1 // Reference-counted versatile string base -*- C++ -*-
3 // Copyright (C) 2005, 2006, 2007, 2008, 2009 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 3, 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 // Under Section 7 of GPL version 3, you are granted additional
17 // permissions described in the GCC Runtime Library Exception, version
18 // 3.1, as published by the Free Software Foundation.
20 // You should have received a copy of the GNU General Public License and
21 // a copy of the GCC Runtime Library Exception along with this program;
22 // see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
23 // <http://www.gnu.org/licenses/>.
25 /** @file ext/rc_string_base.h
26 * This file is a GNU extension to the Standard C++ Library.
27 * This is an internal header file, included by other library headers.
28 * You should not attempt to use it directly.
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 "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 "empty string" _Rep object already "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 if (__exchange_and_add_dispatch(&_M_rep()->_M_info._M_refcount,
203 _M_rep()->_M_destroy(_M_get_allocator());
208 { return _M_rep()->_M_info._M_refcount < 0; }
212 { _M_rep()->_M_info._M_refcount = 0; }
217 // _S_construct_aux is used to implement the 21.3.1 para 15 which
218 // requires special behaviour if _InIterator is an integral type
219 template<typename _InIterator>
221 _S_construct_aux(_InIterator __beg, _InIterator __end,
222 const _Alloc& __a, std::__false_type)
224 typedef typename iterator_traits<_InIterator>::iterator_category _Tag;
225 return _S_construct(__beg, __end, __a, _Tag());
228 // _GLIBCXX_RESOLVE_LIB_DEFECTS
229 // 438. Ambiguity in the "do the right thing" clause
230 template<typename _Integer>
232 _S_construct_aux(_Integer __beg, _Integer __end,
233 const _Alloc& __a, std::__true_type)
234 { return _S_construct_aux_2(static_cast<size_type>(__beg),
238 _S_construct_aux_2(size_type __req, _CharT __c, const _Alloc& __a)
239 { return _S_construct(__req, __c, __a); }
241 template<typename _InIterator>
243 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a)
245 typedef typename std::__is_integer<_InIterator>::__type _Integral;
246 return _S_construct_aux(__beg, __end, __a, _Integral());
249 // For Input Iterators, used in istreambuf_iterators, etc.
250 template<typename _InIterator>
252 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
253 std::input_iterator_tag);
255 // For forward_iterators up to random_access_iterators, used for
256 // string::iterator, _CharT*, etc.
257 template<typename _FwdIterator>
259 _S_construct(_FwdIterator __beg, _FwdIterator __end, const _Alloc& __a,
260 std::forward_iterator_tag);
263 _S_construct(size_type __req, _CharT __c, const _Alloc& __a);
268 { return size_type(_S_max_size); }
272 { return _M_dataplus._M_p; }
276 { return _M_rep()->_M_info._M_length; }
280 { return _M_rep()->_M_info._M_capacity; }
284 { return _M_rep()->_M_info._M_refcount > 0; }
288 { _M_rep()->_M_info._M_refcount = -1; }
291 _M_leak() // for use in begin() & non-const op[]
298 _M_set_length(size_type __n)
299 { _M_rep()->_M_set_length(__n); }
302 : _M_dataplus(_S_empty_rep._M_refcopy()) { }
304 __rc_string_base(const _Alloc& __a);
306 __rc_string_base(const __rc_string_base& __rcs);
308 #ifdef __GXX_EXPERIMENTAL_CXX0X__
309 __rc_string_base(__rc_string_base&& __rcs)
310 : _M_dataplus(__rcs._M_get_allocator(), __rcs._M_data())
311 { __rcs._M_data(_S_empty_rep._M_refcopy()); }
314 __rc_string_base(size_type __n, _CharT __c, const _Alloc& __a);
316 template<typename _InputIterator>
317 __rc_string_base(_InputIterator __beg, _InputIterator __end,
325 { return _M_dataplus; }
327 const allocator_type&
328 _M_get_allocator() const
329 { return _M_dataplus; }
332 _M_swap(__rc_string_base& __rcs);
335 _M_assign(const __rc_string_base& __rcs);
338 _M_reserve(size_type __res);
341 _M_mutate(size_type __pos, size_type __len1, const _CharT* __s,
345 _M_erase(size_type __pos, size_type __n);
349 { _M_erase(size_type(0), _M_length()); }
352 _M_compare(const __rc_string_base&) const
356 template<typename _CharT, typename _Traits, typename _Alloc>
357 typename __rc_string_base<_CharT, _Traits, _Alloc>::_Rep_empty
358 __rc_string_base<_CharT, _Traits, _Alloc>::_S_empty_rep;
360 template<typename _CharT, typename _Traits, typename _Alloc>
361 typename __rc_string_base<_CharT, _Traits, _Alloc>::_Rep*
362 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
363 _S_create(size_type __capacity, size_type __old_capacity,
364 const _Alloc& __alloc)
366 // _GLIBCXX_RESOLVE_LIB_DEFECTS
367 // 83. String::npos vs. string::max_size()
368 if (__capacity > size_type(_S_max_size))
369 std::__throw_length_error(__N("__rc_string_base::_Rep::_S_create"));
371 // The standard places no restriction on allocating more memory
372 // than is strictly needed within this layer at the moment or as
373 // requested by an explicit application call to reserve().
375 // Many malloc implementations perform quite poorly when an
376 // application attempts to allocate memory in a stepwise fashion
377 // growing each allocation size by only 1 char. Additionally,
378 // it makes little sense to allocate less linear memory than the
379 // natural blocking size of the malloc implementation.
380 // Unfortunately, we would need a somewhat low-level calculation
381 // with tuned parameters to get this perfect for any particular
382 // malloc implementation. Fortunately, generalizations about
383 // common features seen among implementations seems to suffice.
385 // __pagesize need not match the actual VM page size for good
386 // results in practice, thus we pick a common value on the low
387 // side. __malloc_header_size is an estimate of the amount of
388 // overhead per memory allocation (in practice seen N * sizeof
389 // (void*) where N is 0, 2 or 4). According to folklore,
390 // picking this value on the high side is better than
391 // low-balling it (especially when this algorithm is used with
392 // malloc implementations that allocate memory blocks rounded up
393 // to a size which is a power of 2).
394 const size_type __pagesize = 4096;
395 const size_type __malloc_header_size = 4 * sizeof(void*);
397 // The below implements an exponential growth policy, necessary to
398 // meet amortized linear time requirements of the library: see
399 // http://gcc.gnu.org/ml/libstdc++/2001-07/msg00085.html.
400 if (__capacity > __old_capacity && __capacity < 2 * __old_capacity)
402 __capacity = 2 * __old_capacity;
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);
408 // NB: Need an array of char_type[__capacity], plus a terminating
409 // null char_type() element, plus enough for the _Rep data structure,
410 // plus sizeof(_Rep) - 1 to upper round to a size multiple of
412 // Whew. Seemingly so needy, yet so elemental.
413 size_type __size = ((__capacity + 1) * sizeof(_CharT)
414 + 2 * sizeof(_Rep) - 1);
416 const size_type __adj_size = __size + __malloc_header_size;
417 if (__adj_size > __pagesize && __capacity > __old_capacity)
419 const size_type __extra = __pagesize - __adj_size % __pagesize;
420 __capacity += __extra / sizeof(_CharT);
421 if (__capacity > size_type(_S_max_size))
422 __capacity = size_type(_S_max_size);
423 __size = (__capacity + 1) * sizeof(_CharT) + 2 * sizeof(_Rep) - 1;
426 // NB: Might throw, but no worries about a leak, mate: _Rep()
428 _Rep* __place = _Rep_alloc_type(__alloc).allocate(__size / sizeof(_Rep));
429 _Rep* __p = new (__place) _Rep;
430 __p->_M_info._M_capacity = __capacity;
434 template<typename _CharT, typename _Traits, typename _Alloc>
436 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
437 _M_destroy(const _Alloc& __a) throw ()
439 const size_type __size = ((_M_info._M_capacity + 1) * sizeof(_CharT)
440 + 2 * sizeof(_Rep) - 1);
441 _Rep_alloc_type(__a).deallocate(this, __size / sizeof(_Rep));
444 template<typename _CharT, typename _Traits, typename _Alloc>
446 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
447 _M_clone(const _Alloc& __alloc, size_type __res)
449 // Requested capacity of the clone.
450 const size_type __requested_cap = _M_info._M_length + __res;
451 _Rep* __r = _Rep::_S_create(__requested_cap, _M_info._M_capacity,
454 if (_M_info._M_length)
455 _S_copy(__r->_M_refdata(), _M_refdata(), _M_info._M_length);
457 __r->_M_set_length(_M_info._M_length);
458 return __r->_M_refdata();
461 template<typename _CharT, typename _Traits, typename _Alloc>
462 __rc_string_base<_CharT, _Traits, _Alloc>::
463 __rc_string_base(const _Alloc& __a)
464 : _M_dataplus(__a, _S_construct(size_type(), _CharT(), __a)) { }
466 template<typename _CharT, typename _Traits, typename _Alloc>
467 __rc_string_base<_CharT, _Traits, _Alloc>::
468 __rc_string_base(const __rc_string_base& __rcs)
469 : _M_dataplus(__rcs._M_get_allocator(),
470 __rcs._M_grab(__rcs._M_get_allocator())) { }
472 template<typename _CharT, typename _Traits, typename _Alloc>
473 __rc_string_base<_CharT, _Traits, _Alloc>::
474 __rc_string_base(size_type __n, _CharT __c, const _Alloc& __a)
475 : _M_dataplus(__a, _S_construct(__n, __c, __a)) { }
477 template<typename _CharT, typename _Traits, typename _Alloc>
478 template<typename _InputIterator>
479 __rc_string_base<_CharT, _Traits, _Alloc>::
480 __rc_string_base(_InputIterator __beg, _InputIterator __end,
482 : _M_dataplus(__a, _S_construct(__beg, __end, __a)) { }
484 template<typename _CharT, typename _Traits, typename _Alloc>
486 __rc_string_base<_CharT, _Traits, _Alloc>::
494 // NB: This is the special case for Input Iterators, used in
495 // istreambuf_iterators, etc.
496 // Input Iterators have a cost structure very different from
497 // pointers, calling for a different coding style.
498 template<typename _CharT, typename _Traits, typename _Alloc>
499 template<typename _InIterator>
501 __rc_string_base<_CharT, _Traits, _Alloc>::
502 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
503 std::input_iterator_tag)
505 if (__beg == __end && __a == _Alloc())
506 return _S_empty_rep._M_refcopy();
508 // Avoid reallocation for common case.
511 while (__beg != __end && __len < sizeof(__buf) / sizeof(_CharT))
513 __buf[__len++] = *__beg;
516 _Rep* __r = _Rep::_S_create(__len, size_type(0), __a);
517 _S_copy(__r->_M_refdata(), __buf, __len);
520 while (__beg != __end)
522 if (__len == __r->_M_info._M_capacity)
524 // Allocate more space.
525 _Rep* __another = _Rep::_S_create(__len + 1, __len, __a);
526 _S_copy(__another->_M_refdata(), __r->_M_refdata(), __len);
527 __r->_M_destroy(__a);
530 __r->_M_refdata()[__len++] = *__beg;
536 __r->_M_destroy(__a);
537 __throw_exception_again;
539 __r->_M_set_length(__len);
540 return __r->_M_refdata();
543 template<typename _CharT, typename _Traits, typename _Alloc>
544 template<typename _InIterator>
546 __rc_string_base<_CharT, _Traits, _Alloc>::
547 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
548 std::forward_iterator_tag)
550 if (__beg == __end && __a == _Alloc())
551 return _S_empty_rep._M_refcopy();
553 // NB: Not required, but considered best practice.
554 if (__is_null_pointer(__beg) && __beg != __end)
555 std::__throw_logic_error(__N("__rc_string_base::"
556 "_S_construct NULL not valid"));
558 const size_type __dnew = static_cast<size_type>(std::distance(__beg,
560 // Check for out_of_range and length_error exceptions.
561 _Rep* __r = _Rep::_S_create(__dnew, size_type(0), __a);
563 { _S_copy_chars(__r->_M_refdata(), __beg, __end); }
566 __r->_M_destroy(__a);
567 __throw_exception_again;
569 __r->_M_set_length(__dnew);
570 return __r->_M_refdata();
573 template<typename _CharT, typename _Traits, typename _Alloc>
575 __rc_string_base<_CharT, _Traits, _Alloc>::
576 _S_construct(size_type __n, _CharT __c, const _Alloc& __a)
578 if (__n == 0 && __a == _Alloc())
579 return _S_empty_rep._M_refcopy();
581 // Check for out_of_range and length_error exceptions.
582 _Rep* __r = _Rep::_S_create(__n, size_type(0), __a);
584 _S_assign(__r->_M_refdata(), __n, __c);
586 __r->_M_set_length(__n);
587 return __r->_M_refdata();
590 template<typename _CharT, typename _Traits, typename _Alloc>
592 __rc_string_base<_CharT, _Traits, _Alloc>::
593 _M_swap(__rc_string_base& __rcs)
597 if (__rcs._M_is_leaked())
598 __rcs._M_set_sharable();
600 _CharT* __tmp = _M_data();
601 _M_data(__rcs._M_data());
602 __rcs._M_data(__tmp);
604 // _GLIBCXX_RESOLVE_LIB_DEFECTS
605 // 431. Swapping containers with unequal allocators.
606 std::__alloc_swap<allocator_type>::_S_do_it(_M_get_allocator(),
607 __rcs._M_get_allocator());
610 template<typename _CharT, typename _Traits, typename _Alloc>
612 __rc_string_base<_CharT, _Traits, _Alloc>::
613 _M_assign(const __rc_string_base& __rcs)
615 if (_M_rep() != __rcs._M_rep())
617 _CharT* __tmp = __rcs._M_grab(_M_get_allocator());
623 template<typename _CharT, typename _Traits, typename _Alloc>
625 __rc_string_base<_CharT, _Traits, _Alloc>::
626 _M_reserve(size_type __res)
628 // Make sure we don't shrink below the current size.
629 if (__res < _M_length())
632 if (__res != _M_capacity() || _M_is_shared())
634 _CharT* __tmp = _M_rep()->_M_clone(_M_get_allocator(),
635 __res - _M_length());
641 template<typename _CharT, typename _Traits, typename _Alloc>
643 __rc_string_base<_CharT, _Traits, _Alloc>::
644 _M_mutate(size_type __pos, size_type __len1, const _CharT* __s,
647 const size_type __how_much = _M_length() - __pos - __len1;
649 _Rep* __r = _Rep::_S_create(_M_length() + __len2 - __len1,
650 _M_capacity(), _M_get_allocator());
653 _S_copy(__r->_M_refdata(), _M_data(), __pos);
655 _S_copy(__r->_M_refdata() + __pos, __s, __len2);
657 _S_copy(__r->_M_refdata() + __pos + __len2,
658 _M_data() + __pos + __len1, __how_much);
661 _M_data(__r->_M_refdata());
664 template<typename _CharT, typename _Traits, typename _Alloc>
666 __rc_string_base<_CharT, _Traits, _Alloc>::
667 _M_erase(size_type __pos, size_type __n)
669 const size_type __new_size = _M_length() - __n;
670 const size_type __how_much = _M_length() - __pos - __n;
675 _Rep* __r = _Rep::_S_create(__new_size, _M_capacity(),
679 _S_copy(__r->_M_refdata(), _M_data(), __pos);
681 _S_copy(__r->_M_refdata() + __pos,
682 _M_data() + __pos + __n, __how_much);
685 _M_data(__r->_M_refdata());
687 else if (__how_much && __n)
690 _S_move(_M_data() + __pos,
691 _M_data() + __pos + __n, __how_much);
694 _M_rep()->_M_set_length(__new_size);
699 __rc_string_base<char, std::char_traits<char>,
700 std::allocator<char> >::
701 _M_compare(const __rc_string_base& __rcs) const
703 if (_M_rep() == __rcs._M_rep())
708 #ifdef _GLIBCXX_USE_WCHAR_T
711 __rc_string_base<wchar_t, std::char_traits<wchar_t>,
712 std::allocator<wchar_t> >::
713 _M_compare(const __rc_string_base& __rcs) const
715 if (_M_rep() == __rcs._M_rep())
721 _GLIBCXX_END_NAMESPACE
723 #endif /* _RC_STRING_BASE_H */