share/gccxml-0.9/GCC/3.3/bits/messages_members.h +108 -0
- data/share/gccxml-0.9/GCC/3.3/bits/sstream.tcc +243 -0
- data/share/gccxml-0.9/GCC/3.3/bits/stl_deque.h +1603 -0
- data/share/gccxml-0.9/GCC/3.3/bits/stl_list.h +1167 -0
- data/share/gccxml-0.9/GCC/3.3/bits/stl_tree.h +1462 -0
- data/share/gccxml-0.9/GCC/3.3/bits/stl_vector.h +992 -0
- data/share/gccxml-0.9/GCC/3.3/bits/valarray_meta.h +1135 -0
- data/share/gccxml-0.9/GCC/3.3/fstream +842 -0
- data/share/gccxml-0.9/GCC/3.3/gccxml_builtins.h +22 -0
- data/share/gccxml-0.9/GCC/3.3/sstream +638 -0
- data/share/gccxml-0.9/GCC/3.4/gccxml_builtins.h +91 -0
- data/share/gccxml-0.9/GCC/4.0/gccxml_builtins.h +128 -0
- data/share/gccxml-0.9/GCC/4.1/bits/gthr-default.h +618 -0
- data/share/gccxml-0.9/GCC/4.1/gccxml_builtins.h +131 -0
- data/share/gccxml-0.9/GCC/4.2/gccxml_builtins.h +131 -0
- data/share/gccxml-0.9/GCC/find_flags +105 -0
- data/share/gccxml-0.9/Intel/find_flags +56 -0
- data/share/gccxml-0.9/Intel/pthread.h +16 -0
- data/share/gccxml-0.9/MIPSpro/7.3/exception +9 -0
- data/share/gccxml-0.9/MIPSpro/7.3/gccxml_mpro_internals.h +21 -0
- data/share/gccxml-0.9/MIPSpro/7.3/iomanip +161 -0
- data/share/gccxml-0.9/MIPSpro/7.3/ostream +9 -0
- data/share/gccxml-0.9/MIPSpro/7.3/stddef.h +9 -0
- data/share/gccxml-0.9/MIPSpro/7.3/stl_config.h +9 -0
- data/share/gccxml-0.9/MIPSpro/7.3/stl_locale.h +17 -0
- data/share/gccxml-0.9/MIPSpro/7.3/stl_monetary.h +14 -0
- data/share/gccxml-0.9/MIPSpro/7.3/stl_numeric_facets.h +13 -0
- data/share/gccxml-0.9/MIPSpro/7.3/stl_threads.h +11 -0
- data/share/gccxml-0.9/MIPSpro/7.3/string +18 -0
- data/share/gccxml-0.9/MIPSpro/find_flags +137 -0
- data/share/gccxml-0.9/Sun/5.8/adaptation.patch +168 -0
- data/share/gccxml-0.9/Sun/README +8 -0
- data/share/gccxml-0.9/Sun/adapt_headers.sh +24 -0
- data/share/gccxml-0.9/Sun/find_flags +69 -0
- data/share/gccxml-0.9/gccxml_config +2 -0
- data/share/gccxml-0.9/gccxml_find_flags +93 -0
- data/share/man/man1/gccxml.1 +246 -0
- metadata +204 -0
@@ -0,0 +1,17 @@
|
|
1
|
+
#ifndef GCCXML_STDEXCEPT
|
2
|
+
#define GCCXML_STDEXCEPT
|
3
|
+
|
4
|
+
#include_next <stdexcept>
|
5
|
+
|
6
|
+
// 19.1/3
|
7
|
+
using std::logic_error;
|
8
|
+
using std::domain_error;
|
9
|
+
using std::invalid_argument;
|
10
|
+
using std::length_error;
|
11
|
+
using std::out_of_range;
|
12
|
+
using std::runtime_error;
|
13
|
+
using std::range_error;
|
14
|
+
using std::overflow_error;
|
15
|
+
using std::underflow_error;
|
16
|
+
|
17
|
+
#endif
|
@@ -0,0 +1,1057 @@
|
|
1
|
+
/*
|
2
|
+
* Copyright (c) 1996-1997
|
3
|
+
* Silicon Graphics Computer Systems, Inc.
|
4
|
+
*
|
5
|
+
* Permission to use, copy, modify, distribute and sell this software
|
6
|
+
* and its documentation for any purpose is hereby granted without fee,
|
7
|
+
* provided that the above copyright notice appear in all copies and
|
8
|
+
* that both that copyright notice and this permission notice appear
|
9
|
+
* in supporting documentation. Silicon Graphics makes no
|
10
|
+
* representations about the suitability of this software for any
|
11
|
+
* purpose. It is provided "as is" without express or implied warranty.
|
12
|
+
*/
|
13
|
+
|
14
|
+
/* NOTE: This is an internal header file, included by other STL headers.
|
15
|
+
* You should not attempt to use it directly.
|
16
|
+
*/
|
17
|
+
|
18
|
+
#ifndef __SGI_STL_INTERNAL_ALLOC_H
|
19
|
+
#define __SGI_STL_INTERNAL_ALLOC_H
|
20
|
+
|
21
|
+
#ifdef __SUNPRO_CC
|
22
|
+
# define __PRIVATE public
|
23
|
+
// Extra access restrictions prevent us from really making some things
|
24
|
+
// private.
|
25
|
+
#else
|
26
|
+
# define __PRIVATE private
|
27
|
+
#endif
|
28
|
+
|
29
|
+
#ifdef __STL_STATIC_TEMPLATE_MEMBER_BUG
|
30
|
+
# define __USE_MALLOC
|
31
|
+
#endif
|
32
|
+
|
33
|
+
|
34
|
+
// This implements some standard node allocators. These are
|
35
|
+
// NOT the same as the allocators in the C++ draft standard or in
|
36
|
+
// in the original STL. They do not encapsulate different pointer
|
37
|
+
// types; indeed we assume that there is only one pointer type.
|
38
|
+
// The allocation primitives are intended to allocate individual objects,
|
39
|
+
// not larger arenas as with the original STL allocators.
|
40
|
+
|
41
|
+
#ifdef __STL_USE_EXCEPTIONS
|
42
|
+
# include <new>
|
43
|
+
# define __THROW_BAD_ALLOC throw bad_alloc()
|
44
|
+
#elif !defined(__THROW_BAD_ALLOC)
|
45
|
+
# include <iostream.h>
|
46
|
+
# define __THROW_BAD_ALLOC cerr << "out of memory" << endl; exit(1)
|
47
|
+
#endif
|
48
|
+
|
49
|
+
#ifdef __STL_WIN32THREADS
|
50
|
+
# include <windows.h>
|
51
|
+
#endif
|
52
|
+
|
53
|
+
#include <stddef.h>
|
54
|
+
#include <stdlib.h>
|
55
|
+
#include <string.h>
|
56
|
+
#include <assert.h>
|
57
|
+
#ifndef __RESTRICT
|
58
|
+
# define __RESTRICT
|
59
|
+
#endif
|
60
|
+
|
61
|
+
#if !defined(__STL_PTHREADS) && !defined(__STL_SOLTHREADS) \
|
62
|
+
&& !defined(_NOTHREADS) \
|
63
|
+
&& !defined(__STL_SGI_THREADS) && !defined(__STL_WIN32THREADS)
|
64
|
+
# define _NOTHREADS
|
65
|
+
#endif
|
66
|
+
|
67
|
+
# ifdef __STL_PTHREADS
|
68
|
+
// POSIX Threads
|
69
|
+
// This is dubious, since this is likely to be a high contention
|
70
|
+
// lock. Performance may not be adequate.
|
71
|
+
# include <pthread.h>
|
72
|
+
# define __NODE_ALLOCATOR_LOCK \
|
73
|
+
if (threads) pthread_mutex_lock(&_S_node_allocator_lock)
|
74
|
+
# define __NODE_ALLOCATOR_UNLOCK \
|
75
|
+
if (threads) pthread_mutex_unlock(&_S_node_allocator_lock)
|
76
|
+
# define __NODE_ALLOCATOR_THREADS true
|
77
|
+
# define __VOLATILE volatile // Needed at -O3 on SGI
|
78
|
+
# endif
|
79
|
+
# ifdef __STL_SOLTHREADS
|
80
|
+
# include <thread.h>
|
81
|
+
# define __NODE_ALLOCATOR_LOCK \
|
82
|
+
if (threads) mutex_lock(&_S_node_allocator_lock)
|
83
|
+
# define __NODE_ALLOCATOR_UNLOCK \
|
84
|
+
if (threads) mutex_unlock(&_S_node_allocator_lock)
|
85
|
+
# define __NODE_ALLOCATOR_THREADS true
|
86
|
+
# define __VOLATILE
|
87
|
+
# endif
|
88
|
+
# ifdef __STL_WIN32THREADS
|
89
|
+
// The lock needs to be initialized by constructing an allocator
|
90
|
+
// objects of the right type. We do that here explicitly for alloc.
|
91
|
+
# define __NODE_ALLOCATOR_LOCK \
|
92
|
+
EnterCriticalSection(&_S_node_allocator_lock)
|
93
|
+
# define __NODE_ALLOCATOR_UNLOCK \
|
94
|
+
LeaveCriticalSection(&_S_node_allocator_lock)
|
95
|
+
# define __NODE_ALLOCATOR_THREADS true
|
96
|
+
# define __VOLATILE volatile // may not be needed
|
97
|
+
# endif /* WIN32THREADS */
|
98
|
+
# ifdef __STL_SGI_THREADS
|
99
|
+
// This should work without threads, with sproc threads, or with
|
100
|
+
// pthreads. It is suboptimal in all cases.
|
101
|
+
// It is unlikely to even compile on nonSGI machines.
|
102
|
+
|
103
|
+
extern "C" {
|
104
|
+
extern int __us_rsthread_malloc;
|
105
|
+
}
|
106
|
+
// The above is copied from malloc.h. Including <malloc.h>
|
107
|
+
// would be cleaner but fails with certain levels of standard
|
108
|
+
// conformance.
|
109
|
+
# define __NODE_ALLOCATOR_LOCK if (threads && __us_rsthread_malloc) \
|
110
|
+
{ _S_lock(&_S_node_allocator_lock); }
|
111
|
+
# define __NODE_ALLOCATOR_UNLOCK if (threads && __us_rsthread_malloc) \
|
112
|
+
{ _S_unlock(&_S_node_allocator_lock); }
|
113
|
+
# define __NODE_ALLOCATOR_THREADS true
|
114
|
+
# define __VOLATILE volatile // Needed at -O3 on SGI
|
115
|
+
# endif
|
116
|
+
# ifdef _NOTHREADS
|
117
|
+
// Thread-unsafe
|
118
|
+
# define __NODE_ALLOCATOR_LOCK
|
119
|
+
# define __NODE_ALLOCATOR_UNLOCK
|
120
|
+
# define __NODE_ALLOCATOR_THREADS false
|
121
|
+
# define __VOLATILE
|
122
|
+
# endif
|
123
|
+
|
124
|
+
__STL_BEGIN_NAMESPACE
|
125
|
+
|
126
|
+
#if defined(__sgi) && !defined(__GNUC__) && (_MIPS_SIM != _MIPS_SIM_ABI32)
|
127
|
+
#pragma set woff 1174
|
128
|
+
#endif
|
129
|
+
|
130
|
+
// Malloc-based allocator. Typically slower than default alloc below.
|
131
|
+
// Typically thread-safe and more storage efficient.
|
132
|
+
#ifdef __STL_STATIC_TEMPLATE_MEMBER_BUG
|
133
|
+
# ifdef __DECLARE_GLOBALS_HERE
|
134
|
+
void (* __malloc_alloc_oom_handler)() = 0;
|
135
|
+
// g++ 2.7.2 does not handle static template data members.
|
136
|
+
# else
|
137
|
+
extern void (* __malloc_alloc_oom_handler)();
|
138
|
+
# endif
|
139
|
+
#endif
|
140
|
+
|
141
|
+
template <int __inst>
|
142
|
+
class __malloc_alloc_template {
|
143
|
+
|
144
|
+
private:
|
145
|
+
|
146
|
+
static void* _S_oom_malloc(size_t);
|
147
|
+
static void* _S_oom_realloc(void*, size_t);
|
148
|
+
|
149
|
+
#ifndef __STL_STATIC_TEMPLATE_MEMBER_BUG
|
150
|
+
static void (* __malloc_alloc_oom_handler)();
|
151
|
+
#endif
|
152
|
+
|
153
|
+
public:
|
154
|
+
|
155
|
+
static void* allocate(size_t __n)
|
156
|
+
{
|
157
|
+
void* __result = malloc(__n);
|
158
|
+
if (0 == __result) __result = _S_oom_malloc(__n);
|
159
|
+
return __result;
|
160
|
+
}
|
161
|
+
|
162
|
+
static void deallocate(void* __p, size_t /* __n */)
|
163
|
+
{
|
164
|
+
free(__p);
|
165
|
+
}
|
166
|
+
|
167
|
+
static void* reallocate(void* __p, size_t /* old_sz */, size_t __new_sz)
|
168
|
+
{
|
169
|
+
void* __result = realloc(__p, __new_sz);
|
170
|
+
if (0 == __result) __result = _S_oom_realloc(__p, __new_sz);
|
171
|
+
return __result;
|
172
|
+
}
|
173
|
+
|
174
|
+
static void (* __set_malloc_handler(void (*__f)()))()
|
175
|
+
{
|
176
|
+
void (* __old)() = __malloc_alloc_oom_handler;
|
177
|
+
__malloc_alloc_oom_handler = __f;
|
178
|
+
return(__old);
|
179
|
+
}
|
180
|
+
|
181
|
+
};
|
182
|
+
|
183
|
+
// malloc_alloc out-of-memory handling
|
184
|
+
|
185
|
+
#ifndef __STL_STATIC_TEMPLATE_MEMBER_BUG
|
186
|
+
template <int __inst>
|
187
|
+
void (* __malloc_alloc_template<__inst>::__malloc_alloc_oom_handler)() = 0;
|
188
|
+
#endif
|
189
|
+
|
190
|
+
template <int __inst>
|
191
|
+
void*
|
192
|
+
__malloc_alloc_template<__inst>::_S_oom_malloc(size_t __n)
|
193
|
+
{
|
194
|
+
void (* __my_malloc_handler)();
|
195
|
+
void* __result;
|
196
|
+
|
197
|
+
for (;;) {
|
198
|
+
__my_malloc_handler = __malloc_alloc_oom_handler;
|
199
|
+
if (0 == __my_malloc_handler) { __THROW_BAD_ALLOC; }
|
200
|
+
(*__my_malloc_handler)();
|
201
|
+
__result = malloc(__n);
|
202
|
+
if (__result) return(__result);
|
203
|
+
}
|
204
|
+
}
|
205
|
+
|
206
|
+
template <int __inst>
|
207
|
+
void* __malloc_alloc_template<__inst>::_S_oom_realloc(void* __p, size_t __n)
|
208
|
+
{
|
209
|
+
void (* __my_malloc_handler)();
|
210
|
+
void* __result;
|
211
|
+
|
212
|
+
for (;;) {
|
213
|
+
__my_malloc_handler = __malloc_alloc_oom_handler;
|
214
|
+
if (0 == __my_malloc_handler) { __THROW_BAD_ALLOC; }
|
215
|
+
(*__my_malloc_handler)();
|
216
|
+
__result = realloc(__p, __n);
|
217
|
+
if (__result) return(__result);
|
218
|
+
}
|
219
|
+
}
|
220
|
+
|
221
|
+
typedef __malloc_alloc_template<0> malloc_alloc;
|
222
|
+
|
223
|
+
template<class _Tp, class _Alloc>
|
224
|
+
class simple_alloc {
|
225
|
+
|
226
|
+
public:
|
227
|
+
static _Tp* allocate(size_t __n)
|
228
|
+
{ return 0 == __n ? 0 : (_Tp*) _Alloc::allocate(__n * sizeof (_Tp)); }
|
229
|
+
static _Tp* allocate(void)
|
230
|
+
{ return (_Tp*) _Alloc::allocate(sizeof (_Tp)); }
|
231
|
+
static void deallocate(_Tp* __p, size_t __n)
|
232
|
+
{ if (0 != __n) _Alloc::deallocate(__p, __n * sizeof (_Tp)); }
|
233
|
+
static void deallocate(_Tp* __p)
|
234
|
+
{ _Alloc::deallocate(__p, sizeof (_Tp)); }
|
235
|
+
};
|
236
|
+
|
237
|
+
// Allocator adaptor to check size arguments for debugging.
|
238
|
+
// Reports errors using assert. Checking can be disabled with
|
239
|
+
// NDEBUG, but it's far better to just use the underlying allocator
|
240
|
+
// instead when no checking is desired.
|
241
|
+
// There is some evidence that this can confuse Purify.
|
242
|
+
template <class _Alloc>
|
243
|
+
class debug_alloc {
|
244
|
+
|
245
|
+
private:
|
246
|
+
|
247
|
+
enum {_S_extra = 8}; // Size of space used to store size. Note
|
248
|
+
// that this must be large enough to preserve
|
249
|
+
// alignment.
|
250
|
+
|
251
|
+
public:
|
252
|
+
|
253
|
+
static void* allocate(size_t __n)
|
254
|
+
{
|
255
|
+
char* __result = (char*)_Alloc::allocate(__n + _S_extra);
|
256
|
+
*(size_t*)__result = __n;
|
257
|
+
return __result + _S_extra;
|
258
|
+
}
|
259
|
+
|
260
|
+
static void deallocate(void* __p, size_t __n)
|
261
|
+
{
|
262
|
+
char* __real_p = (char*)__p - _S_extra;
|
263
|
+
assert(*(size_t*)__real_p == __n);
|
264
|
+
_Alloc::deallocate(__real_p, __n + _S_extra);
|
265
|
+
}
|
266
|
+
|
267
|
+
static void* reallocate(void* __p, size_t __old_sz, size_t __new_sz)
|
268
|
+
{
|
269
|
+
char* __real_p = (char*)__p - _S_extra;
|
270
|
+
assert(*(size_t*)__real_p == __old_sz);
|
271
|
+
char* __result = (char*)
|
272
|
+
_Alloc::reallocate(__real_p, __old_sz + _S_extra, __new_sz + _S_extra);
|
273
|
+
*(size_t*)__result = __new_sz;
|
274
|
+
return __result + _S_extra;
|
275
|
+
}
|
276
|
+
|
277
|
+
};
|
278
|
+
|
279
|
+
|
280
|
+
# ifdef __USE_MALLOC
|
281
|
+
|
282
|
+
typedef malloc_alloc alloc;
|
283
|
+
typedef malloc_alloc single_client_alloc;
|
284
|
+
|
285
|
+
# else
|
286
|
+
|
287
|
+
|
288
|
+
// Default node allocator.
|
289
|
+
// With a reasonable compiler, this should be roughly as fast as the
|
290
|
+
// original STL class-specific allocators, but with less fragmentation.
|
291
|
+
// Default_alloc_template parameters are experimental and MAY
|
292
|
+
// DISAPPEAR in the future. Clients should just use alloc for now.
|
293
|
+
//
|
294
|
+
// Important implementation properties:
|
295
|
+
// 1. If the client request an object of size > _MAX_BYTES, the resulting
|
296
|
+
// object will be obtained directly from malloc.
|
297
|
+
// 2. In all other cases, we allocate an object of size exactly
|
298
|
+
// _S_round_up(requested_size). Thus the client has enough size
|
299
|
+
// information that we can return the object to the proper free list
|
300
|
+
// without permanently losing part of the object.
|
301
|
+
//
|
302
|
+
|
303
|
+
// The first template parameter specifies whether more than one thread
|
304
|
+
// may use this allocator. It is safe to allocate an object from
|
305
|
+
// one instance of a default_alloc and deallocate it with another
|
306
|
+
// one. This effectively transfers its ownership to the second one.
|
307
|
+
// This may have undesirable effects on reference locality.
|
308
|
+
// The second parameter is unreferenced and serves only to allow the
|
309
|
+
// creation of multiple default_alloc instances.
|
310
|
+
// Node that containers built on different allocator instances have
|
311
|
+
// different types, limiting the utility of this approach.
|
312
|
+
#ifdef __SUNPRO_CC
|
313
|
+
// breaks if we make these template class members:
|
314
|
+
enum {_ALIGN = 8};
|
315
|
+
enum {_MAX_BYTES = 128};
|
316
|
+
enum {_NFREELISTS = _MAX_BYTES/_ALIGN};
|
317
|
+
#endif
|
318
|
+
|
319
|
+
template <bool threads, int inst>
|
320
|
+
class __default_alloc_template {
|
321
|
+
|
322
|
+
private:
|
323
|
+
// Really we should use static const int x = N
|
324
|
+
// instead of enum { x = N }, but few compilers accept the former.
|
325
|
+
# ifndef __SUNPRO_CC
|
326
|
+
enum {_ALIGN = 8};
|
327
|
+
enum {_MAX_BYTES = 128};
|
328
|
+
enum {_NFREELISTS = _MAX_BYTES/_ALIGN};
|
329
|
+
# endif
|
330
|
+
static size_t
|
331
|
+
_S_round_up(size_t __bytes)
|
332
|
+
{ return (((__bytes) + _ALIGN-1) & ~(_ALIGN - 1)); }
|
333
|
+
|
334
|
+
__PRIVATE:
|
335
|
+
union _Obj {
|
336
|
+
union _Obj* _M_free_list_link;
|
337
|
+
char _M_client_data[1]; /* The client sees this. */
|
338
|
+
};
|
339
|
+
private:
|
340
|
+
# ifdef __SUNPRO_CC
|
341
|
+
static _Obj* __VOLATILE _S_free_list[];
|
342
|
+
// Specifying a size results in duplicate def for 4.1
|
343
|
+
# else
|
344
|
+
static _Obj* __VOLATILE _S_free_list[_NFREELISTS];
|
345
|
+
# endif
|
346
|
+
static size_t _S_freelist_index(size_t __bytes) {
|
347
|
+
return (((__bytes) + _ALIGN-1)/_ALIGN - 1);
|
348
|
+
}
|
349
|
+
|
350
|
+
// Returns an object of size __n, and optionally adds to size __n free list.
|
351
|
+
static void* _S_refill(size_t __n);
|
352
|
+
// Allocates a chunk for nobjs of size "size". nobjs may be reduced
|
353
|
+
// if it is inconvenient to allocate the requested number.
|
354
|
+
static char* _S_chunk_alloc(size_t __size, int& __nobjs);
|
355
|
+
|
356
|
+
// Chunk allocation state.
|
357
|
+
static char* _S_start_free;
|
358
|
+
static char* _S_end_free;
|
359
|
+
static size_t _S_heap_size;
|
360
|
+
|
361
|
+
# ifdef __STL_SGI_THREADS
|
362
|
+
static volatile unsigned long _S_node_allocator_lock;
|
363
|
+
static void _S_lock(volatile unsigned long*);
|
364
|
+
static inline void _S_unlock(volatile unsigned long*);
|
365
|
+
# endif
|
366
|
+
|
367
|
+
# ifdef __STL_PTHREADS
|
368
|
+
static pthread_mutex_t _S_node_allocator_lock;
|
369
|
+
# endif
|
370
|
+
|
371
|
+
# ifdef __STL_SOLTHREADS
|
372
|
+
static mutex_t _S_node_allocator_lock;
|
373
|
+
# endif
|
374
|
+
|
375
|
+
# ifdef __STL_WIN32THREADS
|
376
|
+
static CRITICAL_SECTION _S_node_allocator_lock;
|
377
|
+
static bool _S_node_allocator_lock_initialized;
|
378
|
+
|
379
|
+
public:
|
380
|
+
__default_alloc_template() {
|
381
|
+
// This assumes the first constructor is called before threads
|
382
|
+
// are started.
|
383
|
+
if (!_S_node_allocator_lock_initialized) {
|
384
|
+
InitializeCriticalSection(&_S_node_allocator_lock);
|
385
|
+
_S_node_allocator_lock_initialized = true;
|
386
|
+
}
|
387
|
+
}
|
388
|
+
private:
|
389
|
+
# endif
|
390
|
+
|
391
|
+
class _Lock {
|
392
|
+
public:
|
393
|
+
_Lock() { __NODE_ALLOCATOR_LOCK; }
|
394
|
+
~_Lock() { __NODE_ALLOCATOR_UNLOCK; }
|
395
|
+
};
|
396
|
+
friend class _Lock;
|
397
|
+
|
398
|
+
public:
|
399
|
+
|
400
|
+
/* __n must be > 0 */
|
401
|
+
static void* allocate(size_t __n)
|
402
|
+
{
|
403
|
+
_Obj* __VOLATILE* __my_free_list;
|
404
|
+
_Obj* __RESTRICT __result;
|
405
|
+
|
406
|
+
if (__n > (size_t) _MAX_BYTES) {
|
407
|
+
return(malloc_alloc::allocate(__n));
|
408
|
+
}
|
409
|
+
__my_free_list = _S_free_list + _S_freelist_index(__n);
|
410
|
+
// Acquire the lock here with a constructor call.
|
411
|
+
// This ensures that it is released in exit or during stack
|
412
|
+
// unwinding.
|
413
|
+
# ifndef _NOTHREADS
|
414
|
+
/*REFERENCED*/
|
415
|
+
_Lock __lock_instance;
|
416
|
+
# endif
|
417
|
+
__result = *__my_free_list;
|
418
|
+
if (__result == 0) {
|
419
|
+
void* __r = _S_refill(_S_round_up(__n));
|
420
|
+
return __r;
|
421
|
+
}
|
422
|
+
*__my_free_list = __result -> _M_free_list_link;
|
423
|
+
return (__result);
|
424
|
+
};
|
425
|
+
|
426
|
+
/* __p may not be 0 */
|
427
|
+
static void deallocate(void* __p, size_t __n)
|
428
|
+
{
|
429
|
+
_Obj* __q = (_Obj*)__p;
|
430
|
+
_Obj* __VOLATILE* __my_free_list;
|
431
|
+
|
432
|
+
if (__n > (size_t) _MAX_BYTES) {
|
433
|
+
malloc_alloc::deallocate(__p, __n);
|
434
|
+
return;
|
435
|
+
}
|
436
|
+
__my_free_list = _S_free_list + _S_freelist_index(__n);
|
437
|
+
// acquire lock
|
438
|
+
# ifndef _NOTHREADS
|
439
|
+
/*REFERENCED*/
|
440
|
+
_Lock __lock_instance;
|
441
|
+
# endif /* _NOTHREADS */
|
442
|
+
__q -> _M_free_list_link = *__my_free_list;
|
443
|
+
*__my_free_list = __q;
|
444
|
+
// lock is released here
|
445
|
+
}
|
446
|
+
|
447
|
+
static void* reallocate(void* __p, size_t __old_sz, size_t __new_sz);
|
448
|
+
|
449
|
+
} ;
|
450
|
+
|
451
|
+
typedef __default_alloc_template<__NODE_ALLOCATOR_THREADS, 0> alloc;
|
452
|
+
typedef __default_alloc_template<false, 0> single_client_alloc;
|
453
|
+
|
454
|
+
|
455
|
+
|
456
|
+
/* We allocate memory in large chunks in order to avoid fragmenting */
|
457
|
+
/* the malloc heap too much. */
|
458
|
+
/* We assume that size is properly aligned. */
|
459
|
+
/* We hold the allocation lock. */
|
460
|
+
template <bool __threads, int __inst>
|
461
|
+
char*
|
462
|
+
__default_alloc_template<__threads, __inst>::_S_chunk_alloc(size_t __size,
|
463
|
+
int& __nobjs)
|
464
|
+
{
|
465
|
+
char* __result;
|
466
|
+
size_t __total_bytes = __size * __nobjs;
|
467
|
+
size_t __bytes_left = _S_end_free - _S_start_free;
|
468
|
+
|
469
|
+
if (__bytes_left >= __total_bytes) {
|
470
|
+
__result = _S_start_free;
|
471
|
+
_S_start_free += __total_bytes;
|
472
|
+
return(__result);
|
473
|
+
} else if (__bytes_left >= __size) {
|
474
|
+
__nobjs = (int)(__bytes_left/__size);
|
475
|
+
__total_bytes = __size * __nobjs;
|
476
|
+
__result = _S_start_free;
|
477
|
+
_S_start_free += __total_bytes;
|
478
|
+
return(__result);
|
479
|
+
} else {
|
480
|
+
size_t __bytes_to_get =
|
481
|
+
2 * __total_bytes + _S_round_up(_S_heap_size >> 4);
|
482
|
+
// Try to make use of the left-over piece.
|
483
|
+
if (__bytes_left > 0) {
|
484
|
+
_Obj* __VOLATILE* __my_free_list =
|
485
|
+
_S_free_list + _S_freelist_index(__bytes_left);
|
486
|
+
|
487
|
+
((_Obj*)_S_start_free) -> _M_free_list_link = *__my_free_list;
|
488
|
+
*__my_free_list = (_Obj*)_S_start_free;
|
489
|
+
}
|
490
|
+
_S_start_free = (char*)malloc(__bytes_to_get);
|
491
|
+
if (0 == _S_start_free) {
|
492
|
+
size_t __i;
|
493
|
+
_Obj* __VOLATILE* __my_free_list;
|
494
|
+
_Obj* __p;
|
495
|
+
// Try to make do with what we have. That can't
|
496
|
+
// hurt. We do not try smaller requests, since that tends
|
497
|
+
// to result in disaster on multi-process machines.
|
498
|
+
for (__i = __size; __i <= _MAX_BYTES; __i += _ALIGN) {
|
499
|
+
__my_free_list = _S_free_list + _S_freelist_index(__i);
|
500
|
+
__p = *__my_free_list;
|
501
|
+
if (0 != __p) {
|
502
|
+
*__my_free_list = __p -> _M_free_list_link;
|
503
|
+
_S_start_free = (char*)__p;
|
504
|
+
_S_end_free = _S_start_free + __i;
|
505
|
+
return(_S_chunk_alloc(__size, __nobjs));
|
506
|
+
// Any leftover piece will eventually make it to the
|
507
|
+
// right free list.
|
508
|
+
}
|
509
|
+
}
|
510
|
+
_S_end_free = 0; // In case of exception.
|
511
|
+
_S_start_free = (char*)malloc_alloc::allocate(__bytes_to_get);
|
512
|
+
// This should either throw an
|
513
|
+
// exception or remedy the situation. Thus we assume it
|
514
|
+
// succeeded.
|
515
|
+
}
|
516
|
+
_S_heap_size += __bytes_to_get;
|
517
|
+
_S_end_free = _S_start_free + __bytes_to_get;
|
518
|
+
return(_S_chunk_alloc(__size, __nobjs));
|
519
|
+
}
|
520
|
+
}
|
521
|
+
|
522
|
+
|
523
|
+
/* Returns an object of size __n, and optionally adds to size __n free list.*/
|
524
|
+
/* We assume that __n is properly aligned. */
|
525
|
+
/* We hold the allocation lock. */
|
526
|
+
template <bool __threads, int __inst>
|
527
|
+
void*
|
528
|
+
__default_alloc_template<__threads, __inst>::_S_refill(size_t __n)
|
529
|
+
{
|
530
|
+
int __nobjs = 20;
|
531
|
+
char* __chunk = _S_chunk_alloc(__n, __nobjs);
|
532
|
+
_Obj* __VOLATILE* __my_free_list;
|
533
|
+
_Obj* __result;
|
534
|
+
_Obj* __current_obj;
|
535
|
+
_Obj* __next_obj;
|
536
|
+
int __i;
|
537
|
+
|
538
|
+
if (1 == __nobjs) return(__chunk);
|
539
|
+
__my_free_list = _S_free_list + _S_freelist_index(__n);
|
540
|
+
|
541
|
+
/* Build free list in chunk */
|
542
|
+
__result = (_Obj*)__chunk;
|
543
|
+
*__my_free_list = __next_obj = (_Obj*)(__chunk + __n);
|
544
|
+
for (__i = 1; ; __i++) {
|
545
|
+
__current_obj = __next_obj;
|
546
|
+
__next_obj = (_Obj*)((char*)__next_obj + __n);
|
547
|
+
if (__nobjs - 1 == __i) {
|
548
|
+
__current_obj -> _M_free_list_link = 0;
|
549
|
+
break;
|
550
|
+
} else {
|
551
|
+
__current_obj -> _M_free_list_link = __next_obj;
|
552
|
+
}
|
553
|
+
}
|
554
|
+
return(__result);
|
555
|
+
}
|
556
|
+
|
557
|
+
template <bool threads, int inst>
|
558
|
+
void*
|
559
|
+
__default_alloc_template<threads, inst>::reallocate(void* __p,
|
560
|
+
size_t __old_sz,
|
561
|
+
size_t __new_sz)
|
562
|
+
{
|
563
|
+
void* __result;
|
564
|
+
size_t __copy_sz;
|
565
|
+
|
566
|
+
if (__old_sz > (size_t) _MAX_BYTES && __new_sz > (size_t) _MAX_BYTES) {
|
567
|
+
return(realloc(__p, __new_sz));
|
568
|
+
}
|
569
|
+
if (_S_round_up(__old_sz) == _S_round_up(__new_sz)) return(__p);
|
570
|
+
__result = allocate(__new_sz);
|
571
|
+
__copy_sz = __new_sz > __old_sz? __old_sz : __new_sz;
|
572
|
+
memcpy(__result, __p, __copy_sz);
|
573
|
+
deallocate(__p, __old_sz);
|
574
|
+
return(__result);
|
575
|
+
}
|
576
|
+
|
577
|
+
#ifdef __STL_PTHREADS
|
578
|
+
template <bool __threads, int __inst>
|
579
|
+
pthread_mutex_t
|
580
|
+
__default_alloc_template<__threads, __inst>::_S_node_allocator_lock
|
581
|
+
= PTHREAD_MUTEX_INITIALIZER;
|
582
|
+
#endif
|
583
|
+
|
584
|
+
#ifdef __STL_SOLTHREADS
|
585
|
+
template <bool __threads, int __inst>
|
586
|
+
mutex_t
|
587
|
+
__default_alloc_template<__threads, __inst>::_S_node_allocator_lock
|
588
|
+
= DEFAULTMUTEX;
|
589
|
+
#endif
|
590
|
+
|
591
|
+
#ifdef __STL_WIN32THREADS
|
592
|
+
template <bool __threads, int __inst>
|
593
|
+
CRITICAL_SECTION
|
594
|
+
__default_alloc_template<__threads, __inst>::
|
595
|
+
_S_node_allocator_lock;
|
596
|
+
|
597
|
+
template <bool __threads, int __inst>
|
598
|
+
bool
|
599
|
+
__default_alloc_template<__threads, __inst>::
|
600
|
+
_S_node_allocator_lock_initialized
|
601
|
+
= false;
|
602
|
+
#endif
|
603
|
+
|
604
|
+
#ifdef __STL_SGI_THREADS
|
605
|
+
__STL_END_NAMESPACE
|
606
|
+
#include <mutex.h>
|
607
|
+
#include <time.h> /* XXX should use <ctime> */
|
608
|
+
__STL_BEGIN_NAMESPACE
|
609
|
+
// Somewhat generic lock implementations. We need only test-and-set
|
610
|
+
// and some way to sleep. These should work with both SGI pthreads
|
611
|
+
// and sproc threads. They may be useful on other systems.
|
612
|
+
template <bool __threads, int __inst>
|
613
|
+
volatile unsigned long
|
614
|
+
__default_alloc_template<__threads, __inst>::_S_node_allocator_lock = 0;
|
615
|
+
|
616
|
+
#if __mips < 3 || !(defined (_ABIN32) || defined(_ABI64)) || defined(__GNUC__)
|
617
|
+
# define __test_and_set(l,v) test_and_set(l,v)
|
618
|
+
#endif
|
619
|
+
|
620
|
+
template <bool __threads, int __inst>
|
621
|
+
void
|
622
|
+
__default_alloc_template<__threads, __inst>::
|
623
|
+
_S_lock(volatile unsigned long* __lock)
|
624
|
+
{
|
625
|
+
const unsigned __low_spin_max = 30; // spins if we suspect uniprocessor
|
626
|
+
const unsigned __high_spin_max = 1000; // spins for multiprocessor
|
627
|
+
static unsigned __spin_max = __low_spin_max;
|
628
|
+
unsigned __my_spin_max;
|
629
|
+
static unsigned __last_spins = 0;
|
630
|
+
unsigned __my_last_spins;
|
631
|
+
unsigned __junk;
|
632
|
+
# define __ALLOC_PAUSE \
|
633
|
+
__junk *= __junk; __junk *= __junk; __junk *= __junk; __junk *= __junk
|
634
|
+
int __i;
|
635
|
+
|
636
|
+
if (!__test_and_set((unsigned long*)__lock, 1)) {
|
637
|
+
return;
|
638
|
+
}
|
639
|
+
__my_spin_max = __spin_max;
|
640
|
+
__my_last_spins = __last_spins;
|
641
|
+
for (__i = 0; __i < __my_spin_max; __i++) {
|
642
|
+
if (__i < __my_last_spins/2 || *__lock) {
|
643
|
+
__ALLOC_PAUSE;
|
644
|
+
continue;
|
645
|
+
}
|
646
|
+
if (!__test_and_set((unsigned long*)__lock, 1)) {
|
647
|
+
// got it!
|
648
|
+
// Spinning worked. Thus we're probably not being scheduled
|
649
|
+
// against the other process with which we were contending.
|
650
|
+
// Thus it makes sense to spin longer the next time.
|
651
|
+
__last_spins = __i;
|
652
|
+
__spin_max = __high_spin_max;
|
653
|
+
return;
|
654
|
+
}
|
655
|
+
}
|
656
|
+
// We are probably being scheduled against the other process. Sleep.
|
657
|
+
__spin_max = __low_spin_max;
|
658
|
+
for (__i = 0 ;; ++__i) {
|
659
|
+
struct timespec __ts;
|
660
|
+
int __log_nsec = __i + 6;
|
661
|
+
|
662
|
+
if (!__test_and_set((unsigned long *)__lock, 1)) {
|
663
|
+
return;
|
664
|
+
}
|
665
|
+
if (__log_nsec > 27) __log_nsec = 27;
|
666
|
+
/* Max sleep is 2**27nsec ~ 60msec */
|
667
|
+
__ts.tv_sec = 0;
|
668
|
+
__ts.tv_nsec = 1 << __log_nsec;
|
669
|
+
nanosleep(&__ts, 0);
|
670
|
+
}
|
671
|
+
}
|
672
|
+
|
673
|
+
template <bool __threads, int __inst>
|
674
|
+
inline void
|
675
|
+
__default_alloc_template<__threads, __inst>::_S_unlock(
|
676
|
+
volatile unsigned long* __lock)
|
677
|
+
{
|
678
|
+
# if defined(__GNUC__) && __mips >= 3
|
679
|
+
asm("sync");
|
680
|
+
*__lock = 0;
|
681
|
+
# elif __mips >= 3 && (defined (_ABIN32) || defined(_ABI64))
|
682
|
+
__lock_release(__lock);
|
683
|
+
# else
|
684
|
+
*__lock = 0;
|
685
|
+
// This is not sufficient on many multiprocessors, since
|
686
|
+
// writes to protected variables and the lock may be reordered.
|
687
|
+
# endif
|
688
|
+
}
|
689
|
+
#endif
|
690
|
+
|
691
|
+
template <bool __threads, int __inst>
|
692
|
+
char* __default_alloc_template<__threads, __inst>::_S_start_free = 0;
|
693
|
+
|
694
|
+
template <bool __threads, int __inst>
|
695
|
+
char* __default_alloc_template<__threads, __inst>::_S_end_free = 0;
|
696
|
+
|
697
|
+
template <bool __threads, int __inst>
|
698
|
+
size_t __default_alloc_template<__threads, __inst>::_S_heap_size = 0;
|
699
|
+
|
700
|
+
template <bool __threads, int __inst>
|
701
|
+
typename __default_alloc_template<__threads, __inst>::_Obj* __VOLATILE
|
702
|
+
__default_alloc_template<__threads, __inst> ::_S_free_list[
|
703
|
+
_NFREELISTS
|
704
|
+
] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, };
|
705
|
+
// The 16 zeros are necessary to make version 4.1 of the SunPro
|
706
|
+
// compiler happy. Otherwise it appears to allocate too little
|
707
|
+
// space for the array.
|
708
|
+
|
709
|
+
# ifdef __STL_WIN32THREADS
|
710
|
+
// Create one to get critical section initialized.
|
711
|
+
// We do this onece per file, but only the first constructor
|
712
|
+
// does anything.
|
713
|
+
static alloc __node_allocator_dummy_instance;
|
714
|
+
# endif
|
715
|
+
|
716
|
+
#endif /* ! __USE_MALLOC */
|
717
|
+
|
718
|
+
// This implements allocators as specified in the C++ standard.
|
719
|
+
//
|
720
|
+
// Note that standard-conforming allocators use many language features
|
721
|
+
// that are not yet widely implemented. In particular, they rely on
|
722
|
+
// member templates, partial specialization, partial ordering of function
|
723
|
+
// templates, the typename keyword, and the use of the template keyword
|
724
|
+
// to refer to a template member of a dependent type.
|
725
|
+
|
726
|
+
#ifdef __STL_USE_STD_ALLOCATORS
|
727
|
+
|
728
|
+
template <class _Tp>
|
729
|
+
class allocator {
|
730
|
+
typedef alloc _Alloc; // The underlying allocator.
|
731
|
+
public:
|
732
|
+
typedef size_t size_type;
|
733
|
+
typedef ptrdiff_t difference_type;
|
734
|
+
typedef _Tp* pointer;
|
735
|
+
typedef const _Tp* const_pointer;
|
736
|
+
typedef _Tp& reference;
|
737
|
+
typedef const _Tp& const_reference;
|
738
|
+
typedef _Tp value_type;
|
739
|
+
|
740
|
+
template <class _Tp1> struct rebind {
|
741
|
+
typedef allocator<_Tp1> other;
|
742
|
+
};
|
743
|
+
|
744
|
+
allocator() __STL_NOTHROW {}
|
745
|
+
allocator(const allocator&) __STL_NOTHROW {}
|
746
|
+
template <class _Tp1> allocator(const allocator<_Tp1>&) __STL_NOTHROW {}
|
747
|
+
~allocator() __STL_NOTHROW {}
|
748
|
+
|
749
|
+
pointer address(reference __x) const { return &__x; }
|
750
|
+
const_pointer address(const_reference __x) const { return &__x; }
|
751
|
+
|
752
|
+
// __n is permitted to be 0. The C++ standard says nothing about what
|
753
|
+
// the return value is when __n == 0.
|
754
|
+
_Tp* allocate(size_type __n, const void* = 0) {
|
755
|
+
return __n != 0 ? static_cast<_Tp*>(_Alloc::allocate(__n * sizeof(_Tp)))
|
756
|
+
: 0;
|
757
|
+
}
|
758
|
+
|
759
|
+
// __p is not permitted to be a null pointer.
|
760
|
+
void deallocate(pointer __p, size_type __n)
|
761
|
+
{ _Alloc::deallocate(__p, __n * sizeof(_Tp)); }
|
762
|
+
|
763
|
+
size_type max_size() const __STL_NOTHROW
|
764
|
+
{ return size_t(-1) / sizeof(_Tp); }
|
765
|
+
|
766
|
+
void construct(pointer __p, const _Tp& __val) { new(__p) _Tp(__val); }
|
767
|
+
void destroy(pointer __p) { __p->~_Tp(); }
|
768
|
+
};
|
769
|
+
|
770
|
+
template<>
|
771
|
+
class allocator<void> {
|
772
|
+
typedef size_t size_type;
|
773
|
+
typedef ptrdiff_t difference_type;
|
774
|
+
typedef void* pointer;
|
775
|
+
typedef const void* const_pointer;
|
776
|
+
typedef void value_type;
|
777
|
+
|
778
|
+
template <class _Tp1> struct rebind {
|
779
|
+
typedef allocator<_Tp1> other;
|
780
|
+
};
|
781
|
+
};
|
782
|
+
|
783
|
+
|
784
|
+
template <class _T1, class _T2>
|
785
|
+
inline bool operator==(const allocator<_T1>&, const allocator<_T2>&)
|
786
|
+
{
|
787
|
+
return true;
|
788
|
+
}
|
789
|
+
|
790
|
+
template <class _T1, class _T2>
|
791
|
+
inline bool operator!=(const allocator<_T1>&, const allocator<_T2>&)
|
792
|
+
{
|
793
|
+
return false;
|
794
|
+
}
|
795
|
+
|
796
|
+
// Allocator adaptor to turn an SGI-style allocator (e.g. alloc, malloc_alloc)
|
797
|
+
// into a standard-conforming allocator. Note that this adaptor does
|
798
|
+
// *not* assume that all objects of the underlying alloc class are
|
799
|
+
// identical, nor does it assume that all of the underlying alloc's
|
800
|
+
// member functions are static member functions. Note, also, that
|
801
|
+
// __allocator<_Tp, alloc> is essentially the same thing as allocator<_Tp>.
|
802
|
+
|
803
|
+
template <class _Tp, class _Alloc>
|
804
|
+
struct __allocator {
|
805
|
+
_Alloc __underlying_alloc;
|
806
|
+
|
807
|
+
typedef size_t size_type;
|
808
|
+
typedef ptrdiff_t difference_type;
|
809
|
+
typedef _Tp* pointer;
|
810
|
+
typedef const _Tp* const_pointer;
|
811
|
+
typedef _Tp& reference;
|
812
|
+
typedef const _Tp& const_reference;
|
813
|
+
typedef _Tp value_type;
|
814
|
+
|
815
|
+
template <class _Tp1> struct rebind {
|
816
|
+
typedef __allocator<_Tp1, _Alloc> other;
|
817
|
+
};
|
818
|
+
|
819
|
+
__allocator() __STL_NOTHROW {}
|
820
|
+
__allocator(const __allocator& __a) __STL_NOTHROW
|
821
|
+
: __underlying_alloc(__a.__underlying_alloc) {}
|
822
|
+
template <class _Tp1>
|
823
|
+
__allocator(const __allocator<_Tp1, _Alloc>& __a) __STL_NOTHROW
|
824
|
+
: __underlying_alloc(__a.__underlying_alloc) {}
|
825
|
+
~__allocator() __STL_NOTHROW {}
|
826
|
+
|
827
|
+
pointer address(reference __x) const { return &__x; }
|
828
|
+
const_pointer address(const_reference __x) const { return &__x; }
|
829
|
+
|
830
|
+
// __n is permitted to be 0.
|
831
|
+
_Tp* allocate(size_type __n, const void* = 0) {
|
832
|
+
return __n != 0
|
833
|
+
? static_cast<_Tp*>(__underlying_alloc.allocate(__n * sizeof(_Tp)))
|
834
|
+
: 0;
|
835
|
+
}
|
836
|
+
|
837
|
+
// __p is not permitted to be a null pointer.
|
838
|
+
void deallocate(pointer __p, size_type __n)
|
839
|
+
{ __underlying_alloc.deallocate(__p, __n * sizeof(_Tp)); }
|
840
|
+
|
841
|
+
size_type max_size() const __STL_NOTHROW
|
842
|
+
{ return size_t(-1) / sizeof(_Tp); }
|
843
|
+
|
844
|
+
void construct(pointer __p, const _Tp& __val) { new(__p) _Tp(__val); }
|
845
|
+
void destroy(pointer __p) { __p->~_Tp(); }
|
846
|
+
};
|
847
|
+
|
848
|
+
template <class _Alloc>
|
849
|
+
class __allocator<void, _Alloc> {
|
850
|
+
typedef size_t size_type;
|
851
|
+
typedef ptrdiff_t difference_type;
|
852
|
+
typedef void* pointer;
|
853
|
+
typedef const void* const_pointer;
|
854
|
+
typedef void value_type;
|
855
|
+
|
856
|
+
template <class _Tp1> struct rebind {
|
857
|
+
typedef __allocator<_Tp1, _Alloc> other;
|
858
|
+
};
|
859
|
+
};
|
860
|
+
|
861
|
+
template <class _Tp, class _Alloc>
|
862
|
+
inline bool operator==(const __allocator<_Tp, _Alloc>& __a1,
|
863
|
+
const __allocator<_Tp, _Alloc>& __a2)
|
864
|
+
{
|
865
|
+
return __a1.__underlying_alloc == __a2.__underlying_alloc;
|
866
|
+
}
|
867
|
+
|
868
|
+
#ifdef __STL_FUNCTION_TMPL_PARTIAL_ORDER
|
869
|
+
template <class _Tp, class _Alloc>
|
870
|
+
inline bool operator!=(const __allocator<_Tp, _Alloc>& __a1,
|
871
|
+
const __allocator<_Tp, _Alloc>& __a2)
|
872
|
+
{
|
873
|
+
return __a1.__underlying_alloc != __a2.__underlying_alloc;
|
874
|
+
}
|
875
|
+
#endif /* __STL_FUNCTION_TMPL_PARTIAL_ORDER */
|
876
|
+
|
877
|
+
// Comparison operators for all of the predifined SGI-style allocators.
|
878
|
+
// This ensures that __allocator<malloc_alloc> (for example) will
|
879
|
+
// work correctly.
|
880
|
+
|
881
|
+
template <int inst>
|
882
|
+
inline bool operator==(const __malloc_alloc_template<inst>&,
|
883
|
+
const __malloc_alloc_template<inst>&)
|
884
|
+
{
|
885
|
+
return true;
|
886
|
+
}
|
887
|
+
|
888
|
+
#ifdef __STL_FUNCTION_TMPL_PARTIAL_ORDER
|
889
|
+
template <int __inst>
|
890
|
+
inline bool operator!=(const __malloc_alloc_template<__inst>&,
|
891
|
+
const __malloc_alloc_template<__inst>&)
|
892
|
+
{
|
893
|
+
return false;
|
894
|
+
}
|
895
|
+
#endif /* __STL_FUNCTION_TMPL_PARTIAL_ORDER */
|
896
|
+
|
897
|
+
#ifndef __USE_MALLOC
|
898
|
+
template <bool __threads, int __inst>
|
899
|
+
inline bool operator==(const __default_alloc_template<__threads, __inst>&,
|
900
|
+
const __default_alloc_template<__threads, __inst>&)
|
901
|
+
{
|
902
|
+
return true;
|
903
|
+
}
|
904
|
+
|
905
|
+
# ifdef __STL_FUNCTION_TMPL_PARTIAL_ORDER
|
906
|
+
template <bool __threads, int __inst>
|
907
|
+
inline bool operator!=(const __default_alloc_template<__threads, __inst>&,
|
908
|
+
const __default_alloc_template<__threads, __inst>&)
|
909
|
+
{
|
910
|
+
return false;
|
911
|
+
}
|
912
|
+
# endif /* __STL_FUNCTION_TMPL_PARTIAL_ORDER */
|
913
|
+
#endif
|
914
|
+
|
915
|
+
template <class _Alloc>
|
916
|
+
inline bool operator==(const debug_alloc<_Alloc>&,
|
917
|
+
const debug_alloc<_Alloc>&) {
|
918
|
+
return true;
|
919
|
+
}
|
920
|
+
|
921
|
+
#ifdef __STL_FUNCTION_TMPL_PARTIAL_ORDER
|
922
|
+
template <class _Alloc>
|
923
|
+
inline bool operator!=(const debug_alloc<_Alloc>&,
|
924
|
+
const debug_alloc<_Alloc>&) {
|
925
|
+
return false;
|
926
|
+
}
|
927
|
+
#endif /* __STL_FUNCTION_TMPL_PARTIAL_ORDER */
|
928
|
+
|
929
|
+
// Another allocator adaptor: _Alloc_traits. This serves two
|
930
|
+
// purposes. First, make it possible to write containers that can use
|
931
|
+
// either SGI-style allocators or standard-conforming allocator.
|
932
|
+
// Second, provide a mechanism so that containers can query whether or
|
933
|
+
// not the allocator has distinct instances. If not, the container
|
934
|
+
// can avoid wasting a word of memory to store an empty object.
|
935
|
+
|
936
|
+
// This adaptor uses partial specialization. The general case of
|
937
|
+
// _Alloc_traits<_Tp, _Alloc> assumes that _Alloc is a
|
938
|
+
// standard-conforming allocator, possibly with non-equal instances
|
939
|
+
// and non-static members. (It still behaves correctly even if _Alloc
|
940
|
+
// has static member and if all instances are equal. Refinements
|
941
|
+
// affect performance, not correctness.)
|
942
|
+
|
943
|
+
// There are always two members: allocator_type, which is a standard-
|
944
|
+
// conforming allocator type for allocating objects of type _Tp, and
|
945
|
+
// _S_instanceless, a static const member of type bool. If
|
946
|
+
// _S_instanceless is true, this means that there is no difference
|
947
|
+
// between any two instances of type allocator_type. Furthermore, if
|
948
|
+
// _S_instanceless is true, then _Alloc_traits has one additional
|
949
|
+
// member: _Alloc_type. This type encapsulates allocation and
|
950
|
+
// deallocation of objects of type _Tp through a static interface; it
|
951
|
+
// has two member functions, whose signatures are
|
952
|
+
// static _Tp* allocate(size_t)
|
953
|
+
// static void deallocate(_Tp*, size_t)
|
954
|
+
|
955
|
+
// The fully general version.
|
956
|
+
|
957
|
+
template <class _Tp, class _Allocator>
|
958
|
+
struct _Alloc_traits
|
959
|
+
{
|
960
|
+
static const bool _S_instanceless = false;
|
961
|
+
typedef typename _Allocator::template rebind<_Tp>::other
|
962
|
+
allocator_type;
|
963
|
+
};
|
964
|
+
|
965
|
+
template <class _Tp, class _Allocator>
|
966
|
+
const bool _Alloc_traits<_Tp, _Allocator>::_S_instanceless;
|
967
|
+
|
968
|
+
// The version for the default allocator.
|
969
|
+
|
970
|
+
template <class _Tp, class _Tp1>
|
971
|
+
struct _Alloc_traits<_Tp, allocator<_Tp1> >
|
972
|
+
{
|
973
|
+
static const bool _S_instanceless = true;
|
974
|
+
typedef simple_alloc<_Tp, alloc> _Alloc_type;
|
975
|
+
typedef allocator<_Tp> allocator_type;
|
976
|
+
};
|
977
|
+
|
978
|
+
// Versions for the predefined SGI-style allocators.
|
979
|
+
|
980
|
+
template <class _Tp, int __inst>
|
981
|
+
struct _Alloc_traits<_Tp, __malloc_alloc_template<__inst> >
|
982
|
+
{
|
983
|
+
static const bool _S_instanceless = true;
|
984
|
+
typedef simple_alloc<_Tp, __malloc_alloc_template<__inst> > _Alloc_type;
|
985
|
+
typedef __allocator<_Tp, __malloc_alloc_template<__inst> > allocator_type;
|
986
|
+
};
|
987
|
+
|
988
|
+
#ifndef __USE_MALLOC
|
989
|
+
template <class _Tp, bool __threads, int __inst>
|
990
|
+
struct _Alloc_traits<_Tp, __default_alloc_template<__threads, __inst> >
|
991
|
+
{
|
992
|
+
static const bool _S_instanceless = true;
|
993
|
+
typedef simple_alloc<_Tp, __default_alloc_template<__threads, __inst> >
|
994
|
+
_Alloc_type;
|
995
|
+
typedef __allocator<_Tp, __default_alloc_template<__threads, __inst> >
|
996
|
+
allocator_type;
|
997
|
+
};
|
998
|
+
#endif
|
999
|
+
|
1000
|
+
template <class _Tp, class _Alloc>
|
1001
|
+
struct _Alloc_traits<_Tp, debug_alloc<_Alloc> >
|
1002
|
+
{
|
1003
|
+
static const bool _S_instanceless = true;
|
1004
|
+
typedef simple_alloc<_Tp, debug_alloc<_Alloc> > _Alloc_type;
|
1005
|
+
typedef __allocator<_Tp, debug_alloc<_Alloc> > allocator_type;
|
1006
|
+
};
|
1007
|
+
|
1008
|
+
// Versions for the __allocator adaptor used with the predefined
|
1009
|
+
// SGI-style allocators.
|
1010
|
+
|
1011
|
+
template <class _Tp, class _Tp1, int __inst>
|
1012
|
+
struct _Alloc_traits<_Tp,
|
1013
|
+
__allocator<_Tp1, __malloc_alloc_template<__inst> > >
|
1014
|
+
{
|
1015
|
+
static const bool _S_instanceless = true;
|
1016
|
+
typedef simple_alloc<_Tp, __malloc_alloc_template<__inst> > _Alloc_type;
|
1017
|
+
typedef __allocator<_Tp, __malloc_alloc_template<__inst> > allocator_type;
|
1018
|
+
};
|
1019
|
+
|
1020
|
+
#ifndef __USE_MALLOC
|
1021
|
+
template <class _Tp, class _Tp1, bool __thr, int __inst>
|
1022
|
+
struct _Alloc_traits<_Tp,
|
1023
|
+
__allocator<_Tp1,
|
1024
|
+
__default_alloc_template<__thr, __inst> > >
|
1025
|
+
{
|
1026
|
+
static const bool _S_instanceless = true;
|
1027
|
+
typedef simple_alloc<_Tp, __default_alloc_template<__thr,__inst> >
|
1028
|
+
_Alloc_type;
|
1029
|
+
typedef __allocator<_Tp, __default_alloc_template<__thr,__inst> >
|
1030
|
+
allocator_type;
|
1031
|
+
};
|
1032
|
+
#endif
|
1033
|
+
|
1034
|
+
template <class _Tp, class _Tp1, class _Alloc>
|
1035
|
+
struct _Alloc_traits<_Tp, __allocator<_Tp1, debug_alloc<_Alloc> > >
|
1036
|
+
{
|
1037
|
+
static const bool _S_instanceless = true;
|
1038
|
+
typedef simple_alloc<_Tp, debug_alloc<_Alloc> > _Alloc_type;
|
1039
|
+
typedef __allocator<_Tp, debug_alloc<_Alloc> > allocator_type;
|
1040
|
+
};
|
1041
|
+
|
1042
|
+
|
1043
|
+
#endif /* __STL_USE_STD_ALLOCATORS */
|
1044
|
+
|
1045
|
+
#if defined(__sgi) && !defined(__GNUC__) && (_MIPS_SIM != _MIPS_SIM_ABI32)
|
1046
|
+
#pragma reset woff 1174
|
1047
|
+
#endif
|
1048
|
+
|
1049
|
+
__STL_END_NAMESPACE
|
1050
|
+
|
1051
|
+
#undef __PRIVATE
|
1052
|
+
|
1053
|
+
#endif /* __SGI_STL_INTERNAL_ALLOC_H */
|
1054
|
+
|
1055
|
+
// Local Variables:
|
1056
|
+
// mode:C++
|
1057
|
+
// End:
|