1 // Reference-counted versatile string base -*- C++ -*-
3 // Copyright (C) 2005, 2006, 2007, 2008, 2009, 2010
4 // Free Software Foundation, Inc.
6 // This file is part of the GNU ISO C++ Library. This library is free
7 // software; you can redistribute it and/or modify it under the
8 // terms of the GNU General Public License as published by the
9 // Free Software Foundation; either version 3, or (at your option)
12 // This library is distributed in the hope that it will be useful,
13 // but WITHOUT ANY WARRANTY; without even the implied warranty of
14 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 // GNU General Public License for more details.
17 // Under Section 7 of GPL version 3, you are granted additional
18 // permissions described in the GCC Runtime Library Exception, version
19 // 3.1, as published by the Free Software Foundation.
21 // You should have received a copy of the GNU General Public License and
22 // a copy of the GCC Runtime Library Exception along with this program;
23 // see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
24 // <http://www.gnu.org/licenses/>.
26 /** @file ext/rc_string_base.h
27 * This is an internal header file, included by other library headers.
28 * Do not attempt to use it directly. @headername{ext/vstring.h}
31 #ifndef _RC_STRING_BASE_H
32 #define _RC_STRING_BASE_H 1
34 #include <ext/atomicity.h>
35 #include <bits/stl_iterator_base_funcs.h>
37 namespace __gnu_cxx
_GLIBCXX_VISIBILITY(default)
39 _GLIBCXX_BEGIN_NAMESPACE_VERSION
42 * Documentation? What's that?
43 * Nathan Myers <ncm@cantrip.org>.
45 * A string looks like this:
50 * [__rc_string_base<char_type>] _M_capacity
51 * _M_dataplus _M_refcount
52 * _M_p ----------------> unnamed array of char_type
55 * Where the _M_p points to the first character in the string, and
56 * you cast it to a pointer-to-_Rep and subtract 1 to get a
57 * pointer to the header.
59 * This approach has the enormous advantage that a string object
60 * requires only one allocation. All the ugliness is confined
61 * within a single pair of inline functions, which each compile to
62 * a single @a add instruction: _Rep::_M_refdata(), and
63 * __rc_string_base::_M_rep(); and the allocation function which gets a
64 * block of raw bytes and with room enough and constructs a _Rep
65 * object at the front.
67 * The reason you want _M_data pointing to the character array and
68 * not the _Rep is so that the debugger can see the string
69 * contents. (Probably we should add a non-inline member to get
70 * the _Rep for the debugger to use, so users can check the actual
73 * Note that the _Rep object is a POD so that you can have a
74 * static <em>empty string</em> _Rep object already @a constructed before
75 * static constructors have run. The reference-count encoding is
76 * chosen so that a 0 indicates one reference, so you never try to
77 * destroy the empty-string _Rep object.
79 * All but the last paragraph is considered pretty conventional
80 * for a C++ string implementation.
82 template<typename _CharT
, typename _Traits
, typename _Alloc
>
83 class __rc_string_base
84 : protected __vstring_utility
<_CharT
, _Traits
, _Alloc
>
87 typedef _Traits traits_type
;
88 typedef typename
_Traits::char_type value_type
;
89 typedef _Alloc allocator_type
;
91 typedef __vstring_utility
<_CharT
, _Traits
, _Alloc
> _Util_Base
;
92 typedef typename
_Util_Base::_CharT_alloc_type _CharT_alloc_type
;
93 typedef typename
_CharT_alloc_type::size_type size_type
;
96 // _Rep: string representation
98 // 1. String really contains _M_length + 1 characters: due to 21.3.4
99 // must be kept null-terminated.
100 // 2. _M_capacity >= _M_length
101 // Allocated memory is always (_M_capacity + 1) * sizeof(_CharT).
102 // 3. _M_refcount has three states:
103 // -1: leaked, one reference, no ref-copies allowed, non-const.
104 // 0: one reference, non-const.
105 // n>0: n + 1 references, operations require a lock, const.
106 // 4. All fields == 0 is an empty string, given the extra storage
107 // beyond-the-end for a null terminator; thus, the shared
108 // empty string representation needs no constructor.
116 size_type _M_capacity
;
117 _Atomic_word _M_refcount
;
120 // Only for alignment purposes.
124 typedef typename
_Alloc::template rebind
<_Rep
>::other _Rep_alloc_type
;
128 { return reinterpret_cast<_CharT
*>(this + 1); }
133 __atomic_add_dispatch(&_M_info
._M_refcount
, 1);
138 _M_set_length(size_type __n
)
140 _M_info
._M_refcount
= 0; // One reference.
141 _M_info
._M_length
= __n
;
142 // grrr. (per 21.3.4)
143 // You cannot leave those LWG people alone for a second.
144 traits_type::assign(_M_refdata()[__n
], _CharT());
149 _S_create(size_type
, size_type
, const _Alloc
&);
152 _M_destroy(const _Alloc
&) throw();
155 _M_clone(const _Alloc
&, size_type __res
= 0);
164 static _Rep_empty _S_empty_rep
;
166 // The maximum number of individual char_type elements of an
167 // individual string is determined by _S_max_size. This is the
168 // value that will be returned by max_size(). (Whereas npos
169 // is the maximum number of bytes the allocator can allocate.)
170 // If one was to divvy up the theoretical largest size string,
171 // with a terminating character and m _CharT elements, it'd
173 // npos = sizeof(_Rep) + (m * sizeof(_CharT)) + sizeof(_CharT)
174 // + sizeof(_Rep) - 1
175 // (NB: last two terms for rounding reasons, see _M_create below)
177 // m = ((npos - 2 * sizeof(_Rep) + 1) / sizeof(_CharT)) - 1
178 // In addition, this implementation halves this amount.
179 enum { _S_max_size
= (((static_cast<size_type
>(-1) - 2 * sizeof(_Rep
)
180 + 1) / sizeof(_CharT
)) - 1) / 2 };
182 // Data Member (private):
183 mutable typename
_Util_Base::template _Alloc_hider
<_Alloc
> _M_dataplus
;
187 { _M_dataplus
._M_p
= __p
; }
191 { return &((reinterpret_cast<_Rep
*>(_M_data()))[-1]); }
194 _M_grab(const _Alloc
& __alloc
) const
196 return (!_M_is_leaked() && _M_get_allocator() == __alloc
)
197 ? _M_rep()->_M_refcopy() : _M_rep()->_M_clone(__alloc
);
203 // Be race-detector-friendly. For more info see bits/c++config.
204 _GLIBCXX_SYNCHRONIZATION_HAPPENS_BEFORE(&_M_rep()->_M_info
.
206 if (__exchange_and_add_dispatch(&_M_rep()->_M_info
._M_refcount
,
209 _GLIBCXX_SYNCHRONIZATION_HAPPENS_AFTER(&_M_rep()->_M_info
.
211 _M_rep()->_M_destroy(_M_get_allocator());
217 { return _M_rep()->_M_info
._M_refcount
< 0; }
221 { _M_rep()->_M_info
._M_refcount
= 0; }
226 // _S_construct_aux is used to implement the 21.3.1 para 15 which
227 // requires special behaviour if _InIterator is an integral type
228 template<typename _InIterator
>
230 _S_construct_aux(_InIterator __beg
, _InIterator __end
,
231 const _Alloc
& __a
, std::__false_type
)
233 typedef typename iterator_traits
<_InIterator
>::iterator_category _Tag
;
234 return _S_construct(__beg
, __end
, __a
, _Tag());
237 // _GLIBCXX_RESOLVE_LIB_DEFECTS
238 // 438. Ambiguity in the "do the right thing" clause
239 template<typename _Integer
>
241 _S_construct_aux(_Integer __beg
, _Integer __end
,
242 const _Alloc
& __a
, std::__true_type
)
243 { return _S_construct_aux_2(static_cast<size_type
>(__beg
),
247 _S_construct_aux_2(size_type __req
, _CharT __c
, const _Alloc
& __a
)
248 { return _S_construct(__req
, __c
, __a
); }
250 template<typename _InIterator
>
252 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
)
254 typedef typename
std::__is_integer
<_InIterator
>::__type _Integral
;
255 return _S_construct_aux(__beg
, __end
, __a
, _Integral());
258 // For Input Iterators, used in istreambuf_iterators, etc.
259 template<typename _InIterator
>
261 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
262 std::input_iterator_tag
);
264 // For forward_iterators up to random_access_iterators, used for
265 // string::iterator, _CharT*, etc.
266 template<typename _FwdIterator
>
268 _S_construct(_FwdIterator __beg
, _FwdIterator __end
, const _Alloc
& __a
,
269 std::forward_iterator_tag
);
272 _S_construct(size_type __req
, _CharT __c
, const _Alloc
& __a
);
277 { return size_type(_S_max_size
); }
281 { return _M_dataplus
._M_p
; }
285 { return _M_rep()->_M_info
._M_length
; }
289 { return _M_rep()->_M_info
._M_capacity
; }
293 { return _M_rep()->_M_info
._M_refcount
> 0; }
297 { _M_rep()->_M_info
._M_refcount
= -1; }
300 _M_leak() // for use in begin() & non-const op[]
307 _M_set_length(size_type __n
)
308 { _M_rep()->_M_set_length(__n
); }
311 : _M_dataplus(_S_empty_rep
._M_refcopy()) { }
313 __rc_string_base(const _Alloc
& __a
);
315 __rc_string_base(const __rc_string_base
& __rcs
);
317 #ifdef __GXX_EXPERIMENTAL_CXX0X__
318 __rc_string_base(__rc_string_base
&& __rcs
)
319 : _M_dataplus(__rcs
._M_dataplus
)
320 { __rcs
._M_data(_S_empty_rep
._M_refcopy()); }
323 __rc_string_base(size_type __n
, _CharT __c
, const _Alloc
& __a
);
325 template<typename _InputIterator
>
326 __rc_string_base(_InputIterator __beg
, _InputIterator __end
,
334 { return _M_dataplus
; }
336 const allocator_type
&
337 _M_get_allocator() const
338 { return _M_dataplus
; }
341 _M_swap(__rc_string_base
& __rcs
);
344 _M_assign(const __rc_string_base
& __rcs
);
347 _M_reserve(size_type __res
);
350 _M_mutate(size_type __pos
, size_type __len1
, const _CharT
* __s
,
354 _M_erase(size_type __pos
, size_type __n
);
358 { _M_erase(size_type(0), _M_length()); }
361 _M_compare(const __rc_string_base
&) const
365 template<typename _CharT
, typename _Traits
, typename _Alloc
>
366 typename __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep_empty
367 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_S_empty_rep
;
369 template<typename _CharT
, typename _Traits
, typename _Alloc
>
370 typename __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep
*
371 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
372 _S_create(size_type __capacity
, size_type __old_capacity
,
373 const _Alloc
& __alloc
)
375 // _GLIBCXX_RESOLVE_LIB_DEFECTS
376 // 83. String::npos vs. string::max_size()
377 if (__capacity
> size_type(_S_max_size
))
378 std::__throw_length_error(__N("__rc_string_base::_Rep::_S_create"));
380 // The standard places no restriction on allocating more memory
381 // than is strictly needed within this layer at the moment or as
382 // requested by an explicit application call to reserve().
384 // Many malloc implementations perform quite poorly when an
385 // application attempts to allocate memory in a stepwise fashion
386 // growing each allocation size by only 1 char. Additionally,
387 // it makes little sense to allocate less linear memory than the
388 // natural blocking size of the malloc implementation.
389 // Unfortunately, we would need a somewhat low-level calculation
390 // with tuned parameters to get this perfect for any particular
391 // malloc implementation. Fortunately, generalizations about
392 // common features seen among implementations seems to suffice.
394 // __pagesize need not match the actual VM page size for good
395 // results in practice, thus we pick a common value on the low
396 // side. __malloc_header_size is an estimate of the amount of
397 // overhead per memory allocation (in practice seen N * sizeof
398 // (void*) where N is 0, 2 or 4). According to folklore,
399 // picking this value on the high side is better than
400 // low-balling it (especially when this algorithm is used with
401 // malloc implementations that allocate memory blocks rounded up
402 // to a size which is a power of 2).
403 const size_type __pagesize
= 4096;
404 const size_type __malloc_header_size
= 4 * sizeof(void*);
406 // The below implements an exponential growth policy, necessary to
407 // meet amortized linear time requirements of the library: see
408 // http://gcc.gnu.org/ml/libstdc++/2001-07/msg00085.html.
409 if (__capacity
> __old_capacity
&& __capacity
< 2 * __old_capacity
)
411 __capacity
= 2 * __old_capacity
;
412 // Never allocate a string bigger than _S_max_size.
413 if (__capacity
> size_type(_S_max_size
))
414 __capacity
= size_type(_S_max_size
);
417 // NB: Need an array of char_type[__capacity], plus a terminating
418 // null char_type() element, plus enough for the _Rep data structure,
419 // plus sizeof(_Rep) - 1 to upper round to a size multiple of
421 // Whew. Seemingly so needy, yet so elemental.
422 size_type __size
= ((__capacity
+ 1) * sizeof(_CharT
)
423 + 2 * sizeof(_Rep
) - 1);
425 const size_type __adj_size
= __size
+ __malloc_header_size
;
426 if (__adj_size
> __pagesize
&& __capacity
> __old_capacity
)
428 const size_type __extra
= __pagesize
- __adj_size
% __pagesize
;
429 __capacity
+= __extra
/ sizeof(_CharT
);
430 if (__capacity
> size_type(_S_max_size
))
431 __capacity
= size_type(_S_max_size
);
432 __size
= (__capacity
+ 1) * sizeof(_CharT
) + 2 * sizeof(_Rep
) - 1;
435 // NB: Might throw, but no worries about a leak, mate: _Rep()
437 _Rep
* __place
= _Rep_alloc_type(__alloc
).allocate(__size
/ sizeof(_Rep
));
438 _Rep
* __p
= new (__place
) _Rep
;
439 __p
->_M_info
._M_capacity
= __capacity
;
443 template<typename _CharT
, typename _Traits
, typename _Alloc
>
445 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
446 _M_destroy(const _Alloc
& __a
) throw ()
448 const size_type __size
= ((_M_info
._M_capacity
+ 1) * sizeof(_CharT
)
449 + 2 * sizeof(_Rep
) - 1);
450 _Rep_alloc_type(__a
).deallocate(this, __size
/ sizeof(_Rep
));
453 template<typename _CharT
, typename _Traits
, typename _Alloc
>
455 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
456 _M_clone(const _Alloc
& __alloc
, size_type __res
)
458 // Requested capacity of the clone.
459 const size_type __requested_cap
= _M_info
._M_length
+ __res
;
460 _Rep
* __r
= _Rep::_S_create(__requested_cap
, _M_info
._M_capacity
,
463 if (_M_info
._M_length
)
464 __rc_string_base::_S_copy(__r
->_M_refdata(), _M_refdata(), _M_info
._M_length
);
466 __r
->_M_set_length(_M_info
._M_length
);
467 return __r
->_M_refdata();
470 template<typename _CharT
, typename _Traits
, typename _Alloc
>
471 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
472 __rc_string_base(const _Alloc
& __a
)
473 : _M_dataplus(__a
, _S_construct(size_type(), _CharT(), __a
)) { }
475 template<typename _CharT
, typename _Traits
, typename _Alloc
>
476 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
477 __rc_string_base(const __rc_string_base
& __rcs
)
478 : _M_dataplus(__rcs
._M_get_allocator(),
479 __rcs
._M_grab(__rcs
._M_get_allocator())) { }
481 template<typename _CharT
, typename _Traits
, typename _Alloc
>
482 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
483 __rc_string_base(size_type __n
, _CharT __c
, const _Alloc
& __a
)
484 : _M_dataplus(__a
, _S_construct(__n
, __c
, __a
)) { }
486 template<typename _CharT
, typename _Traits
, typename _Alloc
>
487 template<typename _InputIterator
>
488 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
489 __rc_string_base(_InputIterator __beg
, _InputIterator __end
,
491 : _M_dataplus(__a
, _S_construct(__beg
, __end
, __a
)) { }
493 template<typename _CharT
, typename _Traits
, typename _Alloc
>
495 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
503 // NB: This is the special case for Input Iterators, used in
504 // istreambuf_iterators, etc.
505 // Input Iterators have a cost structure very different from
506 // pointers, calling for a different coding style.
507 template<typename _CharT
, typename _Traits
, typename _Alloc
>
508 template<typename _InIterator
>
510 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
511 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
512 std::input_iterator_tag
)
514 if (__beg
== __end
&& __a
== _Alloc())
515 return _S_empty_rep
._M_refcopy();
517 // Avoid reallocation for common case.
520 while (__beg
!= __end
&& __len
< sizeof(__buf
) / sizeof(_CharT
))
522 __buf
[__len
++] = *__beg
;
525 _Rep
* __r
= _Rep::_S_create(__len
, size_type(0), __a
);
526 _S_copy(__r
->_M_refdata(), __buf
, __len
);
529 while (__beg
!= __end
)
531 if (__len
== __r
->_M_info
._M_capacity
)
533 // Allocate more space.
534 _Rep
* __another
= _Rep::_S_create(__len
+ 1, __len
, __a
);
535 _S_copy(__another
->_M_refdata(), __r
->_M_refdata(), __len
);
536 __r
->_M_destroy(__a
);
539 __r
->_M_refdata()[__len
++] = *__beg
;
545 __r
->_M_destroy(__a
);
546 __throw_exception_again
;
548 __r
->_M_set_length(__len
);
549 return __r
->_M_refdata();
552 template<typename _CharT
, typename _Traits
, typename _Alloc
>
553 template<typename _InIterator
>
555 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
556 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
557 std::forward_iterator_tag
)
559 if (__beg
== __end
&& __a
== _Alloc())
560 return _S_empty_rep
._M_refcopy();
562 // NB: Not required, but considered best practice.
563 if (__is_null_pointer(__beg
) && __beg
!= __end
)
564 std::__throw_logic_error(__N("__rc_string_base::"
565 "_S_construct null not valid"));
567 const size_type __dnew
= static_cast<size_type
>(std::distance(__beg
,
569 // Check for out_of_range and length_error exceptions.
570 _Rep
* __r
= _Rep::_S_create(__dnew
, size_type(0), __a
);
572 { __rc_string_base::_S_copy_chars(__r
->_M_refdata(), __beg
, __end
); }
575 __r
->_M_destroy(__a
);
576 __throw_exception_again
;
578 __r
->_M_set_length(__dnew
);
579 return __r
->_M_refdata();
582 template<typename _CharT
, typename _Traits
, typename _Alloc
>
584 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
585 _S_construct(size_type __n
, _CharT __c
, const _Alloc
& __a
)
587 if (__n
== 0 && __a
== _Alloc())
588 return _S_empty_rep
._M_refcopy();
590 // Check for out_of_range and length_error exceptions.
591 _Rep
* __r
= _Rep::_S_create(__n
, size_type(0), __a
);
593 __rc_string_base::_S_assign(__r
->_M_refdata(), __n
, __c
);
595 __r
->_M_set_length(__n
);
596 return __r
->_M_refdata();
599 template<typename _CharT
, typename _Traits
, typename _Alloc
>
601 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
602 _M_swap(__rc_string_base
& __rcs
)
606 if (__rcs
._M_is_leaked())
607 __rcs
._M_set_sharable();
609 _CharT
* __tmp
= _M_data();
610 _M_data(__rcs
._M_data());
611 __rcs
._M_data(__tmp
);
613 // _GLIBCXX_RESOLVE_LIB_DEFECTS
614 // 431. Swapping containers with unequal allocators.
615 std::__alloc_swap
<allocator_type
>::_S_do_it(_M_get_allocator(),
616 __rcs
._M_get_allocator());
619 template<typename _CharT
, typename _Traits
, typename _Alloc
>
621 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
622 _M_assign(const __rc_string_base
& __rcs
)
624 if (_M_rep() != __rcs
._M_rep())
626 _CharT
* __tmp
= __rcs
._M_grab(_M_get_allocator());
632 template<typename _CharT
, typename _Traits
, typename _Alloc
>
634 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
635 _M_reserve(size_type __res
)
637 // Make sure we don't shrink below the current size.
638 if (__res
< _M_length())
641 if (__res
!= _M_capacity() || _M_is_shared())
643 _CharT
* __tmp
= _M_rep()->_M_clone(_M_get_allocator(),
644 __res
- _M_length());
650 template<typename _CharT
, typename _Traits
, typename _Alloc
>
652 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
653 _M_mutate(size_type __pos
, size_type __len1
, const _CharT
* __s
,
656 const size_type __how_much
= _M_length() - __pos
- __len1
;
658 _Rep
* __r
= _Rep::_S_create(_M_length() + __len2
- __len1
,
659 _M_capacity(), _M_get_allocator());
662 this->_S_copy(__r
->_M_refdata(), _M_data(), __pos
);
664 this->_S_copy(__r
->_M_refdata() + __pos
, __s
, __len2
);
666 this->_S_copy(__r
->_M_refdata() + __pos
+ __len2
,
667 _M_data() + __pos
+ __len1
, __how_much
);
670 _M_data(__r
->_M_refdata());
673 template<typename _CharT
, typename _Traits
, typename _Alloc
>
675 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
676 _M_erase(size_type __pos
, size_type __n
)
678 const size_type __new_size
= _M_length() - __n
;
679 const size_type __how_much
= _M_length() - __pos
- __n
;
684 _Rep
* __r
= _Rep::_S_create(__new_size
, _M_capacity(),
688 this->_S_copy(__r
->_M_refdata(), _M_data(), __pos
);
690 this->_S_copy(__r
->_M_refdata() + __pos
,
691 _M_data() + __pos
+ __n
, __how_much
);
694 _M_data(__r
->_M_refdata());
696 else if (__how_much
&& __n
)
699 this->_S_move(_M_data() + __pos
,
700 _M_data() + __pos
+ __n
, __how_much
);
703 _M_rep()->_M_set_length(__new_size
);
708 __rc_string_base
<char, std::char_traits
<char>,
709 std::allocator
<char> >::
710 _M_compare(const __rc_string_base
& __rcs
) const
712 if (_M_rep() == __rcs
._M_rep())
717 #ifdef _GLIBCXX_USE_WCHAR_T
720 __rc_string_base
<wchar_t, std::char_traits
<wchar_t>,
721 std::allocator
<wchar_t> >::
722 _M_compare(const __rc_string_base
& __rcs
) const
724 if (_M_rep() == __rcs
._M_rep())
730 _GLIBCXX_END_NAMESPACE_VERSION
733 #endif /* _RC_STRING_BASE_H */