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(static_cast<size_type>(__beg), __end, __a); }
236 template<typename _InIterator>
238 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a)
240 typedef typename std::__is_integer<_InIterator>::__type _Integral;
241 return _S_construct_aux(__beg, __end, __a, _Integral());
244 // For Input Iterators, used in istreambuf_iterators, etc.
245 template<typename _InIterator>
247 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
248 std::input_iterator_tag);
250 // For forward_iterators up to random_access_iterators, used for
251 // string::iterator, _CharT*, etc.
252 template<typename _FwdIterator>
254 _S_construct(_FwdIterator __beg, _FwdIterator __end, const _Alloc& __a,
255 std::forward_iterator_tag);
258 _S_construct(size_type __req, _CharT __c, const _Alloc& __a);
263 { return size_type(_S_max_size); }
267 { return _M_dataplus._M_p; }
271 { return _M_rep()->_M_info._M_length; }
275 { return _M_rep()->_M_info._M_capacity; }
279 { return _M_rep()->_M_info._M_refcount > 0; }
283 { _M_rep()->_M_info._M_refcount = -1; }
286 _M_leak() // for use in begin() & non-const op[]
293 _M_set_length(size_type __n)
294 { _M_rep()->_M_set_length(__n); }
297 : _M_dataplus(_S_empty_rep._M_refcopy()) { }
299 __rc_string_base(const _Alloc& __a);
301 __rc_string_base(const __rc_string_base& __rcs);
303 #ifdef __GXX_EXPERIMENTAL_CXX0X__
304 __rc_string_base(__rc_string_base&& __rcs)
305 : _M_dataplus(__rcs._M_get_allocator(), __rcs._M_data())
306 { __rcs._M_data(_S_empty_rep._M_refcopy()); }
309 __rc_string_base(size_type __n, _CharT __c, const _Alloc& __a);
311 template<typename _InputIterator>
312 __rc_string_base(_InputIterator __beg, _InputIterator __end,
320 { return _M_dataplus; }
322 const allocator_type&
323 _M_get_allocator() const
324 { return _M_dataplus; }
327 _M_swap(__rc_string_base& __rcs);
330 _M_assign(const __rc_string_base& __rcs);
333 _M_reserve(size_type __res);
336 _M_mutate(size_type __pos, size_type __len1, const _CharT* __s,
340 _M_erase(size_type __pos, size_type __n);
344 { _M_erase(size_type(0), _M_length()); }
347 _M_compare(const __rc_string_base&) const
351 template<typename _CharT, typename _Traits, typename _Alloc>
352 typename __rc_string_base<_CharT, _Traits, _Alloc>::_Rep_empty
353 __rc_string_base<_CharT, _Traits, _Alloc>::_S_empty_rep;
355 template<typename _CharT, typename _Traits, typename _Alloc>
356 typename __rc_string_base<_CharT, _Traits, _Alloc>::_Rep*
357 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
358 _S_create(size_type __capacity, size_type __old_capacity,
359 const _Alloc& __alloc)
361 // _GLIBCXX_RESOLVE_LIB_DEFECTS
362 // 83. String::npos vs. string::max_size()
363 if (__capacity > size_type(_S_max_size))
364 std::__throw_length_error(__N("__rc_string_base::_Rep::_S_create"));
366 // The standard places no restriction on allocating more memory
367 // than is strictly needed within this layer at the moment or as
368 // requested by an explicit application call to reserve().
370 // Many malloc implementations perform quite poorly when an
371 // application attempts to allocate memory in a stepwise fashion
372 // growing each allocation size by only 1 char. Additionally,
373 // it makes little sense to allocate less linear memory than the
374 // natural blocking size of the malloc implementation.
375 // Unfortunately, we would need a somewhat low-level calculation
376 // with tuned parameters to get this perfect for any particular
377 // malloc implementation. Fortunately, generalizations about
378 // common features seen among implementations seems to suffice.
380 // __pagesize need not match the actual VM page size for good
381 // results in practice, thus we pick a common value on the low
382 // side. __malloc_header_size is an estimate of the amount of
383 // overhead per memory allocation (in practice seen N * sizeof
384 // (void*) where N is 0, 2 or 4). According to folklore,
385 // picking this value on the high side is better than
386 // low-balling it (especially when this algorithm is used with
387 // malloc implementations that allocate memory blocks rounded up
388 // to a size which is a power of 2).
389 const size_type __pagesize = 4096;
390 const size_type __malloc_header_size = 4 * sizeof(void*);
392 // The below implements an exponential growth policy, necessary to
393 // meet amortized linear time requirements of the library: see
394 // http://gcc.gnu.org/ml/libstdc++/2001-07/msg00085.html.
395 if (__capacity > __old_capacity && __capacity < 2 * __old_capacity)
397 __capacity = 2 * __old_capacity;
398 // Never allocate a string bigger than _S_max_size.
399 if (__capacity > size_type(_S_max_size))
400 __capacity = size_type(_S_max_size);
403 // NB: Need an array of char_type[__capacity], plus a terminating
404 // null char_type() element, plus enough for the _Rep data structure,
405 // plus sizeof(_Rep) - 1 to upper round to a size multiple of
407 // Whew. Seemingly so needy, yet so elemental.
408 size_type __size = ((__capacity + 1) * sizeof(_CharT)
409 + 2 * sizeof(_Rep) - 1);
411 const size_type __adj_size = __size + __malloc_header_size;
412 if (__adj_size > __pagesize && __capacity > __old_capacity)
414 const size_type __extra = __pagesize - __adj_size % __pagesize;
415 __capacity += __extra / sizeof(_CharT);
416 if (__capacity > size_type(_S_max_size))
417 __capacity = size_type(_S_max_size);
418 __size = (__capacity + 1) * sizeof(_CharT) + 2 * sizeof(_Rep) - 1;
421 // NB: Might throw, but no worries about a leak, mate: _Rep()
423 _Rep* __place = _Rep_alloc_type(__alloc).allocate(__size / sizeof(_Rep));
424 _Rep* __p = new (__place) _Rep;
425 __p->_M_info._M_capacity = __capacity;
429 template<typename _CharT, typename _Traits, typename _Alloc>
431 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
432 _M_destroy(const _Alloc& __a) throw ()
434 const size_type __size = ((_M_info._M_capacity + 1) * sizeof(_CharT)
435 + 2 * sizeof(_Rep) - 1);
436 _Rep_alloc_type(__a).deallocate(this, __size / sizeof(_Rep));
439 template<typename _CharT, typename _Traits, typename _Alloc>
441 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
442 _M_clone(const _Alloc& __alloc, size_type __res)
444 // Requested capacity of the clone.
445 const size_type __requested_cap = _M_info._M_length + __res;
446 _Rep* __r = _Rep::_S_create(__requested_cap, _M_info._M_capacity,
449 if (_M_info._M_length)
450 _S_copy(__r->_M_refdata(), _M_refdata(), _M_info._M_length);
452 __r->_M_set_length(_M_info._M_length);
453 return __r->_M_refdata();
456 template<typename _CharT, typename _Traits, typename _Alloc>
457 __rc_string_base<_CharT, _Traits, _Alloc>::
458 __rc_string_base(const _Alloc& __a)
459 : _M_dataplus(__a, _S_construct(size_type(), _CharT(), __a)) { }
461 template<typename _CharT, typename _Traits, typename _Alloc>
462 __rc_string_base<_CharT, _Traits, _Alloc>::
463 __rc_string_base(const __rc_string_base& __rcs)
464 : _M_dataplus(__rcs._M_get_allocator(),
465 __rcs._M_grab(__rcs._M_get_allocator())) { }
467 template<typename _CharT, typename _Traits, typename _Alloc>
468 __rc_string_base<_CharT, _Traits, _Alloc>::
469 __rc_string_base(size_type __n, _CharT __c, const _Alloc& __a)
470 : _M_dataplus(__a, _S_construct(__n, __c, __a)) { }
472 template<typename _CharT, typename _Traits, typename _Alloc>
473 template<typename _InputIterator>
474 __rc_string_base<_CharT, _Traits, _Alloc>::
475 __rc_string_base(_InputIterator __beg, _InputIterator __end,
477 : _M_dataplus(__a, _S_construct(__beg, __end, __a)) { }
479 template<typename _CharT, typename _Traits, typename _Alloc>
481 __rc_string_base<_CharT, _Traits, _Alloc>::
489 // NB: This is the special case for Input Iterators, used in
490 // istreambuf_iterators, etc.
491 // Input Iterators have a cost structure very different from
492 // pointers, calling for a different coding style.
493 template<typename _CharT, typename _Traits, typename _Alloc>
494 template<typename _InIterator>
496 __rc_string_base<_CharT, _Traits, _Alloc>::
497 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
498 std::input_iterator_tag)
500 if (__beg == __end && __a == _Alloc())
501 return _S_empty_rep._M_refcopy();
503 // Avoid reallocation for common case.
506 while (__beg != __end && __len < sizeof(__buf) / sizeof(_CharT))
508 __buf[__len++] = *__beg;
511 _Rep* __r = _Rep::_S_create(__len, size_type(0), __a);
512 _S_copy(__r->_M_refdata(), __buf, __len);
515 while (__beg != __end)
517 if (__len == __r->_M_info._M_capacity)
519 // Allocate more space.
520 _Rep* __another = _Rep::_S_create(__len + 1, __len, __a);
521 _S_copy(__another->_M_refdata(), __r->_M_refdata(), __len);
522 __r->_M_destroy(__a);
525 __r->_M_refdata()[__len++] = *__beg;
531 __r->_M_destroy(__a);
532 __throw_exception_again;
534 __r->_M_set_length(__len);
535 return __r->_M_refdata();
538 template<typename _CharT, typename _Traits, typename _Alloc>
539 template<typename _InIterator>
541 __rc_string_base<_CharT, _Traits, _Alloc>::
542 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
543 std::forward_iterator_tag)
545 if (__beg == __end && __a == _Alloc())
546 return _S_empty_rep._M_refcopy();
548 // NB: Not required, but considered best practice.
549 if (__builtin_expect(__is_null_pointer(__beg) && __beg != __end, 0))
550 std::__throw_logic_error(__N("__rc_string_base::"
551 "_S_construct NULL not valid"));
553 const size_type __dnew = static_cast<size_type>(std::distance(__beg,
555 // Check for out_of_range and length_error exceptions.
556 _Rep* __r = _Rep::_S_create(__dnew, size_type(0), __a);
558 { _S_copy_chars(__r->_M_refdata(), __beg, __end); }
561 __r->_M_destroy(__a);
562 __throw_exception_again;
564 __r->_M_set_length(__dnew);
565 return __r->_M_refdata();
568 template<typename _CharT, typename _Traits, typename _Alloc>
570 __rc_string_base<_CharT, _Traits, _Alloc>::
571 _S_construct(size_type __n, _CharT __c, const _Alloc& __a)
573 if (__n == 0 && __a == _Alloc())
574 return _S_empty_rep._M_refcopy();
576 // Check for out_of_range and length_error exceptions.
577 _Rep* __r = _Rep::_S_create(__n, size_type(0), __a);
579 _S_assign(__r->_M_refdata(), __n, __c);
581 __r->_M_set_length(__n);
582 return __r->_M_refdata();
585 template<typename _CharT, typename _Traits, typename _Alloc>
587 __rc_string_base<_CharT, _Traits, _Alloc>::
588 _M_swap(__rc_string_base& __rcs)
592 if (__rcs._M_is_leaked())
593 __rcs._M_set_sharable();
595 _CharT* __tmp = _M_data();
596 _M_data(__rcs._M_data());
597 __rcs._M_data(__tmp);
599 // _GLIBCXX_RESOLVE_LIB_DEFECTS
600 // 431. Swapping containers with unequal allocators.
601 std::__alloc_swap<allocator_type>::_S_do_it(_M_get_allocator(),
602 __rcs._M_get_allocator());
605 template<typename _CharT, typename _Traits, typename _Alloc>
607 __rc_string_base<_CharT, _Traits, _Alloc>::
608 _M_assign(const __rc_string_base& __rcs)
610 if (_M_rep() != __rcs._M_rep())
612 _CharT* __tmp = __rcs._M_grab(_M_get_allocator());
618 template<typename _CharT, typename _Traits, typename _Alloc>
620 __rc_string_base<_CharT, _Traits, _Alloc>::
621 _M_reserve(size_type __res)
623 // Make sure we don't shrink below the current size.
624 if (__res < _M_length())
627 if (__res != _M_capacity() || _M_is_shared())
629 _CharT* __tmp = _M_rep()->_M_clone(_M_get_allocator(),
630 __res - _M_length());
636 template<typename _CharT, typename _Traits, typename _Alloc>
638 __rc_string_base<_CharT, _Traits, _Alloc>::
639 _M_mutate(size_type __pos, size_type __len1, const _CharT* __s,
642 const size_type __how_much = _M_length() - __pos - __len1;
644 _Rep* __r = _Rep::_S_create(_M_length() + __len2 - __len1,
645 _M_capacity(), _M_get_allocator());
648 _S_copy(__r->_M_refdata(), _M_data(), __pos);
650 _S_copy(__r->_M_refdata() + __pos, __s, __len2);
652 _S_copy(__r->_M_refdata() + __pos + __len2,
653 _M_data() + __pos + __len1, __how_much);
656 _M_data(__r->_M_refdata());
659 template<typename _CharT, typename _Traits, typename _Alloc>
661 __rc_string_base<_CharT, _Traits, _Alloc>::
662 _M_erase(size_type __pos, size_type __n)
664 const size_type __new_size = _M_length() - __n;
665 const size_type __how_much = _M_length() - __pos - __n;
670 _Rep* __r = _Rep::_S_create(__new_size, _M_capacity(),
674 _S_copy(__r->_M_refdata(), _M_data(), __pos);
676 _S_copy(__r->_M_refdata() + __pos,
677 _M_data() + __pos + __n, __how_much);
680 _M_data(__r->_M_refdata());
682 else if (__how_much && __n)
685 _S_move(_M_data() + __pos,
686 _M_data() + __pos + __n, __how_much);
689 _M_rep()->_M_set_length(__new_size);
694 __rc_string_base<char, std::char_traits<char>,
695 std::allocator<char> >::
696 _M_compare(const __rc_string_base& __rcs) const
698 if (_M_rep() == __rcs._M_rep())
703 #ifdef _GLIBCXX_USE_WCHAR_T
706 __rc_string_base<wchar_t, std::char_traits<wchar_t>,
707 std::allocator<wchar_t> >::
708 _M_compare(const __rc_string_base& __rcs) const
710 if (_M_rep() == __rcs._M_rep())
716 _GLIBCXX_END_NAMESPACE
718 #endif /* _RC_STRING_BASE_H */