OpenCores
URL https://opencores.org/ocsvn/openrisc/openrisc/trunk

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [libstdc++-v3/] [doc/] [xml/] [manual/] [containers.xml] - Blame information for rev 750

Go to most recent revision | Details | Compare with Previous | View Log

Line No. Rev Author Line
1 742 jeremybenn
2
         xml:id="std.containers" xreflabel="Containers">
3
4
 
5
</code></pre></td>
      </tr>
      <tr valign="middle">
         <td>6</td>
         <td></td>
         <td></td>
         <td class="code"><pre><code>  Containers</code></pre></td>
      </tr>
      <tr valign="middle">
         <td>7</td>
         <td></td>
         <td></td>
         <td class="code"><pre><code>  <indexterm><primary>Containers</primary></indexterm></code></pre></td>
      </tr>
      <tr valign="middle">
         <td>8</td>
         <td></td>
         <td></td>
         <td class="code"><pre><code>
9
  
10
    
11
      ISO C++
12
    
13
    
14
      library
15
    
16
  
17
18
 
19
 
20
 
21
22
Sequences
23
24
 
25
 
26
list
27
28
 
29
  
list::size() is O(n)
30
 
31
   
32
     Yes it is, and that's okay.  This is a decision that we preserved
33
     when we imported SGI's STL implementation.  The following is
34
     quoted from their FAQ:
35
   
36
   
37
     
38
       The size() member function, for list and slist, takes time
39
       proportional to the number of elements in the list.  This was a
40
       deliberate tradeoff.  The only way to get a constant-time
41
       size() for linked lists would be to maintain an extra member
42
       variable containing the list's size.  This would require taking
43
       extra time to update that variable (it would make splice() a
44
       linear time operation, for example), and it would also make the
45
       list larger.  Many list algorithms don't require that extra
46
       word (algorithms that do require it might do better with
47
       vectors than with lists), and, when it is necessary to maintain
48
       an explicit size count, it's something that users can do
49
       themselves.
50
     
51
     
52
       This choice is permitted by the C++ standard. The standard says
53
       that size() should be constant time, and
54
       should does not mean the same thing as
55
       shall.  This is the officially recommended ISO
56
       wording for saying that an implementation is supposed to do
57
       something unless there is a good reason not to.
58
      
59
      
60
        One implication of linear time size(): you should never write
61
      
62
         
63
         if (L.size() == 0)
64
             ...
65
         
66
 
67
         
68
         Instead, you should write
69
         
70
 
71
         
72
         if (L.empty())
73
             ...
74
         
75
   
76
  
77
78
 
79
vector
80
81
 
82
  
83
  
84
  
Space Overhead Management
85
 
86
   
87
     In this
88
     message to the list, Daniel Kostecky announced work on an
89
     alternate form of std::vector that would support
90
     hints on the number of elements to be over-allocated.  The design
91
     was also described, along with possible implementation choices.
92
   
93
   
94
     The first two alpha releases were announced here
95
     and here.
96
   
97
 
98
  
99
100
 
101
102
Associative
103
104
 
105
 
106
  
Insertion Hints
107
 
108
   
109
     Section [23.1.2], Table 69, of the C++ standard lists this
110
     function for all of the associative containers (map, set, etc):
111
   
112
   
113
      a.insert(p,t);
114
   
115
   
116
     where 'p' is an iterator into the container 'a', and 't' is the
117
     item to insert.  The standard says that t is
118
     inserted as close as possible to the position just prior to
119
     p. (Library DR #233 addresses this topic,
120
     referring to N1780.
121
     Since version 4.2 GCC implements the resolution to DR 233, so
122
     that insertions happen as close as possible to the hint. For
123
     earlier releases the hint was only used as described below.
124
   
125
   
126
     Here we'll describe how the hinting works in the libstdc++
127
     implementation, and what you need to do in order to take
128
     advantage of it.  (Insertions can change from logarithmic
129
     complexity to amortized constant time, if the hint is properly
130
     used.)  Also, since the current implementation is based on the
131
     SGI STL one, these points may hold true for other library
132
     implementations also, since the HP/SGI code is used in a lot of
133
     places.
134
   
135
   
136
     In the following text, the phrases greater
137
     than and less than refer to the
138
     results of the strict weak ordering imposed on the container by
139
     its comparison object, which defaults to (basically)
140
     <.  Using those phrases is semantically sloppy,
141
     but I didn't want to get bogged down in syntax.  I assume that if
142
     you are intelligent enough to use your own comparison objects,
143
     you are also intelligent enough to assign greater
144
     and lesser their new meanings in the next
145
     paragraph.  *grin*
146
   
147
   
148
     If the hint parameter ('p' above) is equivalent to:
149
   
150
     
151
      
152
        
153
          begin(), then the item being inserted should
154
          have a key less than all the other keys in the container.
155
          The item will be inserted at the beginning of the container,
156
          becoming the new entry at begin().
157
      
158
      
159
      
160
        
161
          end(), then the item being inserted should have
162
          a key greater than all the other keys in the container.  The
163
          item will be inserted at the end of the container, becoming
164
          the new entry before end().
165
      
166
      
167
      
168
        
169
          neither begin() nor end(), then:
170
          Let h be the entry in the container pointed to
171
          by hint, that is, h = *hint.  Then
172
          the item being inserted should have a key less than that of
173
          h, and greater than that of the item preceding
174
          h.  The new item will be inserted between
175
          h and h's predecessor.
176
          
177
      
178
     
179
   
180
     For multimap and multiset, the
181
     restrictions are slightly looser: greater than
182
     should be replaced by not less thanand less
183
     than should be replaced by not greater
184
     than. (Why not replace greater with
185
     greater-than-or-equal-to?  You probably could in your head, but
186
     the mathematicians will tell you that it isn't the same thing.)
187
   
188
   
189
     If the conditions are not met, then the hint is not used, and the
190
     insertion proceeds as if you had called  a.insert(t)
191
      instead.  (Note  that GCC releases
192
     prior to 3.0.2 had a bug in the case with hint ==
193
     begin() for the map and set
194
     classes.  You should not use a hint argument in those releases.)
195
   
196
   
197
     This behavior goes well with other containers'
198
     insert() functions which take an iterator: if used,
199
     the new item will be inserted before the iterator passed as an
200
     argument, same as the other containers.
201
   
202
   
203
     Note  also that the hint in this
204
     implementation is a one-shot.  The older insertion-with-hint
205
     routines check the immediately surrounding entries to ensure that
206
     the new item would in fact belong there.  If the hint does not
207
     point to the correct place, then no further local searching is
208
     done; the search begins from scratch in logarithmic time.
209
   
210
  
211
 
212
 
213
  
bitset
214
    
215
 
216
    
Size Variable
217
 
218
      
219
        No, you cannot write code of the form
220
      
221
      
222
   
223
      #include <bitset>
224
 
225
      void foo (size_t n)
226
      {
227
          std::bitset<n>   bits;
228
          ....
229
      }
230
   
231
   
232
     because n must be known at compile time.  Your
233
     compiler is correct; it is not a bug.  That's the way templates
234
     work.  (Yes, it is a feature.)
235
   
236
   
237
     There are a couple of ways to handle this kind of thing.  Please
238
     consider all of them before passing judgement.  They include, in
239
     no chaptericular order:
240
   
241
      
242
        A very large N in bitset<N>.
243
        A container<bool>.
244
        Extremely weird solutions.
245
      
246
   
247
     A very large N in
248
     bitset<N>.   It has been
249
     pointed out a few times in newsgroups that N bits only takes up
250
     (N/8) bytes on most systems, and division by a factor of eight is
251
     pretty impressive when speaking of memory.  Half a megabyte given
252
     over to a bitset (recall that there is zero space overhead for
253
     housekeeping info; it is known at compile time exactly how large
254
     the set is) will hold over four million bits.  If you're using
255
     those bits as status flags (e.g.,
256
     changed/unchanged flags), that's a
257
     lot of state.
258
   
259
   
260
     You can then keep track of the maximum bit used
261
     during some testing runs on representative data, make note of how
262
     many of those bits really need to be there, and then reduce N to
263
     a smaller number.  Leave some extra space, of course.  (If you
264
     plan to write code like the incorrect example above, where the
265
     bitset is a local variable, then you may have to talk your
266
     compiler into allowing that much stack space; there may be zero
267
     space overhead, but it's all allocated inside the object.)
268
   
269
   
270
     A container<bool>.   The
271
     Committee made provision for the space savings possible with that
272
     (N/8) usage previously mentioned, so that you don't have to do
273
     wasteful things like Container<char> or
274
     Container<short int>.  Specifically,
275
     vector<bool> is required to be specialized for
276
     that space savings.
277
   
278
   
279
     The problem is that vector<bool> doesn't
280
     behave like a normal vector anymore.  There have been
281
     journal articles which discuss the problems (the ones by Herb
282
     Sutter in the May and July/August 1999 issues of C++ Report cover
283
     it well).  Future revisions of the ISO C++ Standard will change
284
     the requirement for vector<bool>
285
     specialization.  In the meantime, deque<bool>
286
     is recommended (although its behavior is sane, you probably will
287
     not get the space savings, but the allocation scheme is different
288
     than that of vector).
289
   
290
   
291
     Extremely weird solutions.   If
292
     you have access to the compiler and linker at runtime, you can do
293
     something insane, like figuring out just how many bits you need,
294
     then writing a temporary source code file.  That file contains an
295
     instantiation of bitset for the required number of
296
     bits, inside some wrapper functions with unchanging signatures.
297
     Have your program then call the compiler on that file using
298
     Position Independent Code, then open the newly-created object
299
     file and load those wrapper functions.  You'll have an
300
     instantiation of bitset<N> for the exact
301
     N that you need at the time.  Don't forget to delete
302
     the temporary files.  (Yes, this can be, and
303
     has been, done.)
304
   
305
   
306
   
307
     This would be the approach of either a visionary genius or a
308
     raving lunatic, depending on your programming and management
309
     style.  Probably the latter.
310
   
311
   
312
     Which of the above techniques you use, if any, are up to you and
313
     your intended application.  Some time/space profiling is
314
     indicated if it really matters (don't just guess).  And, if you
315
     manage to do anything along the lines of the third category, the
316
     author would love to hear from you...
317
   
318
   
319
     Also note that the implementation of bitset used in libstdc++ has
320
     some extensions.
321
   
322
 
323
    
324
    
Type String
325
 
326
      
327
      
328
   
329
     Bitmasks do not take char* nor const char* arguments in their
330
     constructors.  This is something of an accident, but you can read
331
     about the problem: follow the library's Links from
332
     the homepage, and from the C++ information defect
333
     reflector link, select the library issues list.  Issue
334
     number 116 describes the problem.
335
   
336
   
337
     For now you can simply make a temporary string object using the
338
     constructor expression:
339
   
340
   
341
      std::bitset<5> b ( std::string(10110) );
342
   
343
 
344
   
345
     instead of
346
   
347
 
348
    
349
      std::bitset<5> b ( 10110 );    // invalid
350
    
351
    
352
  
353
 
354
355
 
356
357
Interacting with C
358
359
 
360
 
361
  
Containers vs. Arrays
362
 
363
   
364
     You're writing some code and can't decide whether to use builtin
365
     arrays or some kind of container.  There are compelling reasons
366
     to use one of the container classes, but you're afraid that
367
     you'll eventually run into difficulties, change everything back
368
     to arrays, and then have to change all the code that uses those
369
     data types to keep up with the change.
370
   
371
   
372
     If your code makes use of the standard algorithms, this isn't as
373
     scary as it sounds.  The algorithms don't know, nor care, about
374
     the kind of container on which they work, since
375
     the algorithms are only given endpoints to work with.  For the
376
     container classes, these are iterators (usually
377
     begin() and end(), but not always).
378
     For builtin arrays, these are the address of the first element
379
     and the past-the-end element.
380
   
381
   
382
     Some very simple wrapper functions can hide all of that from the
383
     rest of the code.  For example, a pair of functions called
384
     beginof can be written, one that takes an array,
385
     another that takes a vector.  The first returns a pointer to the
386
     first element, and the second returns the vector's
387
     begin() iterator.
388
   
389
   
390
     The functions should be made template functions, and should also
391
     be declared inline.  As pointed out in the comments in the code
392
     below, this can lead to beginof being optimized out
393
     of existence, so you pay absolutely nothing in terms of increased
394
     code size or execution time.
395
   
396
   
397
     The result is that if all your algorithm calls look like
398
   
399
   
400
   std::transform(beginof(foo), endof(foo), beginof(foo), SomeFunction);
401
   
402
   
403
     then the type of foo can change from an array of ints to a vector
404
     of ints to a deque of ints and back again, without ever changing
405
     any client code.
406
   
407
 
408
409
// beginof
410
template<typename T>
411
  inline typename vector<T>::iterator
412
  beginof(vector<T> &v)
413
  { return v.begin(); }
414
 
415
template<typename T, unsigned int sz>
416
  inline T*
417
  beginof(T (&array)[sz]) { return array; }
418
 
419
// endof
420
template<typename T>
421
  inline typename vector<T>::iterator
422
  endof(vector<T> &v)
423
  { return v.end(); }
424
 
425
template<typename T, unsigned int sz>
426
  inline T*
427
  endof(T (&array)[sz]) { return array + sz; }
428
 
429
// lengthof
430
template<typename T>
431
  inline typename vector<T>::size_type
432
  lengthof(vector<T> &v)
433
  { return v.size(); }
434
 
435
template<typename T, unsigned int sz>
436
  inline unsigned int
437
  lengthof(T (&)[sz]) { return sz; }
438
439
 
440
   
441
     Astute readers will notice two things at once: first, that the
442
     container class is still a vector<T> instead
443
     of a more general Container<T>.  This would
444
     mean that three functions for deque would have to be
445
     added, another three for list, and so on.  This is
446
     due to problems with getting template resolution correct; I find
447
     it easier just to give the extra three lines and avoid confusion.
448
   
449
   
450
     Second, the line
451
   
452
   
453
    inline unsigned int lengthof (T (&)[sz]) { return sz; }
454
   
455
   
456
     looks just weird!  Hint:  unused parameters can be left nameless.
457
   
458
  
459
 
460
461
 
462

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.