Chapter 23: Containers

Chapter 23 deals with container classes and what they offer.


Contents


Making code unaware of the container/array difference

You're writing some code and can't decide whether to use builtin arrays or some kind of container. There are compelling reasons to use one of the container classes, but you're afraid that you'll eventually run into difficulties, change everything back to arrays, and then have to change all the code that uses those data types to keep up with the change.

If your code makes use of the standard algorithms, this isn't as scary as it sounds. The algorithms don't know, nor care, about the kind of "container" on which they work, since the algorithms are only given endpoints to work with. For the container classes, these are iterators (usually begin() and end(), but not always). For builtin arrays, these are the address of the first element and the past-the-end element.

Some very simple wrapper functions can hide all of that from the rest of the code. For example, a pair of functions called beginof can be written, one that takes an array, another that takes a vector. The first returns a pointer to the first element, and the second returns the vector's begin() iterator.

The functions should be made template functions, and should also be declared inline. As pointed out in the comments in the code below, this can lead to beginof being optimized out of existence, so you pay absolutely nothing in terms of increased code size or execution time.

The result is that if all your algorithm calls look like

   std::transform(beginof(foo), endof(foo), beginof(foo), SomeFunction);

then the type of foo can change from an array of ints to a vector of ints to a deque of ints and back again, without ever changing any client code.

This author has a collection of such functions, called "*of" because they all extend the builtin "sizeof". It started with some Usenet discussions on a transparent way to find the length of an array. A simplified and much-reduced version for easier reading is given here.

Astute readers will notice two things at once: first, that the container class is still a vector<T> instead of a more general Container<T>. This would mean that three functions for deque would have to be added, another three for list, and so on. This is due to problems with getting template resolution correct; I find it easier just to give the extra three lines and avoid confusion.

Second, the line

    inline unsigned int lengthof (T (&)[sz]) { return sz; } 

looks just weird! Hint: unused parameters can be left nameless.

Return to top of page or to the FAQ.


Variable-sized bitmasks

No, you cannot write code of the form

      #include <bitset>

      void foo (size_t n)
      {
          std::bitset<n>   bits;
          ....
      } 

because n must be known at compile time. Your compiler is correct; it is not a bug. That's the way templates work. (Yes, it is a feature.)

There are a couple of ways to handle this kind of thing. Please consider all of them before passing judgement. They include, in no particular order:

A very large N in bitset<N>.   It has been pointed out a few times in newsgroups that N bits only takes up (N/8) bytes on most systems, and division by a factor of eight is pretty impressive when speaking of memory. Half a megabyte given over to a bitset (recall that there is zero space overhead for housekeeping info; it is known at compile time exactly how large the set is) will hold over four million bits. If you're using those bits as status flags (e.g., "changed"/"unchanged" flags), that's a lot of state.

You can then keep track of the "maximum bit used" during some testing runs on representative data, make note of how many of those bits really need to be there, and then reduce N to a smaller number. Leave some extra space, of course. (If you plan to write code like the incorrect example above, where the bitset is a local variable, then you may have to talk your compiler into allowing that much stack space; there may be zero space overhead, but it's all allocated inside the object.)

A container<bool>.   The Committee made provision for the space savings possible with that (N/8) usage previously mentioned, so that you don't have to do wasteful things like Container<char> or Container<short int>. Specifically, vector<bool> is required to be specialized for that space savings.

The problem is that vector<bool> doesn't behave like a normal vector anymore. There have been recent journal articles which discuss the problems (the ones by Herb Sutter in the May and July/August 1999 issues of C++ Report cover it well). Future revisions of the ISO C++ Standard will change the requirement for vector<bool> specialization. In the meantime, deque<bool> is recommended (although its behavior is sane, you probably will not get the space savings, but the allocation scheme is different than that of vector).

Extremely weird solutions.   If you have access to the compiler and linker at runtime, you can do something insane, like figuring out just how many bits you need, then writing a temporary source code file. That file contains an instantiation of bitset for the required number of bits, inside some wrapper functions with unchanging signatures. Have your program then call the compiler on that file using Position Independent Code, then open the newly-created object file and load those wrapper functions. You'll have an instantiation of bitset<N> for the exact N that you need at the time. Don't forget to delete the temporary files. (Yes, this can be, and has been, done.)

This would be the approach of either a visionary genius or a raving lunatic, depending on your programming and management style. Probably the latter.

Which of the above techniques you use, if any, are up to you and your intended application. Some time/space profiling is indicated if it really matters (don't just guess). And, if you manage to do anything along the lines of the third category, the author would love to hear from you...

Also note that the implementation of bitset used in libstdc++-v3 has some extensions.

Return to top of page or to the FAQ.


Containers and multithreading

This section discusses issues surrounding the design of multithreaded applications which use Standard C++ containers. All information in this section is current as of the gcc 3.0 release and all later point releases. Although earlier gcc releases had a different approach to threading configuration and proper compilation, the basic code design rules presented here were similar. For information on all other aspects of multithreading as it relates to libstdc++, including details on the proper compilation of threaded code (and compatibility between threaded and non-threaded code), see Chapter 17.

Two excellent pages to read when working with the Standard C++ containers and threads are SGI's http://www.sgi.com/tech/stl/thread_safety.html and SGI's http://www.sgi.com/tech/stl/Allocators.html.

However, please ignore all discussions about the user-level configuration of the lock implementation inside the STL container-memory allocator on those pages. For the sake of this discussion, libstdc++-v3 configures the SGI STL implementation, not you. This is quite different from how gcc pre-3.0 worked. In particular, past advice was for people using g++ to explicitly define _PTHREADS or other macros or port-specific compilation options on the command line to get a thread-safe STL. This is no longer required for any port and should no longer be done unless you really know what you are doing and assume all responsibility.

Since the container implementation of libstdc++-v3 uses the SGI code, we use the same definition of thread safety as SGI when discussing design. A key point that beginners may miss is the fourth major paragraph of the first page mentioned above ("For most clients,"...), which points out that locking must nearly always be done outside the container, by client code (that'd be you, not us). There is a notable exceptions to this rule. Allocators called while a container or element is constructed uses an internal lock obtained and released solely within libstdc++-v3 code (in fact, this is the reason STL requires any knowledge of the thread configuration).

For implementing a container which does its own locking, it is trivial to provide a wrapper class which obtains the lock (as SGI suggests), performs the container operation, and then releases the lock. This could be templatized to a certain extent, on the underlying container and/or a locking mechanism. Trying to provide a catch-all general template solution would probably be more trouble than it's worth.

The STL implementation is currently configured to use the high-speed caching memory allocator. If you absolutely think you must change this on a global basis for your platform to better support multi-threading, then please consult all commentary in include/bits/stl_alloc.h and the allocators link below.

(Explicit warning since so many people get confused while attempting this:)

Adding -D__USE_MALLOC on the command line is almost certainly a bad idea. Memory efficiency is almost guaranteed to suffer as a result; this is why we disabled it for 3.0 in the first place.

Related to threading or otherwise, the current recommendation is that users not add any macro defines on the command line to remove or otherwise disable features of libstdc++-v3. There is no condition under which it will help you without causing other issues to perhaps raise up (possible linkage/ABI problems). In particular, __USE_MALLOC should only be added to a libstdc++-v3 configuration file, include/bits/c++config (where such user action is cautioned against), and the entire library should be rebuilt. If you do not, then you might be violating the one-definition rule of C/C++ and you might cause yourself untold problems.

If you find any platform where gcc reports a threading model other than single, and where libstdc++-v3 builds a buggy container allocator when used with threads unless you define __USE_MALLOC, we want to hear about it ASAP. In the past, correctness was the main reason people were led to believe that they should define __USE_MALLOC when using threads.

There is a better way (not standardized yet): It is possible to force the malloc-based allocator on a per-case-basis for some application code. The library team generally believes that this is a better way to tune an application for high-speed using this implementation of the STL. There is more information on allocators here.

Return to top of page or to the FAQ.


"Hinting" during insertion

Section [23.1.2], Table 69, of the C++ standard lists this function for all of the associative containers (map, set, etc):

      a.insert(p,t);

where 'p' is an iterator into the container 'a', and 't' is the item to insert. The standard says that "iterator p is a hint pointing to where the insert should start to search," but specifies nothing more. (LWG Issue #233, currently in review, addresses this topic, but I will ignore it here because it is not yet finalized.)

Here we'll describe how the hinting works in the libstdc++-v3 implementation, and what you need to do in order to take advantage of it. (Insertions can change from logarithmic complexity to amortized constant time, if the hint is properly used.) Also, since the current implementation is based on the SGI STL one, these points may hold true for other library implementations also, since the HP/SGI code is used in a lot of places.

In the following text, the phrases greater than and less than refer to the results of the strict weak ordering imposed on the container by its comparison object, which defaults to (basically) "<". Using those phrases is semantically sloppy, but I didn't want to get bogged down in syntax. I assume that if you are intelligent enough to use your own comparison objects, you are also intelligent enough to assign "greater" and "lesser" their new meanings in the next paragraph. *grin*

If the hint parameter ('p' above) is equivalent to:

For multimap and multiset, the restrictions are slightly looser: "greater than" should be replaced by "not less than" and "less than" should be replaced by "not greater than." (Why not replace greater with greater-than-or-equal-to? You probably could in your head, but the mathematicians will tell you that it isn't the same thing.)

If the conditions are not met, then the hint is not used, and the insertion proceeds as if you had called a.insert(t) instead. (Note that GCC releases prior to 3.0.2 had a bug in the case with hint == begin() for the map and set classes. You should not use a hint argument in those releases.)

This behavior goes well with other container's insert() functions which take an iterator: if used, the new item will be inserted before the iterator passed as an argument, same as the other containers. The exception (in a sense) is with a hint of end(): the new item will actually be inserted after end(), but it also becomes the new end().

Note also that the hint in this implementation is a one-shot. The insertion-with-hint routines check the immediately surrounding entries to ensure that the new item would in fact belong there. If the hint does not point to the correct place, then no further local searching is done; the search begins from scratch in logarithmic time. (Further local searching would only increase the time required when the hint is too far off.)

Return to top of page or to the FAQ.


Bitmasks and string arguments

Bitmasks do not take char* nor const char* arguments in their constructors. This is something of an accident, but you can read about the problem: follow the library's "Links" from the homepage, and from the C++ information "defect reflector" link, select the library issues list. Issue number 116 describes the problem.

For now you can simply make a temporary string object using the constructor expression:

      std::bitset<5> b ( std::string("10110") );
      
instead of
      std::bitset<5> b ( "10110" );    // invalid
      

Return to top of page or to the FAQ.


std::list::size() is O(n)!

Yes it is, and that's okay. This is a decision that we preserved when we imported SGI's STL implementation. The following is quoted from their FAQ:

The size() member function, for list and slist, takes time proportional to the number of elements in the list. This was a deliberate tradeoff. The only way to get a constant-time size() for linked lists would be to maintain an extra member variable containing the list's size. This would require taking extra time to update that variable (it would make splice() a linear time operation, for example), and it would also make the list larger. Many list algorithms don't require that extra word (algorithms that do require it might do better with vectors than with lists), and, when it is necessary to maintain an explicit size count, it's something that users can do themselves.

This choice is permitted by the C++ standard. The standard says that size() "should" be constant time, and "should" does not mean the same thing as "shall". This is the officially recommended ISO wording for saying that an implementation is supposed to do something unless there is a good reason not to.

One implication of linear time size(): you should never write

         if (L.size() == 0)
             ...
Instead, you should write
         if (L.empty())
             ...

Return to top of page or to the FAQ.


Space overhead management for vectors

In this message to the list, Daniel Kostecky announced work on an alternate form of std::vector that would support hints on the number of elements to be over-allocated. The design was also described, along with possible implementation choices.

The first two alpha releases were announced here and here. The releases themselves are available at http://www.kotelna.sk/dk/sw/caphint/.

Return to top of page or to the FAQ.


See license.html for copying conditions. Comments and suggestions are welcome, and may be sent to the libstdc++ mailing list.