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>
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& __alloc1, const _Alloc& __alloc2) const
198 return (!_M_is_leaked() && __alloc1 == __alloc2)
199 ? _M_rep()->_M_refcopy() : _M_rep()->_M_clone(__alloc1);
203 _M_dispose(const _Alloc& __a)
205 if (__exchange_and_add(&_M_rep()->_M_info._M_refcount, -1) <= 0)
206 _M_rep()->_M_destroy(__a);
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<class _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<class _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<class _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<class _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<class _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,
312 { _M_dispose(_M_get_allocator()); }
315 _M_get_allocator() const
316 { return _M_dataplus; }
319 _M_swap(__rc_string_base& __rcs);
322 _M_assign(const __rc_string_base& __rcs);
325 _M_reserve(size_type __res);
328 _M_mutate(size_type __pos, size_type __len1, size_type __len2);
331 template<typename _CharT, typename _Traits, typename _Alloc>
332 typename __rc_string_base<_CharT, _Traits, _Alloc>::_Rep_empty
333 __rc_string_base<_CharT, _Traits, _Alloc>::_S_empty_rep;
335 template<typename _CharT, typename _Traits, typename _Alloc>
336 typename __rc_string_base<_CharT, _Traits, _Alloc>::_Rep*
337 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
338 _S_create(size_type __capacity, size_type __old_capacity,
339 const _Alloc& __alloc)
341 // _GLIBCXX_RESOLVE_LIB_DEFECTS
342 // 83. String::npos vs. string::max_size()
343 if (__capacity > size_type(_S_max_size))
344 std::__throw_length_error(__N("__rc_string_base::_Rep::_S_create"));
346 // The standard places no restriction on allocating more memory
347 // than is strictly needed within this layer at the moment or as
348 // requested by an explicit application call to reserve().
350 // Many malloc implementations perform quite poorly when an
351 // application attempts to allocate memory in a stepwise fashion
352 // growing each allocation size by only 1 char. Additionally,
353 // it makes little sense to allocate less linear memory than the
354 // natural blocking size of the malloc implementation.
355 // Unfortunately, we would need a somewhat low-level calculation
356 // with tuned parameters to get this perfect for any particular
357 // malloc implementation. Fortunately, generalizations about
358 // common features seen among implementations seems to suffice.
360 // __pagesize need not match the actual VM page size for good
361 // results in practice, thus we pick a common value on the low
362 // side. __malloc_header_size is an estimate of the amount of
363 // overhead per memory allocation (in practice seen N * sizeof
364 // (void*) where N is 0, 2 or 4). According to folklore,
365 // picking this value on the high side is better than
366 // low-balling it (especially when this algorithm is used with
367 // malloc implementations that allocate memory blocks rounded up
368 // to a size which is a power of 2).
369 const size_type __pagesize = 4096;
370 const size_type __malloc_header_size = 4 * sizeof(void*);
372 // The below implements an exponential growth policy, necessary to
373 // meet amortized linear time requirements of the library: see
374 // http://gcc.gnu.org/ml/libstdc++/2001-07/msg00085.html.
375 // It's active for allocations requiring an amount of memory above
376 // system pagesize. This is consistent with the requirements of the
377 // standard: http://gcc.gnu.org/ml/libstdc++/2001-07/msg00130.html
378 if (__capacity > __old_capacity && __capacity < 2 * __old_capacity)
379 __capacity = 2 * __old_capacity;
381 // NB: Need an array of char_type[__capacity], plus a terminating
382 // null char_type() element, plus enough for the _Rep data structure,
383 // plus sizeof(size_type) - 1 to upper round to a size multiple
384 // of sizeof(size_type).
385 // Whew. Seemingly so needy, yet so elemental.
386 size_type __size = ((__capacity + 1) * sizeof(_CharT)
387 + 2 * sizeof(_Rep) - 1);
389 const size_type __adj_size = __size + __malloc_header_size;
390 if (__adj_size > __pagesize && __capacity > __old_capacity)
392 const size_type __extra = __pagesize - __adj_size % __pagesize;
393 __capacity += __extra / sizeof(_CharT);
394 // Never allocate a string bigger than _S_max_size.
395 if (__capacity > size_type(_S_max_size))
396 __capacity = size_type(_S_max_size);
397 __size = (__capacity + 1) * sizeof(_CharT) + 2 * sizeof(_Rep) - 1;
400 // NB: Might throw, but no worries about a leak, mate: _Rep()
402 _Rep* __place = _Rep_alloc_type(__alloc).allocate(__size / sizeof(_Rep));
403 _Rep* __p = new (__place) _Rep;
404 __p->_M_info._M_capacity = __capacity;
408 template<typename _CharT, typename _Traits, typename _Alloc>
410 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
411 _M_destroy(const _Alloc& __a) throw ()
413 const size_type __size = ((_M_info._M_capacity + 1) * sizeof(_CharT)
414 + 2 * sizeof(_Rep) - 1);
415 _Rep_alloc_type(__a).deallocate(this, __size / sizeof(_Rep));
418 template<typename _CharT, typename _Traits, typename _Alloc>
420 __rc_string_base<_CharT, _Traits, _Alloc>::_Rep::
421 _M_clone(const _Alloc& __alloc, size_type __res)
423 // Requested capacity of the clone.
424 const size_type __requested_cap = _M_info._M_length + __res;
425 _Rep* __r = _Rep::_S_create(__requested_cap, _M_info._M_capacity,
428 if (_M_info._M_length)
429 _S_copy(__r->_M_refdata(), _M_refdata(), _M_info._M_length);
431 __r->_M_set_length(_M_info._M_length);
432 return __r->_M_refdata();
435 template<typename _CharT, typename _Traits, typename _Alloc>
436 __rc_string_base<_CharT, _Traits, _Alloc>::
437 __rc_string_base(const _Alloc& __a)
438 : _M_dataplus(__a, _S_construct(size_type(), _CharT(), __a)) { }
440 template<typename _CharT, typename _Traits, typename _Alloc>
441 __rc_string_base<_CharT, _Traits, _Alloc>::
442 __rc_string_base(const __rc_string_base& __rcs)
443 : _M_dataplus(__rcs._M_get_allocator(),
444 __rcs._M_grab(_Alloc(__rcs._M_get_allocator()),
445 __rcs._M_get_allocator())) { }
447 template<typename _CharT, typename _Traits, typename _Alloc>
448 __rc_string_base<_CharT, _Traits, _Alloc>::
449 __rc_string_base(size_type __n, _CharT __c, const _Alloc& __a)
450 : _M_dataplus(__a, _S_construct(__n, __c, __a)) { }
452 template<typename _CharT, typename _Traits, typename _Alloc>
453 template<typename _InputIterator>
454 __rc_string_base<_CharT, _Traits, _Alloc>::
455 __rc_string_base(_InputIterator __beg, _InputIterator __end,
457 : _M_dataplus(__a, _S_construct(__beg, __end, __a)) { }
459 template<typename _CharT, typename _Traits, typename _Alloc>
461 __rc_string_base<_CharT, _Traits, _Alloc>::
469 // NB: This is the special case for Input Iterators, used in
470 // istreambuf_iterators, etc.
471 // Input Iterators have a cost structure very different from
472 // pointers, calling for a different coding style.
473 template<typename _CharT, typename _Traits, typename _Alloc>
474 template<typename _InIterator>
476 __rc_string_base<_CharT, _Traits, _Alloc>::
477 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
478 std::input_iterator_tag)
480 if (__beg == __end && __a == _Alloc())
481 return _S_empty_rep._M_refcopy();
483 // Avoid reallocation for common case.
486 while (__beg != __end && __len < sizeof(__buf) / sizeof(_CharT))
488 __buf[__len++] = *__beg;
491 _Rep* __r = _Rep::_S_create(__len, size_type(0), __a);
492 _S_copy(__r->_M_refdata(), __buf, __len);
495 while (__beg != __end)
497 if (__len == __r->_M_capacity)
499 // Allocate more space.
500 _Rep* __another = _Rep::_S_create(__len + 1, __len, __a);
501 _S_copy(__another->_M_refdata(), __r->_M_refdata(), __len);
502 __r->_M_destroy(__a);
505 __r->_M_refdata()[__len++] = *__beg;
511 __r->_M_destroy(__a);
512 __throw_exception_again;
514 __r->_M_set_length(__len);
515 return __r->_M_refdata();
518 template<typename _CharT, typename _Traits, typename _Alloc>
519 template <typename _InIterator>
521 __rc_string_base<_CharT, _Traits, _Alloc>::
522 _S_construct(_InIterator __beg, _InIterator __end, const _Alloc& __a,
523 std::forward_iterator_tag)
525 if (__beg == __end && __a == _Alloc())
526 return _S_empty_rep._M_refcopy();
528 // NB: Not required, but considered best practice.
529 if (__builtin_expect(_S_is_null_pointer(__beg) && __beg != __end, 0))
530 std::__throw_logic_error(__N("__rc_string_base::"
531 "_S_construct NULL not valid"));
533 const size_type __dnew = static_cast<size_type>(std::distance(__beg,
535 // Check for out_of_range and length_error exceptions.
536 _Rep* __r = _Rep::_S_create(__dnew, size_type(0), __a);
538 { _S_copy_chars(__r->_M_refdata(), __beg, __end); }
541 __r->_M_destroy(__a);
542 __throw_exception_again;
544 __r->_M_set_length(__dnew);
545 return __r->_M_refdata();
548 template<typename _CharT, typename _Traits, typename _Alloc>
550 __rc_string_base<_CharT, _Traits, _Alloc>::
551 _S_construct(size_type __n, _CharT __c, const _Alloc& __a)
553 if (__n == 0 && __a == _Alloc())
554 return _S_empty_rep._M_refcopy();
556 // Check for out_of_range and length_error exceptions.
557 _Rep* __r = _Rep::_S_create(__n, size_type(0), __a);
559 _S_assign(__r->_M_refdata(), __n, __c);
561 __r->_M_set_length(__n);
562 return __r->_M_refdata();
565 template<typename _CharT, typename _Traits, typename _Alloc>
567 __rc_string_base<_CharT, _Traits, _Alloc>::
568 _M_swap(__rc_string_base& __rcs)
572 if (__rcs._M_is_leaked())
573 __rcs._M_set_sharable();
575 _CharT* __tmp = _M_data();
576 _M_data(__rcs._M_data());
577 __rcs._M_data(__tmp);
579 // NB: Implement Option 3 of DR 431 (see N1599).
580 _M_dataplus._M_alloc_swap(__rcs._M_dataplus);
583 template<typename _CharT, typename _Traits, typename _Alloc>
585 __rc_string_base<_CharT, _Traits, _Alloc>::
586 _M_assign(const __rc_string_base& __rcs)
588 if (_M_rep() != __rcs._M_rep())
590 const allocator_type __a = _M_get_allocator();
591 _CharT* __tmp = __rcs._M_grab(__a, __rcs._M_get_allocator());
597 template<typename _CharT, typename _Traits, typename _Alloc>
599 __rc_string_base<_CharT, _Traits, _Alloc>::
600 _M_reserve(size_type __res)
602 if (__res != _M_capacity() || _M_is_shared())
604 // Make sure we don't shrink below the current size.
605 if (__res < _M_length())
608 const allocator_type __a = _M_get_allocator();
609 _CharT* __tmp = _M_rep()->_M_clone(__a, __res - _M_length());
615 template<typename _CharT, typename _Traits, typename _Alloc>
617 __rc_string_base<_CharT, _Traits, _Alloc>::
618 _M_mutate(size_type __pos, size_type __len1, size_type __len2)
620 const size_type __old_size = _M_length();
621 const size_type __new_size = __old_size + __len2 - __len1;
622 const size_type __how_much = __old_size - __pos - __len1;
624 if (__new_size > _M_capacity() || _M_is_shared())
627 const allocator_type __a = _M_get_allocator();
628 _Rep* __r = _Rep::_S_create(__new_size, _M_capacity(), __a);
631 _S_copy(__r->_M_refdata(), _M_data(), __pos);
633 _S_copy(__r->_M_refdata() + __pos + __len2,
634 _M_data() + __pos + __len1, __how_much);
637 _M_data(__r->_M_refdata());
639 else if (__how_much && __len1 != __len2)
642 _S_move(_M_data() + __pos + __len2,
643 _M_data() + __pos + __len1, __how_much);
645 _M_rep()->_M_set_length(__new_size);
647 } // namespace __gnu_cxx
649 #endif /* _RC_STRING_BASE_H */