1 // Reference-counted versatile string base -*- C++ -*-
3 // Copyright (C) 2005 Free Software Foundation, Inc.
5 // This file is part of the GNU ISO C++ Library. This library is free
6 // software; you can redistribute it and/or modify it under the
7 // terms of the GNU General Public License as published by the
8 // Free Software Foundation; either version 2, or (at your option)
11 // This library is distributed in the hope that it will be useful,
12 // but WITHOUT ANY WARRANTY; without even the implied warranty of
13 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 // GNU General Public License for more details.
16 // You should have received a copy of the GNU General Public License along
17 // with this library; see the file COPYING. If not, write to the Free
18 // Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
21 // As a special exception, you may use this file as part of a free software
22 // library without restriction. Specifically, if other files instantiate
23 // templates or use macros or inline functions from this file, or you compile
24 // this file and link it with other files to produce an executable, this
25 // file does not by itself cause the resulting executable to be covered by
26 // the GNU General Public License. This exception does not however
27 // invalidate any other reasons why the executable file might be covered by
28 // the GNU General Public License.
30 /** @file ext/rc_string_base.h
31 * This file is a GNU extension to the Standard C++ Library.
32 * This is an internal header file, included by other library headers.
33 * You should not attempt to use it directly.
36 #ifndef _RC_STRING_BASE_H
37 #define _RC_STRING_BASE_H 1
39 #include <bits/atomicity.h>
41 _GLIBCXX_BEGIN_NAMESPACE(__gnu_cxx
)
45 * Documentation? What's that?
46 * Nathan Myers <ncm@cantrip.org>.
48 * A string looks like this:
53 * [__rc_string_base<char_type>] _M_capacity
54 * _M_dataplus _M_refcount
55 * _M_p ----------------> unnamed array of char_type
58 * Where the _M_p points to the first character in the string, and
59 * you cast it to a pointer-to-_Rep and subtract 1 to get a
60 * pointer to the header.
62 * This approach has the enormous advantage that a string object
63 * requires only one allocation. All the ugliness is confined
64 * within a single pair of inline functions, which each compile to
65 * a single "add" instruction: _Rep::_M_refdata(), and
66 * __rc_string_base::_M_rep(); and the allocation function which gets a
67 * block of raw bytes and with room enough and constructs a _Rep
68 * object at the front.
70 * The reason you want _M_data pointing to the character array and
71 * not the _Rep is so that the debugger can see the string
72 * contents. (Probably we should add a non-inline member to get
73 * the _Rep for the debugger to use, so users can check the actual
76 * Note that the _Rep object is a POD so that you can have a
77 * static "empty string" _Rep object already "constructed" before
78 * static constructors have run. The reference-count encoding is
79 * chosen so that a 0 indicates one reference, so you never try to
80 * destroy the empty-string _Rep object.
82 * All but the last paragraph is considered pretty conventional
83 * for a C++ string implementation.
86 template<typename _CharT
, typename _Traits
, typename _Alloc
>
87 class __rc_string_base
88 : protected __vstring_utility
<_CharT
, _Traits
, _Alloc
>
91 typedef _Traits traits_type
;
92 typedef typename
_Traits::char_type value_type
;
93 typedef _Alloc allocator_type
;
95 typedef __vstring_utility
<_CharT
, _Traits
, _Alloc
> _Util_Base
;
96 typedef typename
_Util_Base::_CharT_alloc_type _CharT_alloc_type
;
97 typedef typename
_CharT_alloc_type::size_type size_type
;
100 // _Rep: string representation
102 // 1. String really contains _M_length + 1 characters: due to 21.3.4
103 // must be kept null-terminated.
104 // 2. _M_capacity >= _M_length
105 // Allocated memory is always (_M_capacity + 1) * sizeof(_CharT).
106 // 3. _M_refcount has three states:
107 // -1: leaked, one reference, no ref-copies allowed, non-const.
108 // 0: one reference, non-const.
109 // n>0: n + 1 references, operations require a lock, const.
110 // 4. All fields == 0 is an empty string, given the extra storage
111 // beyond-the-end for a null terminator; thus, the shared
112 // empty string representation needs no constructor.
120 size_type _M_capacity
;
121 _Atomic_word _M_refcount
;
124 // Only for alignment purposes.
128 typedef typename
_Alloc::template rebind
<_Rep
>::other _Rep_alloc_type
;
132 { return reinterpret_cast<_CharT
*>(this + 1); }
137 __atomic_add(&_M_info
._M_refcount
, 1);
142 _M_set_length(size_type __n
)
144 _M_info
._M_refcount
= 0; // One reference.
145 _M_info
._M_length
= __n
;
146 // grrr. (per 21.3.4)
147 // You cannot leave those LWG people alone for a second.
148 traits_type::assign(_M_refdata()[__n
], _CharT());
153 _S_create(size_type
, size_type
, const _Alloc
&);
156 _M_destroy(const _Alloc
&) throw();
159 _M_clone(const _Alloc
&, size_type __res
= 0);
168 static _Rep_empty _S_empty_rep
;
170 // The maximum number of individual char_type elements of an
171 // individual string is determined by _S_max_size. This is the
172 // value that will be returned by max_size(). (Whereas npos
173 // is the maximum number of bytes the allocator can allocate.)
174 // If one was to divvy up the theoretical largest size string,
175 // with a terminating character and m _CharT elements, it'd
177 // npos = sizeof(_Rep) + (m * sizeof(_CharT)) + sizeof(_CharT)
179 // m = ((npos - sizeof(_Rep)) / sizeof(_CharT)) - 1
180 // In addition, this implementation quarters this amount.
181 enum { _S_max_size
= (((static_cast<size_type
>(-1) - sizeof(_Rep
))
182 / sizeof(_CharT
)) - 1) / 4 };
184 // Data Member (private):
185 mutable typename
_Util_Base::template _Alloc_hider
<_Alloc
> _M_dataplus
;
189 { _M_dataplus
._M_p
= __p
; }
193 { return &((reinterpret_cast<_Rep
*>(_M_data()))[-1]); }
196 _M_grab(const _Alloc
& __alloc
) const
198 return (!_M_is_leaked() && _M_get_allocator() == __alloc
)
199 ? _M_rep()->_M_refcopy() : _M_rep()->_M_clone(__alloc
);
205 if (__exchange_and_add(&_M_rep()->_M_info
._M_refcount
, -1) <= 0)
206 _M_rep()->_M_destroy(_M_get_allocator());
211 { return _M_rep()->_M_info
._M_refcount
< 0; }
215 { _M_rep()->_M_info
._M_refcount
= 0; }
220 // _S_construct_aux is used to implement the 21.3.1 para 15 which
221 // requires special behaviour if _InIterator is an integral type
222 template<typename _InIterator
>
224 _S_construct_aux(_InIterator __beg
, _InIterator __end
,
225 const _Alloc
& __a
, __false_type
)
227 typedef typename iterator_traits
<_InIterator
>::iterator_category _Tag
;
228 return _S_construct(__beg
, __end
, __a
, _Tag());
231 template<typename _InIterator
>
233 _S_construct_aux(_InIterator __beg
, _InIterator __end
,
234 const _Alloc
& __a
, __true_type
)
235 { return _S_construct(static_cast<size_type
>(__beg
),
236 static_cast<value_type
>(__end
), __a
); }
238 template<typename _InIterator
>
240 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
)
242 typedef typename
std::__is_integer
<_InIterator
>::__type _Integral
;
243 return _S_construct_aux(__beg
, __end
, __a
, _Integral());
246 // For Input Iterators, used in istreambuf_iterators, etc.
247 template<typename _InIterator
>
249 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
250 std::input_iterator_tag
);
252 // For forward_iterators up to random_access_iterators, used for
253 // string::iterator, _CharT*, etc.
254 template<typename _FwdIterator
>
256 _S_construct(_FwdIterator __beg
, _FwdIterator __end
, const _Alloc
& __a
,
257 std::forward_iterator_tag
);
260 _S_construct(size_type __req
, _CharT __c
, const _Alloc
& __a
);
265 { return size_type(_S_max_size
); }
269 { return _M_dataplus
._M_p
; }
273 { return _M_rep()->_M_info
._M_length
; }
277 { return _M_rep()->_M_info
._M_capacity
; }
281 { return _M_rep()->_M_info
._M_refcount
> 0; }
285 { _M_rep()->_M_info
._M_refcount
= -1; }
288 _M_leak() // for use in begin() & non-const op[]
295 _M_set_length(size_type __n
)
296 { _M_rep()->_M_set_length(__n
); }
299 : _M_dataplus(_Alloc(), _S_empty_rep
._M_refcopy()) { }
301 __rc_string_base(const _Alloc
& __a
);
303 __rc_string_base(const __rc_string_base
& __rcs
);
305 __rc_string_base(size_type __n
, _CharT __c
, const _Alloc
& __a
);
307 template<typename _InputIterator
>
308 __rc_string_base(_InputIterator __beg
, _InputIterator __end
,
316 { return _M_dataplus
; }
318 const allocator_type
&
319 _M_get_allocator() const
320 { return _M_dataplus
; }
323 _M_swap(__rc_string_base
& __rcs
);
326 _M_assign(const __rc_string_base
& __rcs
);
329 _M_reserve(size_type __res
);
332 _M_mutate(size_type __pos
, size_type __len1
, const _CharT
* __s
,
336 _M_erase(size_type __pos
, size_type __n
);
339 _M_compare(const __rc_string_base
&) const
343 template<typename _CharT
, typename _Traits
, typename _Alloc
>
344 typename __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep_empty
345 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_S_empty_rep
;
347 template<typename _CharT
, typename _Traits
, typename _Alloc
>
348 typename __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep
*
349 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
350 _S_create(size_type __capacity
, size_type __old_capacity
,
351 const _Alloc
& __alloc
)
353 // _GLIBCXX_RESOLVE_LIB_DEFECTS
354 // 83. String::npos vs. string::max_size()
355 if (__capacity
> size_type(_S_max_size
))
356 std::__throw_length_error(__N("__rc_string_base::_Rep::_S_create"));
358 // The standard places no restriction on allocating more memory
359 // than is strictly needed within this layer at the moment or as
360 // requested by an explicit application call to reserve().
362 // Many malloc implementations perform quite poorly when an
363 // application attempts to allocate memory in a stepwise fashion
364 // growing each allocation size by only 1 char. Additionally,
365 // it makes little sense to allocate less linear memory than the
366 // natural blocking size of the malloc implementation.
367 // Unfortunately, we would need a somewhat low-level calculation
368 // with tuned parameters to get this perfect for any particular
369 // malloc implementation. Fortunately, generalizations about
370 // common features seen among implementations seems to suffice.
372 // __pagesize need not match the actual VM page size for good
373 // results in practice, thus we pick a common value on the low
374 // side. __malloc_header_size is an estimate of the amount of
375 // overhead per memory allocation (in practice seen N * sizeof
376 // (void*) where N is 0, 2 or 4). According to folklore,
377 // picking this value on the high side is better than
378 // low-balling it (especially when this algorithm is used with
379 // malloc implementations that allocate memory blocks rounded up
380 // to a size which is a power of 2).
381 const size_type __pagesize
= 4096;
382 const size_type __malloc_header_size
= 4 * sizeof(void*);
384 // The below implements an exponential growth policy, necessary to
385 // meet amortized linear time requirements of the library: see
386 // http://gcc.gnu.org/ml/libstdc++/2001-07/msg00085.html.
387 if (__capacity
> __old_capacity
&& __capacity
< 2 * __old_capacity
)
388 __capacity
= 2 * __old_capacity
;
390 // NB: Need an array of char_type[__capacity], plus a terminating
391 // null char_type() element, plus enough for the _Rep data structure,
392 // plus sizeof(_Rep) - 1 to upper round to a size multiple of
394 // Whew. Seemingly so needy, yet so elemental.
395 size_type __size
= ((__capacity
+ 1) * sizeof(_CharT
)
396 + 2 * sizeof(_Rep
) - 1);
398 const size_type __adj_size
= __size
+ __malloc_header_size
;
399 if (__adj_size
> __pagesize
&& __capacity
> __old_capacity
)
401 const size_type __extra
= __pagesize
- __adj_size
% __pagesize
;
402 __capacity
+= __extra
/ sizeof(_CharT
);
403 // Never allocate a string bigger than _S_max_size.
404 if (__capacity
> size_type(_S_max_size
))
405 __capacity
= size_type(_S_max_size
);
406 __size
= (__capacity
+ 1) * sizeof(_CharT
) + 2 * sizeof(_Rep
) - 1;
409 // NB: Might throw, but no worries about a leak, mate: _Rep()
411 _Rep
* __place
= _Rep_alloc_type(__alloc
).allocate(__size
/ sizeof(_Rep
));
412 _Rep
* __p
= new (__place
) _Rep
;
413 __p
->_M_info
._M_capacity
= __capacity
;
417 template<typename _CharT
, typename _Traits
, typename _Alloc
>
419 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
420 _M_destroy(const _Alloc
& __a
) throw ()
422 const size_type __size
= ((_M_info
._M_capacity
+ 1) * sizeof(_CharT
)
423 + 2 * sizeof(_Rep
) - 1);
424 _Rep_alloc_type(__a
).deallocate(this, __size
/ sizeof(_Rep
));
427 template<typename _CharT
, typename _Traits
, typename _Alloc
>
429 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::_Rep::
430 _M_clone(const _Alloc
& __alloc
, size_type __res
)
432 // Requested capacity of the clone.
433 const size_type __requested_cap
= _M_info
._M_length
+ __res
;
434 _Rep
* __r
= _Rep::_S_create(__requested_cap
, _M_info
._M_capacity
,
437 if (_M_info
._M_length
)
438 _S_copy(__r
->_M_refdata(), _M_refdata(), _M_info
._M_length
);
440 __r
->_M_set_length(_M_info
._M_length
);
441 return __r
->_M_refdata();
444 template<typename _CharT
, typename _Traits
, typename _Alloc
>
445 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
446 __rc_string_base(const _Alloc
& __a
)
447 : _M_dataplus(__a
, _S_construct(size_type(), _CharT(), __a
)) { }
449 template<typename _CharT
, typename _Traits
, typename _Alloc
>
450 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
451 __rc_string_base(const __rc_string_base
& __rcs
)
452 : _M_dataplus(__rcs
._M_get_allocator(),
453 __rcs
._M_grab(__rcs
._M_get_allocator())) { }
455 template<typename _CharT
, typename _Traits
, typename _Alloc
>
456 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
457 __rc_string_base(size_type __n
, _CharT __c
, const _Alloc
& __a
)
458 : _M_dataplus(__a
, _S_construct(__n
, __c
, __a
)) { }
460 template<typename _CharT
, typename _Traits
, typename _Alloc
>
461 template<typename _InputIterator
>
462 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
463 __rc_string_base(_InputIterator __beg
, _InputIterator __end
,
465 : _M_dataplus(__a
, _S_construct(__beg
, __end
, __a
)) { }
467 template<typename _CharT
, typename _Traits
, typename _Alloc
>
469 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
477 // NB: This is the special case for Input Iterators, used in
478 // istreambuf_iterators, etc.
479 // Input Iterators have a cost structure very different from
480 // pointers, calling for a different coding style.
481 template<typename _CharT
, typename _Traits
, typename _Alloc
>
482 template<typename _InIterator
>
484 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
485 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
486 std::input_iterator_tag
)
488 if (__beg
== __end
&& __a
== _Alloc())
489 return _S_empty_rep
._M_refcopy();
491 // Avoid reallocation for common case.
494 while (__beg
!= __end
&& __len
< sizeof(__buf
) / sizeof(_CharT
))
496 __buf
[__len
++] = *__beg
;
499 _Rep
* __r
= _Rep::_S_create(__len
, size_type(0), __a
);
500 _S_copy(__r
->_M_refdata(), __buf
, __len
);
503 while (__beg
!= __end
)
505 if (__len
== __r
->_M_info
._M_capacity
)
507 // Allocate more space.
508 _Rep
* __another
= _Rep::_S_create(__len
+ 1, __len
, __a
);
509 _S_copy(__another
->_M_refdata(), __r
->_M_refdata(), __len
);
510 __r
->_M_destroy(__a
);
513 __r
->_M_refdata()[__len
++] = *__beg
;
519 __r
->_M_destroy(__a
);
520 __throw_exception_again
;
522 __r
->_M_set_length(__len
);
523 return __r
->_M_refdata();
526 template<typename _CharT
, typename _Traits
, typename _Alloc
>
527 template<typename _InIterator
>
529 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
530 _S_construct(_InIterator __beg
, _InIterator __end
, const _Alloc
& __a
,
531 std::forward_iterator_tag
)
533 if (__beg
== __end
&& __a
== _Alloc())
534 return _S_empty_rep
._M_refcopy();
536 // NB: Not required, but considered best practice.
537 if (__builtin_expect(_S_is_null_pointer(__beg
) && __beg
!= __end
, 0))
538 std::__throw_logic_error(__N("__rc_string_base::"
539 "_S_construct NULL not valid"));
541 const size_type __dnew
= static_cast<size_type
>(std::distance(__beg
,
543 // Check for out_of_range and length_error exceptions.
544 _Rep
* __r
= _Rep::_S_create(__dnew
, size_type(0), __a
);
546 { _S_copy_chars(__r
->_M_refdata(), __beg
, __end
); }
549 __r
->_M_destroy(__a
);
550 __throw_exception_again
;
552 __r
->_M_set_length(__dnew
);
553 return __r
->_M_refdata();
556 template<typename _CharT
, typename _Traits
, typename _Alloc
>
558 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
559 _S_construct(size_type __n
, _CharT __c
, const _Alloc
& __a
)
561 if (__n
== 0 && __a
== _Alloc())
562 return _S_empty_rep
._M_refcopy();
564 // Check for out_of_range and length_error exceptions.
565 _Rep
* __r
= _Rep::_S_create(__n
, size_type(0), __a
);
567 _S_assign(__r
->_M_refdata(), __n
, __c
);
569 __r
->_M_set_length(__n
);
570 return __r
->_M_refdata();
573 template<typename _CharT
, typename _Traits
, typename _Alloc
>
575 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
576 _M_swap(__rc_string_base
& __rcs
)
580 if (__rcs
._M_is_leaked())
581 __rcs
._M_set_sharable();
583 _CharT
* __tmp
= _M_data();
584 _M_data(__rcs
._M_data());
585 __rcs
._M_data(__tmp
);
587 // _GLIBCXX_RESOLVE_LIB_DEFECTS
588 // 431. Swapping containers with unequal allocators.
589 std::__alloc_swap
<allocator_type
>::_S_do_it(_M_get_allocator(),
590 __rcs
._M_get_allocator());
593 template<typename _CharT
, typename _Traits
, typename _Alloc
>
595 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
596 _M_assign(const __rc_string_base
& __rcs
)
598 if (_M_rep() != __rcs
._M_rep())
600 _CharT
* __tmp
= __rcs
._M_grab(_M_get_allocator());
606 template<typename _CharT
, typename _Traits
, typename _Alloc
>
608 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
609 _M_reserve(size_type __res
)
611 // Make sure we don't shrink below the current size.
612 if (__res
< _M_length())
615 if (__res
!= _M_capacity() || _M_is_shared())
617 _CharT
* __tmp
= _M_rep()->_M_clone(_M_get_allocator(),
618 __res
- _M_length());
624 template<typename _CharT
, typename _Traits
, typename _Alloc
>
626 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
627 _M_mutate(size_type __pos
, size_type __len1
, const _CharT
* __s
,
630 const size_type __how_much
= _M_length() - __pos
- __len1
;
632 _Rep
* __r
= _Rep::_S_create(_M_length() + __len2
- __len1
,
633 _M_capacity(), _M_get_allocator());
636 _S_copy(__r
->_M_refdata(), _M_data(), __pos
);
638 _S_copy(__r
->_M_refdata() + __pos
, __s
, __len2
);
640 _S_copy(__r
->_M_refdata() + __pos
+ __len2
,
641 _M_data() + __pos
+ __len1
, __how_much
);
644 _M_data(__r
->_M_refdata());
647 template<typename _CharT
, typename _Traits
, typename _Alloc
>
649 __rc_string_base
<_CharT
, _Traits
, _Alloc
>::
650 _M_erase(size_type __pos
, size_type __n
)
652 const size_type __new_size
= _M_length() - __n
;
653 const size_type __how_much
= _M_length() - __pos
- __n
;
658 _Rep
* __r
= _Rep::_S_create(__new_size
, _M_capacity(),
662 _S_copy(__r
->_M_refdata(), _M_data(), __pos
);
664 _S_copy(__r
->_M_refdata() + __pos
,
665 _M_data() + __pos
+ __n
, __how_much
);
668 _M_data(__r
->_M_refdata());
670 else if (__how_much
&& __n
)
673 _S_move(_M_data() + __pos
,
674 _M_data() + __pos
+ __n
, __how_much
);
677 _M_rep()->_M_set_length(__new_size
);
682 __rc_string_base
<char, std::char_traits
<char>,
683 std::allocator
<char> >::
684 _M_compare(const __rc_string_base
& __rcs
) const
686 if (_M_rep() == __rcs
._M_rep())
693 __rc_string_base
<wchar_t, std::char_traits
<wchar_t>,
694 std::allocator
<wchar_t> >::
695 _M_compare(const __rc_string_base
& __rcs
) const
697 if (_M_rep() == __rcs
._M_rep())
702 _GLIBCXX_END_NAMESPACE
704 #endif /* _RC_STRING_BASE_H */