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 _GLIBCXX_BEGIN_NAMESPACE(__gnu_cxx
)
40 * Documentation? What's that?
41 * Nathan Myers <ncm@cantrip.org>.
43 * A string looks like this:
48 * [__rc_string_base<char_type>] _M_capacity
49 * _M_dataplus _M_refcount
50 * _M_p ----------------> unnamed array of char_type
53 * Where the _M_p points to the first character in the string, and
54 * you cast it to a pointer-to-_Rep and subtract 1 to get a
55 * pointer to the header.
57 * This approach has the enormous advantage that a string object
58 * requires only one allocation. All the ugliness is confined
59 * within a single pair of inline functions, which each compile to
60 * a single @a add instruction: _Rep::_M_refdata(), and
61 * __rc_string_base::_M_rep(); and the allocation function which gets a
62 * block of raw bytes and with room enough and constructs a _Rep
63 * object at the front.
65 * The reason you want _M_data pointing to the character array and
66 * not the _Rep is so that the debugger can see the string
67 * contents. (Probably we should add a non-inline member to get
68 * the _Rep for the debugger to use, so users can check the actual
71 * Note that the _Rep object is a POD so that you can have a
72 * static <em>empty string</em> _Rep object already @a constructed before
73 * static constructors have run. The reference-count encoding is
74 * chosen so that a 0 indicates one reference, so you never try to
75 * destroy the empty-string _Rep object.
77 * All but the last paragraph is considered pretty conventional
78 * for a C++ string implementation.
80 template<typename _CharT
, typename _Traits
, typename _Alloc
>
81 class __rc_string_base
82 : protected __vstring_utility
<_CharT
, _Traits
, _Alloc
>
85 typedef _Traits traits_type
;
86 typedef typename
_Traits::char_type value_type
;
87 typedef _Alloc allocator_type
;
89 typedef __vstring_utility
<_CharT
, _Traits
, _Alloc
> _Util_Base
;
90 typedef typename
_Util_Base::_CharT_alloc_type _CharT_alloc_type
;
91 typedef typename
_CharT_alloc_type::size_type size_type
;
94 // _Rep: string representation
96 // 1. String really contains _M_length + 1 characters: due to 21.3.4
97 // must be kept null-terminated.
98 // 2. _M_capacity >= _M_length
99 // Allocated memory is always (_M_capacity + 1) * sizeof(_CharT).
100 // 3. _M_refcount has three states:
101 // -1: leaked, one reference, no ref-copies allowed, non-const.
102 // 0: one reference, non-const.
103 // n>0: n + 1 references, operations require a lock, const.
104 // 4. All fields == 0 is an empty string, given the extra storage
105 // beyond-the-end for a null terminator; thus, the shared
106 // empty string representation needs no constructor.
114 size_type _M_capacity
;
115 _Atomic_word _M_refcount
;
118 // Only for alignment purposes.
122 typedef typename
_Alloc::template rebind
<_Rep
>::other _Rep_alloc_type
;
126 { return reinterpret_cast<_CharT
*>(this + 1); }
131 __atomic_add_dispatch(&_M_info
._M_refcount
, 1);
136 _M_set_length(size_type __n
)
138 _M_info
._M_refcount
= 0; // One reference.
139 _M_info
._M_length
= __n
;
140 // grrr. (per 21.3.4)
141 // You cannot leave those LWG people alone for a second.
142 traits_type::assign(_M_refdata()[__n
], _CharT());
147 _S_create(size_type
, size_type
, const _Alloc
&);
150 _M_destroy(const _Alloc
&) throw();
153 _M_clone(const _Alloc
&, size_type __res
= 0);
162 static _Rep_empty _S_empty_rep
;
164 // The maximum number of individual char_type elements of an
165 // individual string is determined by _S_max_size. This is the
166 // value that will be returned by max_size(). (Whereas npos
167 // is the maximum number of bytes the allocator can allocate.)
168 // If one was to divvy up the theoretical largest size string,
169 // with a terminating character and m _CharT elements, it'd
171 // npos = sizeof(_Rep) + (m * sizeof(_CharT)) + sizeof(_CharT)
172 // + sizeof(_Rep) - 1
173 // (NB: last two terms for rounding reasons, see _M_create below)
175 // m = ((npos - 2 * sizeof(_Rep) + 1) / sizeof(_CharT)) - 1
176 // In addition, this implementation halves this amount.
177 enum { _S_max_size
= (((static_cast<size_type
>(-1) - 2 * sizeof(_Rep
)
178 + 1) / sizeof(_CharT
)) - 1) / 2 };
180 // Data Member (private):
181 mutable typename
_Util_Base::template _Alloc_hider
<_Alloc
> _M_dataplus
;
185 { _M_dataplus
._M_p
= __p
; }
189 { return &((reinterpret_cast<_Rep
*>(_M_data()))[-1]); }
192 _M_grab(const _Alloc
& __alloc
) const
194 return (!_M_is_leaked() && _M_get_allocator() == __alloc
)
195 ? _M_rep()->_M_refcopy() : _M_rep()->_M_clone(__alloc
);
201 // Be race-detector-friendly. For more info see bits/c++config.
202 _GLIBCXX_SYNCHRONIZATION_HAPPENS_BEFORE(&_M_rep()->_M_info
.
204 if (__exchange_and_add_dispatch(&_M_rep()->_M_info
._M_refcount
,
207 _GLIBCXX_SYNCHRONIZATION_HAPPENS_AFTER(&_M_rep()->_M_info
.
209 _M_rep()->_M_destroy(_M_get_allocator());
215 { return _M_rep()->_M_info
._M_refcount
< 0; }
219 { _M_rep()->_M_info
._M_refcount
= 0; }
224 // _S_construct_aux is used to implement the 21.3.1 para 15 which
225 // requires special behaviour if _InIterator is an integral type
226 template<typename _InIterator
>
228 _S_construct_aux(_InIterator __beg
, _InIterator __end
,
229 const _Alloc
& __a
, std::__false_type
)
231 typedef typename iterator_traits
<_InIterator
>::iterator_category _Tag
;
232 return _S_construct(__beg
, __end
, __a
, _Tag());
235 // _GLIBCXX_RESOLVE_LIB_DEFECTS
236 // 438. Ambiguity in the "do the right thing" clause
237 template<typename _Integer
>
239 _S_construct_aux(_Integer __beg
, _Integer __end
,
240 const _Alloc
& __a
, std::__true_type
)
241 { return _S_construct_aux_2(static_cast<size_type
>(__beg
),
245 _S_construct_aux_2(size_type __req
, _CharT __c
, const _Alloc
& __a
)
246 { return _S_construct(__req
, __c
, __a
); }
248 template<typename _InIterator
>
250 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
)
252 typedef typename
std::__is_integer
<_InIterator
>::__type _Integral
;
253 return _S_construct_aux(__beg
, __end
, __a
, _Integral());
256 // For Input Iterators, used in istreambuf_iterators, etc.
257 template<typename _InIterator
>
259 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
260 std::input_iterator_tag
);
262 // For forward_iterators up to random_access_iterators, used for
263 // string::iterator, _CharT*, etc.
264 template<typename _FwdIterator
>
266 _S_construct(_FwdIterator __beg
, _FwdIterator __end
, const _Alloc
& __a
,
267 std::forward_iterator_tag
);
270 _S_construct(size_type __req
, _CharT __c
, const _Alloc
& __a
);
275 { return size_type(_S_max_size
); }
279 { return _M_dataplus
._M_p
; }
283 { return _M_rep()->_M_info
._M_length
; }
287 { return _M_rep()->_M_info
._M_capacity
; }
291 { return _M_rep()->_M_info
._M_refcount
> 0; }
295 { _M_rep()->_M_info
._M_refcount
= -1; }
298 _M_leak() // for use in begin() & non-const op[]
305 _M_set_length(size_type __n
)
306 { _M_rep()->_M_set_length(__n
); }
309 : _M_dataplus(_S_empty_rep
._M_refcopy()) { }
311 __rc_string_base(const _Alloc
& __a
);
313 __rc_string_base(const __rc_string_base
& __rcs
);
315 #ifdef __GXX_EXPERIMENTAL_CXX0X__
316 __rc_string_base(__rc_string_base
&& __rcs
)
317 : _M_dataplus(__rcs
._M_dataplus
)
318 { __rcs
._M_data(_S_empty_rep
._M_refcopy()); }
321 __rc_string_base(size_type __n
, _CharT __c
, const _Alloc
& __a
);
323 template<typename _InputIterator
>
324 __rc_string_base(_InputIterator __beg
, _InputIterator __end
,
332 { return _M_dataplus
; }
334 const allocator_type
&
335 _M_get_allocator() const
336 { return _M_dataplus
; }
339 _M_swap(__rc_string_base
& __rcs
);
342 _M_assign(const __rc_string_base
& __rcs
);
345 _M_reserve(size_type __res
);
348 _M_mutate(size_type __pos
, size_type __len1
, const _CharT
* __s
,
352 _M_erase(size_type __pos
, size_type __n
);
356 { _M_erase(size_type(0), _M_length()); }
359 _M_compare(const __rc_string_base
&) const
363 template<typename _CharT
, typename _Traits
, typename _Alloc
>
364 typename __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep_empty
365 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_S_empty_rep
;
367 template<typename _CharT
, typename _Traits
, typename _Alloc
>
368 typename __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep
*
369 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
370 _S_create(size_type __capacity
, size_type __old_capacity
,
371 const _Alloc
& __alloc
)
373 // _GLIBCXX_RESOLVE_LIB_DEFECTS
374 // 83. String::npos vs. string::max_size()
375 if (__capacity
> size_type(_S_max_size
))
376 std::__throw_length_error(__N("__rc_string_base::_Rep::_S_create"));
378 // The standard places no restriction on allocating more memory
379 // than is strictly needed within this layer at the moment or as
380 // requested by an explicit application call to reserve().
382 // Many malloc implementations perform quite poorly when an
383 // application attempts to allocate memory in a stepwise fashion
384 // growing each allocation size by only 1 char. Additionally,
385 // it makes little sense to allocate less linear memory than the
386 // natural blocking size of the malloc implementation.
387 // Unfortunately, we would need a somewhat low-level calculation
388 // with tuned parameters to get this perfect for any particular
389 // malloc implementation. Fortunately, generalizations about
390 // common features seen among implementations seems to suffice.
392 // __pagesize need not match the actual VM page size for good
393 // results in practice, thus we pick a common value on the low
394 // side. __malloc_header_size is an estimate of the amount of
395 // overhead per memory allocation (in practice seen N * sizeof
396 // (void*) where N is 0, 2 or 4). According to folklore,
397 // picking this value on the high side is better than
398 // low-balling it (especially when this algorithm is used with
399 // malloc implementations that allocate memory blocks rounded up
400 // to a size which is a power of 2).
401 const size_type __pagesize
= 4096;
402 const size_type __malloc_header_size
= 4 * sizeof(void*);
404 // The below implements an exponential growth policy, necessary to
405 // meet amortized linear time requirements of the library: see
406 // http://gcc.gnu.org/ml/libstdc++/2001-07/msg00085.html.
407 if (__capacity
> __old_capacity
&& __capacity
< 2 * __old_capacity
)
409 __capacity
= 2 * __old_capacity
;
410 // Never allocate a string bigger than _S_max_size.
411 if (__capacity
> size_type(_S_max_size
))
412 __capacity
= size_type(_S_max_size
);
415 // NB: Need an array of char_type[__capacity], plus a terminating
416 // null char_type() element, plus enough for the _Rep data structure,
417 // plus sizeof(_Rep) - 1 to upper round to a size multiple of
419 // Whew. Seemingly so needy, yet so elemental.
420 size_type __size
= ((__capacity
+ 1) * sizeof(_CharT
)
421 + 2 * sizeof(_Rep
) - 1);
423 const size_type __adj_size
= __size
+ __malloc_header_size
;
424 if (__adj_size
> __pagesize
&& __capacity
> __old_capacity
)
426 const size_type __extra
= __pagesize
- __adj_size
% __pagesize
;
427 __capacity
+= __extra
/ sizeof(_CharT
);
428 if (__capacity
> size_type(_S_max_size
))
429 __capacity
= size_type(_S_max_size
);
430 __size
= (__capacity
+ 1) * sizeof(_CharT
) + 2 * sizeof(_Rep
) - 1;
433 // NB: Might throw, but no worries about a leak, mate: _Rep()
435 _Rep
* __place
= _Rep_alloc_type(__alloc
).allocate(__size
/ sizeof(_Rep
));
436 _Rep
* __p
= new (__place
) _Rep
;
437 __p
->_M_info
._M_capacity
= __capacity
;
441 template<typename _CharT
, typename _Traits
, typename _Alloc
>
443 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
444 _M_destroy(const _Alloc
& __a
) throw ()
446 const size_type __size
= ((_M_info
._M_capacity
+ 1) * sizeof(_CharT
)
447 + 2 * sizeof(_Rep
) - 1);
448 _Rep_alloc_type(__a
).deallocate(this, __size
/ sizeof(_Rep
));
451 template<typename _CharT
, typename _Traits
, typename _Alloc
>
453 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
454 _M_clone(const _Alloc
& __alloc
, size_type __res
)
456 // Requested capacity of the clone.
457 const size_type __requested_cap
= _M_info
._M_length
+ __res
;
458 _Rep
* __r
= _Rep::_S_create(__requested_cap
, _M_info
._M_capacity
,
461 if (_M_info
._M_length
)
462 _S_copy(__r
->_M_refdata(), _M_refdata(), _M_info
._M_length
);
464 __r
->_M_set_length(_M_info
._M_length
);
465 return __r
->_M_refdata();
468 template<typename _CharT
, typename _Traits
, typename _Alloc
>
469 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
470 __rc_string_base(const _Alloc
& __a
)
471 : _M_dataplus(__a
, _S_construct(size_type(), _CharT(), __a
)) { }
473 template<typename _CharT
, typename _Traits
, typename _Alloc
>
474 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
475 __rc_string_base(const __rc_string_base
& __rcs
)
476 : _M_dataplus(__rcs
._M_get_allocator(),
477 __rcs
._M_grab(__rcs
._M_get_allocator())) { }
479 template<typename _CharT
, typename _Traits
, typename _Alloc
>
480 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
481 __rc_string_base(size_type __n
, _CharT __c
, const _Alloc
& __a
)
482 : _M_dataplus(__a
, _S_construct(__n
, __c
, __a
)) { }
484 template<typename _CharT
, typename _Traits
, typename _Alloc
>
485 template<typename _InputIterator
>
486 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
487 __rc_string_base(_InputIterator __beg
, _InputIterator __end
,
489 : _M_dataplus(__a
, _S_construct(__beg
, __end
, __a
)) { }
491 template<typename _CharT
, typename _Traits
, typename _Alloc
>
493 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
501 // NB: This is the special case for Input Iterators, used in
502 // istreambuf_iterators, etc.
503 // Input Iterators have a cost structure very different from
504 // pointers, calling for a different coding style.
505 template<typename _CharT
, typename _Traits
, typename _Alloc
>
506 template<typename _InIterator
>
508 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
509 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
510 std::input_iterator_tag
)
512 if (__beg
== __end
&& __a
== _Alloc())
513 return _S_empty_rep
._M_refcopy();
515 // Avoid reallocation for common case.
518 while (__beg
!= __end
&& __len
< sizeof(__buf
) / sizeof(_CharT
))
520 __buf
[__len
++] = *__beg
;
523 _Rep
* __r
= _Rep::_S_create(__len
, size_type(0), __a
);
524 _S_copy(__r
->_M_refdata(), __buf
, __len
);
527 while (__beg
!= __end
)
529 if (__len
== __r
->_M_info
._M_capacity
)
531 // Allocate more space.
532 _Rep
* __another
= _Rep::_S_create(__len
+ 1, __len
, __a
);
533 _S_copy(__another
->_M_refdata(), __r
->_M_refdata(), __len
);
534 __r
->_M_destroy(__a
);
537 __r
->_M_refdata()[__len
++] = *__beg
;
543 __r
->_M_destroy(__a
);
544 __throw_exception_again
;
546 __r
->_M_set_length(__len
);
547 return __r
->_M_refdata();
550 template<typename _CharT
, typename _Traits
, typename _Alloc
>
551 template<typename _InIterator
>
553 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
554 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
555 std::forward_iterator_tag
)
557 if (__beg
== __end
&& __a
== _Alloc())
558 return _S_empty_rep
._M_refcopy();
560 // NB: Not required, but considered best practice.
561 if (__is_null_pointer(__beg
) && __beg
!= __end
)
562 std::__throw_logic_error(__N("__rc_string_base::"
563 "_S_construct null not valid"));
565 const size_type __dnew
= static_cast<size_type
>(std::distance(__beg
,
567 // Check for out_of_range and length_error exceptions.
568 _Rep
* __r
= _Rep::_S_create(__dnew
, size_type(0), __a
);
570 { _S_copy_chars(__r
->_M_refdata(), __beg
, __end
); }
573 __r
->_M_destroy(__a
);
574 __throw_exception_again
;
576 __r
->_M_set_length(__dnew
);
577 return __r
->_M_refdata();
580 template<typename _CharT
, typename _Traits
, typename _Alloc
>
582 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
583 _S_construct(size_type __n
, _CharT __c
, const _Alloc
& __a
)
585 if (__n
== 0 && __a
== _Alloc())
586 return _S_empty_rep
._M_refcopy();
588 // Check for out_of_range and length_error exceptions.
589 _Rep
* __r
= _Rep::_S_create(__n
, size_type(0), __a
);
591 _S_assign(__r
->_M_refdata(), __n
, __c
);
593 __r
->_M_set_length(__n
);
594 return __r
->_M_refdata();
597 template<typename _CharT
, typename _Traits
, typename _Alloc
>
599 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
600 _M_swap(__rc_string_base
& __rcs
)
604 if (__rcs
._M_is_leaked())
605 __rcs
._M_set_sharable();
607 _CharT
* __tmp
= _M_data();
608 _M_data(__rcs
._M_data());
609 __rcs
._M_data(__tmp
);
611 // _GLIBCXX_RESOLVE_LIB_DEFECTS
612 // 431. Swapping containers with unequal allocators.
613 std::__alloc_swap
<allocator_type
>::_S_do_it(_M_get_allocator(),
614 __rcs
._M_get_allocator());
617 template<typename _CharT
, typename _Traits
, typename _Alloc
>
619 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
620 _M_assign(const __rc_string_base
& __rcs
)
622 if (_M_rep() != __rcs
._M_rep())
624 _CharT
* __tmp
= __rcs
._M_grab(_M_get_allocator());
630 template<typename _CharT
, typename _Traits
, typename _Alloc
>
632 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
633 _M_reserve(size_type __res
)
635 // Make sure we don't shrink below the current size.
636 if (__res
< _M_length())
639 if (__res
!= _M_capacity() || _M_is_shared())
641 _CharT
* __tmp
= _M_rep()->_M_clone(_M_get_allocator(),
642 __res
- _M_length());
648 template<typename _CharT
, typename _Traits
, typename _Alloc
>
650 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
651 _M_mutate(size_type __pos
, size_type __len1
, const _CharT
* __s
,
654 const size_type __how_much
= _M_length() - __pos
- __len1
;
656 _Rep
* __r
= _Rep::_S_create(_M_length() + __len2
- __len1
,
657 _M_capacity(), _M_get_allocator());
660 _S_copy(__r
->_M_refdata(), _M_data(), __pos
);
662 _S_copy(__r
->_M_refdata() + __pos
, __s
, __len2
);
664 _S_copy(__r
->_M_refdata() + __pos
+ __len2
,
665 _M_data() + __pos
+ __len1
, __how_much
);
668 _M_data(__r
->_M_refdata());
671 template<typename _CharT
, typename _Traits
, typename _Alloc
>
673 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
674 _M_erase(size_type __pos
, size_type __n
)
676 const size_type __new_size
= _M_length() - __n
;
677 const size_type __how_much
= _M_length() - __pos
- __n
;
682 _Rep
* __r
= _Rep::_S_create(__new_size
, _M_capacity(),
686 _S_copy(__r
->_M_refdata(), _M_data(), __pos
);
688 _S_copy(__r
->_M_refdata() + __pos
,
689 _M_data() + __pos
+ __n
, __how_much
);
692 _M_data(__r
->_M_refdata());
694 else if (__how_much
&& __n
)
697 _S_move(_M_data() + __pos
,
698 _M_data() + __pos
+ __n
, __how_much
);
701 _M_rep()->_M_set_length(__new_size
);
706 __rc_string_base
<char, std::char_traits
<char>,
707 std::allocator
<char> >::
708 _M_compare(const __rc_string_base
& __rcs
) const
710 if (_M_rep() == __rcs
._M_rep())
715 #ifdef _GLIBCXX_USE_WCHAR_T
718 __rc_string_base
<wchar_t, std::char_traits
<wchar_t>,
719 std::allocator
<wchar_t> >::
720 _M_compare(const __rc_string_base
& __rcs
) const
722 if (_M_rep() == __rcs
._M_rep())
728 _GLIBCXX_END_NAMESPACE
730 #endif /* _RC_STRING_BASE_H */