File: tuple_users_guide.html

package info (click to toggle)
boost1.35 1.35.0-5
links: PTS
area: main
in suites: lenny
size: 203,856 kB
ctags: 337,867
sloc: cpp: 938,683; xml: 56,847; ansic: 41,589; python: 18,999; sh: 11,566; makefile: 664; perl: 494; yacc: 456; asm: 353; csh: 6
file content (529 lines) | stat: -rw-r--r-- 21,900 bytes
<html>
<head>
<title>The Boost Tuple Library</title>
</head>
<body bgcolor="#FFFFFF" text="#000000">

<IMG SRC="../../../boost.png" 
     ALT="C++ Boost" width="277" height="86">

<h1>The Boost Tuple Library</h1>

<p>
A tuple (or <i>n</i>-tuple) is a fixed size collection of elements. 
Pairs, triples, quadruples etc. are tuples. 
In a programming language, a tuple is a data object containing other objects as elements. 
These element objects may be of different types.
</p>

<p>Tuples are convenient in many circumstances. 
For instance, tuples make it easy to define functions that return more than one value.
</p>

<p>
Some programming languages, such as ML, Python and Haskell, have built-in tuple constructs. 
Unfortunately C++ does not.
To compensate for this &quot;deficiency&quot;, the Boost Tuple Library implements a tuple construct using templates.
</p>

<h2>Table of Contents</h2>

<ol>
<li><a href = "#using_library">Using the library</a></li>
<li><a href = "#tuple_types">Tuple types</a></li>
<li><a href = "#constructing_tuples">Constructing tuples</a></li>
<li><a href = "#accessing_elements">Accessing tuple elements</a></li>
<li><a href = "#construction_and_assignment">Copy construction and tuple assignment</a></li>
<li><a href = "#relational_operators">Relational operators</a></li>
<li><a href = "#tiers">Tiers</a></li>
<li><a href = "#streaming">Streaming</a></li>
<li><a href = "#performance">Performance</a></li>
<li><a href = "#portability">Portability</a></li>
<li><a href = "#thanks">Acknowledgements</a></li>
<li><a href = "#references">References</a></li>
</ol>

<h4>More details</h4>

<p>
<a href = "tuple_advanced_interface.html">Advanced features</a> (describes some metafunctions etc.).</p>
<p>
<a href = "design_decisions_rationale.html">Rationale behind some design/implementation decisions.</a></p>


<h2><a name="using_library">Using the library</a></h2>

<p>To use the library, just include:

<pre><code>#include &quot;boost/tuple/tuple.hpp&quot;</code></pre>

<p>Comparison operators can be included with:
<pre><code>#include &quot;boost/tuple/tuple_comparison.hpp&quot;</code></pre>

<p>To use tuple input and output operators,

<pre><code>#include &quot;boost/tuple/tuple_io.hpp&quot;</code></pre>

Both <code>tuple_io.hpp</code> and <code>tuple_comparison.hpp</code> include <code>tuple.hpp</code>.

<p>All definitions are in namespace <code>::boost::tuples</code>, but the most common names are lifted to namespace <code>::boost</code> with using declarations. These names are: <code>tuple</code>, <code>make_tuple</code>, <code>tie</code> and <code>get</code>. Further, <code>ref</code> and <code>cref</code> are defined directly under the <code>::boost</code> namespace.

<h2><a name = "tuple_types">Tuple types</a></h2>

<p>A tuple type is an instantiation of the <code>tuple</code> template. 
The template parameters specify the types of the tuple elements.
The current version supports tuples with 0-10 elements. 
If necessary, the upper limit can be increased up to, say, a few dozen elements.
The data element can be any C++ type.
Note that <code>void</code> and plain function types are valid 
C++ types, but objects of such types cannot exist. 
Hence, if a tuple type contains such types as elements, the tuple type
can exist, but not an object of that type.
There are natural limitations for element types that cannot
be be copied, or that are not default constructible (see 'Constructing tuples'
 below). 

<p>
For example, the following definitions are valid tuple instantiations (<code>A</code>, <code>B</code> and <code>C</code> are some user defined classes):

<pre><code>tuple&lt;int&gt;
tuple&lt;double&amp;, const double&amp;, const double, double*, const double*&gt;
tuple&lt;A, int(*)(char, int), B(A::*)(C&amp;), C&gt;
tuple&lt;std::string, std::pair&lt;A, B&gt; &gt;
tuple&lt;A*, tuple&lt;const A*, const B&amp;, C&gt;, bool, void*&gt;
</code></pre>

<h2><a name = "constructing_tuples">Constructing tuples</a></h2>

<p>
The tuple constructor takes the tuple elements as arguments. 
For an <i>n</i>-element tuple, the constructor can be invoked with <i>k</i> arguments, where 0 &lt;= <i>k</i> &lt;= <i>n</i>.
For example:
<pre><code>tuple&lt;int, double&gt;() 
tuple&lt;int, double&gt;(1) 
tuple&lt;int, double&gt;(1, 3.14)
</code></pre>

<p>
If no initial value for an element is provided, it is default initialized (and hence must be default initializable).
For example.

<pre><code>class X {
  X(); 
public:
  X(std::string);
};

tuple&lt;X,X,X&gt;()                                              // error: no default constructor for X
tuple&lt;X,X,X&gt;(string(&quot;Jaba&quot;), string(&quot;Daba&quot;), string(&quot;Duu&quot;)) // ok
</code></pre>

In particular, reference types do not have a default initialization: 

<pre><code>tuple&lt;double&amp;&gt;()                // error: reference must be 
                                // initialized explicitly

double d = 5; 
tuple&lt;double&amp;&gt;(d)               // ok

tuple&lt;double&amp;&gt;(d+3.14)          // error: cannot initialize 
                                // non-const reference with a temporary

tuple&lt;const double&amp;&gt;(d+3.14)    // ok, but dangerous: 
                                // the element becomes a dangling reference 
</code></pre>

<p>Using an initial value for an element that cannot be copied, is a compile
time error:

<pre><code>class Y { 
  Y(const Y&amp;); 
public:
  Y();
};

char a[10];

tuple&lt;char[10], Y&gt;(a, Y()); // error, neither arrays nor Y can be copied
tuple&lt;char[10], Y&gt;();       // ok
</code></pre>

Note particularly that the following is perfectly ok:
<code><pre>Y y;
tuple&lt;char(&amp;)[10], Y&amp;&gt;(a, y); 
</code></pre>

It is possible to come up with a tuple type that cannot be constructed.
This occurs if an element that cannot be initialized has a lower
index than an element that requires initialization. 
For example: <code>tuple&lt;char[10], int&amp;&gt;</code>.

<p>In sum, the tuple construction is semantically just a group of individual elementary constructions.
</p>

<h4><a name="make_tuple">The <code>make_tuple</code> function</a></h4>

<p>
Tuples can also be constructed using the <code>make_tuple</code> (cf. <code>std::make_pair</code>) helper functions.
This makes the construction more convenient, saving the programmer from explicitly specifying the element types: 
<pre><code>tuple&lt;int, int, double&gt; add_multiply_divide(int a, int b) {
  return make_tuple(a+b, a*b, double(a)/double(b));
}
</code></pre>

<p>
By default, the element types are deduced to the plain non-reference types. E.g: 
<pre><code>void foo(const A&amp; a, B&amp; b) { 
  ...
  make_tuple(a, b);
</code></pre>
The <code>make_tuple</code> invocation results in a tuple of type <code>tuple&lt;A, B&gt;</code>.

<p>
Sometimes the plain non-reference type is not desired, e.g. if the element type cannot be copied. 
Therefore, the programmer can control the type deduction and state that a reference to const or reference to
non-const type should be used as the element type instead.
This is accomplished with two helper template functions: <code>ref</code> and <code>cref</code>. 
Any argument can be wrapped with these functions to get the desired type. 
The mechanism does not compromise const correctness since a const object wrapped with <code>ref</code> results in a tuple element with const reference type (see the fifth code line below).
For example:

<pre><code>A a; B b; const A ca = a;
make_tuple(cref(a), b);      // creates tuple&lt;const A&amp;, B&gt;
make_tuple(ref(a), b);       // creates tuple&lt;A&amp;, B&gt;
make_tuple(ref(a), cref(b)); // creates tuple&lt;A&amp;, const B&amp;&gt;
make_tuple(cref(ca));        // creates tuple&lt;const A&amp;&gt;
make_tuple(ref(ca));         // creates tuple&lt;const A&amp;&gt;
</code></pre>


<p>
Array arguments to <code>make_tuple</code> functions are deduced to reference to const types by default; there is no need to wrap them with <code>cref</code>. For example:
<pre><code>make_tuple(&quot;Donald&quot;, &quot;Daisy&quot;);
</code></pre>

This creates an object of type <code>tuple&lt;const char (&amp;)[7], const char (&amp;)[6]&gt;</code> 
(note that the type of a string literal is an array of const characters, not <code>const char*</code>). 
However, to get <code>make_tuple</code> to create a tuple with an element of a
non-const array type one must use the <code>ref</code> wrapper.

<p>
Function pointers are deduced to the plain non-reference type, that is, to plain function pointer. 
A tuple can also hold a reference to a function, 
but such a tuple cannot be constructed with <code>make_tuple</code> (a const qualified function type would result, which is illegal):
<pre><code>void f(int i);
  ...
make_tuple(&amp;f); // tuple&lt;void (*)(int)&gt;
  ...
tuple&lt;tuple&lt;void (&amp;)(int)&gt; &gt; a(f) // ok
make_tuple(f);                    // not ok
</code></pre>

<h2><a name = "accessing_elements">Accessing tuple elements</a></h2>

<p>
Tuple elements are accessed with the expression:

<pre><code>t.get&lt;N&gt;()
</code></pre>
or
<pre><code>get&lt;N&gt;(t)
</code></pre>
where <code>t</code> is a tuple object and <code>N</code> is a constant integral expression specifying the index of the element to be accessed.
Depending on whether <code>t</code> is const or not, <code>get</code> returns the <code>N</code>th element as a reference to const or
non-const type.
The index of the first element is 0 and thus<code>
N</code> must be between 0 and <code>k-1</code>, where <code>k</code> is the number of elements in the tuple. 
Violations of  these constrains are detected at compile time. Examples:

<pre><code>double d = 2.7; A a;
tuple&lt;int, double&amp;, const A&amp;&gt; t(1, d, a);
const tuple&lt;int, double&amp;, const A&amp;&gt; ct = t;
  ...
int i = get&lt;0&gt;(t); i = t.get&lt;0&gt;();        // ok
int j = get&lt;0&gt;(ct);                       // ok
get&lt;0&gt;(t) = 5;                            // ok 
get&lt;0&gt;(ct) = 5;                           // error, can't assign to const 
  ...
double e = get&lt;1&gt;(t); // ok   
get&lt;1&gt;(t) = 3.14;     // ok 
get&lt;2&gt;(t) = A();      // error, can't assign to const 
A aa = get&lt;3&gt;(t);     // error: index out of bounds 
  ...
++get&lt;0&gt;(t);  // ok, can be used as any variable
</code></pre>

Note! The member get functions are not supported with MS Visual C++ compiler.
Further, the compiler has trouble with finding the non-member get functions without an explicit namespace qualifier. 
Hence, all <code>get</code> calls should be qualified as: <code>tuples::get&lt;N&gt;(a_tuple)</code> when writing code that shoud compile with MSVC++ 6.0.

<h2><a name = "construction_and_assignment">Copy construction and tuple assignment</a></h2>

<p>
A tuple can be copy constructed from another tuple, provided that the element types are element-wise copy constructible.
Analogously, a tuple can be assigned to another tuple, provided that the element types are element-wise assignable.
For example:

<pre><code>class A {};
class B : public A {};
struct C { C(); C(const B&amp;); };
struct D { operator C() const; };
tuple&lt;char, B*, B, D&gt; t;
  ...
tuple&lt;int, A*, C, C&gt; a(t); // ok 
a = t;                     // ok 
</code></pre>

In both cases, the conversions performed are: <code>char -> int</code>, <code>B* -> A*</code> (derived class pointer to base class pointer),  <code>B -> C</code> (a user defined conversion) and <code>D -> C</code> (a user defined conversion).

<p>
Note that assignment is also defined from <code>std::pair</code> types:

<pre><code>tuple&lt;float, int&gt; a = std::make_pair(1, 'a');
</code></pre>

<h2><a name = "relational_operators">Relational operators</a></h2>
<p>
Tuples reduce the operators <code>==, !=, &lt;, >, &lt;=</code> and <code>>=</code> to the corresponding elementary operators. 
This means, that if any of these operators is defined between all elements of two tuples, then the same operator is defined between the tuples as well.

The equality operators for two tuples <code>a</code> and <code>b</code> are defined as:
<ul>
<li><code>a == b</code> iff for each <code>i</code>: <code>a<sub>i</sub> == b<sub>i</sub></code></li>
<li><code>a != b</code> iff exists <code>i</code>: <code>a<sub>i</sub> != b<sub>i</sub></code></li>
</ul>

The operators <code>&lt;, >, &lt;=</code> and <code>>=</code> implement a lexicographical ordering.

<p>
Note that an attempt to compare two tuples of different lengths results in a compile time error.</p>
Also, the comparison operators are <i>"short-circuited"</i>: elementary comparisons start from the first elements and are performed only until the result is clear.

<p>Examples:

<pre><code>tuple&lt;std::string, int, A&gt; t1(std::string(&quot;same?&quot;), 2, A());
tuple&lt;std::string, long, A&gt; t2(std::string(&quot;same?&quot;), 2, A());
tuple&lt;std::string, long, A&gt; t3(std::string(&quot;different&quot;), 3, A());

bool operator==(A, A) { std::cout &lt;&lt; &quot;All the same to me...&quot;; return true; }

t1 == t2; 		// true
t1 == t3;               // false, does not print &quot;All the...&quot;
</code></pre>


<h2><a name = "tiers">Tiers</a></h2>

<p>
<i>Tiers</i> are tuples, where all elements are of non-const reference types.
They are constructed with a call to the <code>tie</code> function template (cf. <code>make_tuple</code>):

<pre><code>int i; char c; double d; 
  ...
tie(i, c, a);
</code></pre>

<p>
The above <code>tie</code> function creates a tuple of type <code>tuple&lt;int&amp;, char&amp;, double&amp;&gt;</code>. 
The same result could be achieved with the call <code>make_tuple(ref(i), ref(c), ref(a))</code>.
</p>

<p>
A tuple that contains non-const references as elements can be used to 'unpack' another tuple into variables. E.g.:

<pre><code>int i; char c; double d; 
tie(i, c, d) = make_tuple(1,'a', 5.5);
std::cout &lt;&lt; i &lt;&lt; &quot; &quot; &lt;&lt;  c &lt;&lt; &quot; &quot; &lt;&lt; d;
</code></pre>
This code prints <code>1 a 5.5</code> to the standard output stream.

A tuple unpacking operation like this is found for example in ML and Python. 
It is convenient when calling functions which return tuples.

<p>
The tying mechanism works with <code>std::pair</code> templates as well:

<pre><code>int i; char c;
tie(i, c) = std::make_pair(1, 'a');
</code></pre>
<h4>Ignore</h4>
There is also an object called <code>ignore</code> which allows you to ignore an element assigned by a tuple.
The idea is that a function may return a tuple, only part of which you are interested in. For example (note, that <code>ignore</code> is under the <code>tuples</code> subnamespace):

<pre><code>char c;
tie(tuples::ignore, c) = std::make_pair(1, 'a');
</code></pre>

<h2><a name = "streaming">Streaming</a></h2>

<p>
The global <code>operator&lt;&lt;</code> has been overloaded for <code>std::ostream</code> such that tuples are 
output by recursively calling <code>operator&lt;&lt;</code> for each element. 
</p>

<p>
Analogously, the global <code>operator&gt;&gt;</code> has been overloaded to extract tuples from <code>std::istream</code>  by recursively calling <code>operator&gt;&gt;</code> for each element. 
</p>

<p>
The default delimiter between the elements is space, and the tuple is enclosed
in parenthesis.
For Example:

<pre><code>tuple&lt;float, int, std::string&gt; a(1.0f,  2, std::string(&quot;Howdy folks!&quot;);

cout &lt;&lt; a; 
</code></pre>
outputs the tuple as: <code>(1.0 2 Howdy folks!)</code>

<p>
The library defines three <i>manipulators</i> for changing the default behavior:
<ul>
<li><code>set_open(char)</code> defines the character that is output before the first
element.</li>
<li><code>set_close(char)</code> defines the character that is output after the
last element.</li>
<li><code>set_delimiter(char)</code> defines the delimiter character between
elements.</li>
</ul>

Note, that these manipulators are defined in the <code>tuples</code> subnamespace. 
For example:
<code><pre>cout &lt;&lt; tuples::set_open('[') &lt;&lt; tuples::set_close(']') &lt;&lt; tuples::set_delimiter(',') &lt;&lt; a; 
</code></pre>
outputs the same tuple <code>a</code> as: <code>[1.0,2,Howdy folks!]</code>

<p>The same manipulators work with <code>operator&gt;&gt;</code> and <code>istream</code> as well. Suppose the <code>cin</code> stream contains the following data:

<pre><code>(1 2 3) [4:5]</code></pre>

The code:

<code><pre>tuple&lt;int, int, int&gt; i;
tuple&lt;int, int&gt; j;

cin &gt;&gt; i;
cin &gt;&gt; tuples::set_open('[') &gt;&gt; tuples::set_close(']') &gt;&gt; tules::set_delimiter(':');
cin &gt;&gt; j;
</code></pre>

reads the data into the tuples <code>i</code> and <code>j</code>.

<p>
Note that extracting tuples with <code>std::string</code> or C-style string
elements does not generally work, since the streamed tuple representation may not be unambiguously
parseable.
</p>

<h2><a name = "performance">Performance</a></h2>

All tuple access and construction functions are small inlined one-liners. 
Therefore, a decent compiler can eliminate any extra cost of using tuples compared to using hand written tuple like classes. 
Particularly, with a decent compiler there is no performance difference between this code:

<pre><code>class hand_made_tuple { 
  A a; B b; C c;
public:
  hand_made_tuple(const A&amp; aa, const B&amp; bb, const C&amp; cc) 
    : a(aa), b(bb), c(cc) {};
  A&amp; getA() { return a; };
  B&amp; getB() { return b; };
  C&amp; getC() { return c; };
};

hand_made_tuple hmt(A(), B(), C()); 
hmt.getA(); hmt.getB(); hmt.getC();
</code></pre>

and this code:

<pre><code>tuple&lt;A, B, C&gt; t(A(), B(), C());
t.get&lt;0&gt;(); t.get&lt;1&gt;(); t.get&lt;2&gt;(); 
</code></pre>

<p>Note, that there are widely used compilers (e.g. bcc 5.5.1) which fail to optimize this kind of tuple usage.
</p>  
<p>
Depending on the optimizing ability of the compiler, the tier mechanism may have a small performance penalty compared to using
non-const reference parameters as a mechanism for returning multiple values from a function. 
For example, suppose that the following functions <code>f1</code> and <code>f2</code> have equivalent functionalities:

<pre><code>void f1(int&amp;, double&amp;);
tuple&lt;int, double&gt; f2();
</code></pre>

Then, the call #1 may be slightly faster than #2 in the code below:

<pre><code>int i; double d;
  ...
f1(i,d);         // #1
tie(i,d) = f2(); // #2
</code></pre>
See 
[<a href="#publ_1">1</a>,
<a href="#publ_2">2</a>]
 for more in-depth discussions about efficiency.

<h4>Effect on Compile Time</h4>

<p>
Compiling tuples can be slow due to the excessive amount of template instantiations.
Depending on the compiler and the tuple length, it may be more than 10 times slower to compile a tuple construct, compared to compiling an equivalent explicitly written class, such as the <code>hand_made_tuple</code> class above.
However, as a realistic program is likely to contain a lot of code in addition to tuple definitions, the difference is probably unnoticeable.
Compile time increases between 5 to 10 percentages were measured for programs which used tuples very frequently.
With the same test programs, memory consumption of compiling increased between 22% to 27%. See  
[<a href="#publ_1">1</a>,
<a href="#publ_2">2</a>]
for details.
</p>

<h2><a name = "portability">Portability</a></h2>

<p>The library code is(?) standard C++ and thus the library works with a standard conforming compiler. 
Below is a list of compilers and known problems with each compiler:
</p>
<table>
<tr><td><u>Compiler</u></td><td><u>Problems</u></td></tr>
<tr><td>gcc 2.95</td><td>-</td></tr>
<tr><td>edg 2.44</td><td>-</td></tr>
<tr><td>Borland 5.5</td><td>Can't use function  pointers or member pointers as tuple elements</td></tr>
<tr><td>Metrowerks 6.2</td><td>Can't use <code>ref</code> and <code>cref</code> wrappers</td></tr>
<tr><td>MS Visual C++</td><td>No reference elements (<code>tie</code> still works). Can't use <code>ref</code> and <code>cref</code> wrappers</td></tr>
</table>

<h2><a name = "thanks">Acknowledgements</a></h2>
Gary Powell has been an indispensable helping hand. In particular, stream manipulators for tuples were his idea. Doug Gregor came up with a working version for MSVC, David Abrahams found a way to get rid of most of the restrictions for compilers not supporting partial specialization. Thanks to Jeremy Siek, William Kempf and Jens Maurer for their help and suggestions. 
The comments by Vesa Karvonen, John Max Skaller, Ed Brey, Beman Dawes, David Abrahams and Hartmut Kaiser helped to improve the
library.
The idea for the tie mechanism came from an old usenet article by Ian McCulloch, where he proposed something similar for std::pairs.
<h2><a name = "references">References</a></h2>

<p>
<a name="publ_1"></a>[1]
J&auml;rvi J.: <i>Tuples and multiple return values in C++</i>, TUCS Technical Report No 249, 1999<!-- (<a href="http://www.tucs.fi/Publications">http://www.tucs.fi/Publications</a>)-->.
</p>

<p>
<a name="publ_2"></a>[2]
J&auml;rvi J.: <i>ML-Style Tuple Assignment in Standard C++ - Extending the Multiple Return Value Formalism</i>, TUCS Technical Report No 267, 1999<!-- (<a href="http://www.tucs.fi/Publications">http://www.tucs.fi/Publications</a>)-->.
</p>

<p>
[3] J&auml;rvi J.:<i>Tuple Types and Multiple Return Values</i>, C/C++ Users Journal, August 2001.
</p>

<hr>

<p>Last modified 2003-09-07</p>

<p>&copy; Copyright <a href="http://www.boost.org/people/jaakko_jarvi.htm"> Jaakko J&auml;rvi</a> 2001. 

Permission to copy, use, modify, sell and distribute this software and its documentation is granted provided this copyright notice appears in all copies. 
This software and its documentation is provided "as is" without express or implied warranty, and with no claim as to its suitability for any purpose.
</p>
</body>
</html>