Files
boost_integer/integer.htm

506 lines
18 KiB
HTML

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<html>
<head>
<title>Integer Type Selection Templates</title>
</head>
<body bgcolor="white" text="black">
<h1>
<img src="../../boost.png" alt="boost.png (6897 bytes)"
align="middle" width="277" height="86">Integer Type Selection
Templates</h1>
<p>The <cite><a
href="../../boost/integer.hpp">&lt;boost/integer.hpp&gt;</a></cite> type
selection templates allow integer types to be selected based on desired
characteristics such as number of bits or maximum value. This facility
is particularly useful for solving generic programming problems.</p>
<h2><a name="contents">Contents</a></h2>
<ul>
<li><a href="#contents">Contents</a></li>
<li><a href="#synopsis">Synopsis</a></li>
<li><a href="#easy">Processor-Optimized Types</a></li>
<li><a href="#sized">Sized Types</a></li>
<li><a href="#mpl">MPL-Compatible Variants</a></li>
<li><a href="#example">Example</a></li>
<li><a href="#demo">Demonstration Program</a></li>
<li><a href="#rationale">Rationale</a></li>
<li><a href="#alternative">Alternative</a></li>
<li><a href="#credits">Credits</a></li>
</ul>
<h2><a name="synopsis">Synopsis</a></h2>
<blockquote><pre>
#include &lt;<a href="../../boost/integer_fwd.hpp">boost/integer_fwd.hpp</a>&gt; // forwarding header
#include &lt;<a href="cstdint.htm">boost/cstdint.hpp</a>&gt; // for boost::uintmax_t, intmax_t
namespace boost
{
// fast integers from least integers
template&lt; typename BaseInt &gt;
struct fast_integral
{
typedef <em>implementation_supplied</em> type;
};
template&lt; typename LeastInt &gt;
struct int_fast_t
{
typedef typename fast_integral&lt;LeastInt&gt;::type fast;
};
// MPL-compatible
template&lt; int Bits, typename Signedness &gt;
struct sized_integral
{
static bool const is_specialized = <em>implementation_supplied</em>;
static bool const is_signed = <em>implementation_supplied</em>;
static int const bit_count = Bits;
typedef <em>implementation_supplied</em> type;
};
template&lt; int Bits, typename Signedness &gt;
struct exact_integral
{
static bool const is_specialized = <em>implementation_supplied</em>;
static bool const is_signed = <em>implementation_supplied</em>;
static int const bit_count = Bits;
typedef <em>implementation_supplied</em> type;
};
template&lt; intmax_t MaxValue &gt;
struct maximum_signed_integral
{
static bool const is_specialized = <em>implementation_supplied</em>;
static bool const is_signed = true;
static intmax_t const bound = MaxValue;
typedef <em>implementation_supplied</em> type;
};
template&lt; intmax_t MinValue &gt;
struct minimum_signed_integral
{
static bool const is_specialized = <em>implementation_supplied</em>;
static bool const is_signed = true;
static intmax_t const bound = MinValue;
typedef <em>implementation_supplied</em> type;
};
template&lt; uintmax_t Value &gt;
struct maximum_unsigned_integral
{
static bool const is_specialized = <em>implementation_supplied</em>;
static bool const is_signed = false;
static uintmax_t const bound = Value;
typedef <em>implementation_supplied</em> type;
};
// signed
template&lt; int Bits &gt;
struct int_t
{
typedef typename sized_integral&lt;Bits, signed&gt;::type least;
typedef int_fast_t&lt;least&gt;::fast fast;
};
template&lt; int Bits &gt;
struct int_exact_t
{
typedef typename exact_integral&lt;Bits, signed&gt;::type exact;
};
// unsigned
template&lt; int Bits &gt;
struct uint_t
{
typedef typename sized_integral&lt;Bits, unsigned&gt;::type least;
typedef int_fast_t&lt;least&gt;::fast fast;
};
template&lt; int Bits &gt;
struct uint_exact_t
{
typedef typename exact_integral&lt;Bits, unsigned&gt;::type exact;
};
// signed
template&lt; intmax_t MaxValue &gt;
struct int_max_value_t
{
typedef typename maximum_signed_integral&lt;MaxValue&gt;::type least;
typedef int_fast_t&lt;least&gt;::fast fast;
};
template&lt; intmax_t MinValue &gt;
struct int_min_value_t
{
typedef typename minimum_signed_integral&lt;MinValue&gt;::type least;
typedef int_fast_t&lt;least&gt;::fast fast;
};
// unsigned
template&lt; uintmax_t Value &gt;
struct uint_value_t
{
typedef typename maximum_unsigned_integral&lt;Value&gt;::type least;
typedef int_fast_t&lt;least&gt;::fast fast;
};
} // namespace boost
</pre></blockquote>
<h2><a name="easy">Processor-Optimized Types</a></h2>
<p>The <code>fast_integral</code> class template maps its input type to the
next-largest type that the processor can manipulate the easiest, or to
itself if the input type is already an easy-to-manipulate type. For
instance, processing a bunch of <code>char</code> objects may go faster
if they were converted to <code>int</code> objects before processing.
The input type, passed as the only template parameter, can be any built-in
integral type besides <code>bool</code>. The output type is given as the class
member <code>type</code>.</p>
<p>The <code>int_fast_t</code> class template is the classic meta-function for
this operation. Despite the name, it works for unsigned integral types just
like it works for the signed integral types. The output type is given as the
class member <code>fast</code>, defined to be the same as the corresponding
result from the <code>fast_integral</code> meta-function.</p>
<p><strong>Implementation Notes</strong><br>
By default, the output type is identical to the input type. Eventually,
this code's implementation should be conditionalized for each platform
to give accurate mappings between the built-in types and the
easiest-to-manipulate built-in types. Also, there is no guarantee that
the output type actually is easier to manipulate than the input
type.</p>
<h2><a name="sized">Sized Types</a></h2>
<p>The <code>int_t</code>, <code>int_exact_t</code>, <code>uint_t</code>,
<code>uint_exact_t</code>, <code>int_max_value_t</code>,
<code>int_min_value_t</code>, and <code>uint_value_t</code> class templates find
the most appropriate built-in integral type for the given template parameter.
This type is given by the class member <code>least</code> or <code>exact</code>.
For the non-exact class templates, the easiest-to-manipulate version of that
type is given by the class member <code>fast</code>. The following table
describes each template's criteria.</p>
<table border="2" cellpadding="5">
<caption>Criteria for the Sized Type Class Templates</caption>
<tr>
<th>Class Template (all in name-space <code>boost</code>)</th>
<th>Template Parameter Mapping</th>
</tr>
<tr>
<td><code>int_t</code></td>
<td>The smallest built-in signed integral type with at least the
given number of bits, including the sign bit. The parameter
<em>must</em> be a positive number. A compile-time error results if
the parameter is larger than the number of bits in a
<code>boost::intmax_t</code>.</td>
</tr>
<tr>
<td><code>int_exact_t</code></td>
<td>The smallest built-in signed integral type with exactly the
given number of bits, including the sign bit. A compile-time error
results if no qualifying type exists.</td>
</tr>
<tr>
<td><code>uint_t</code></td>
<td>The smallest built-in unsigned integral type with at least
the given number of bits. The parameter <em>must</em> be a
non-negative number. A compile-time error results if the parameter
is larger than the number of bits in a
<code>boost::uintmax_t</code>.</td>
</tr>
<tr>
<td><code>uint_exact_t</code></td>
<td>The smallest built-in unsigned integral type with exactly the given
number of bits. A compile-time error results if no qualifying type
exists.</td>
</tr>
<tr>
<td><code>int_max_value_t</code></td>
<td>The smallest built-in signed integral type that supports the
given value as a maximum. The parameter <em>must</em> be a
positive number.</td>
</tr>
<tr>
<td><code>int_min_value_t</code></td>
<td>The smallest built-in signed integral type that supports the
given value as a minimum. The parameter <em>must</em> be a
negative number.</td>
</tr>
<tr>
<td><code>uint_value_t</code></td>
<td>The smallest built-in unsigned integral type that supports
the given value as a maximum. The parameter should be a
positive number.</td>
</tr>
</table>
<h2><a name="mpl">MPL-Compatible Variants</a></h2>
<p>The bit-length sized-type class templates have several drawbacks:</p>
<ul>
<li>You must know the valid bit-lengths in advance.</li>
<li>There is no way to inspect the parameter used after a size-type template
class is aliased.</li>
<li>Using an inappropriate parameter value results in a compiler
diagnostic.</li>
<li>The type names used are inconsistent with other transformations in
Boost, like in <a href="../mpl/">MPL</a>.</li>
<li>The above two facts make use of the size-type class templates
incompatible with template meta-programming techniques.</li>
</ul>
<p>The <code>sized_integral</code>, <code>exact_integral</code>,
<code>maximum_signed_integral</code>, <code>minimum_signed_integral</code>, and
<code>maximum_unsigned_integral</code> class templates provide MPL-compatible
alternatives. These alternatives generally have the form:</p>
<blockquote><pre>
template&lt; <var>SwitchType</var> <var>SwitchValue</var>, typename Signedness &gt;
struct <var>name</var>
{
static bool const is_specialized = <em>implementation_supplied</em>;
static bool const is_signed = <em>implementation_supplied</em>;
static <var>SwitchType</var> const <var>switch_id</var> = <var>SwitchValue</var>;
typedef <em>implementation_supplied</em> type;
};
</pre></blockquote>
<p>Each member, if present, is defined by:</p>
<table border="2" cellpadding="5">
<caption>Members in MPL-Compatible Class Templates</caption>
<tr>
<th>Class Template Member</th>
<th>When Defined</th>
<th>Meaning</th>
</tr>
<tr>
<td><code>is_specialized</code></td>
<td>Always</td>
<td>Flag indicating when a particular template class instantiation is a
valid meta-function (<code>true</code>) or not (<code>false</code>).</td>
</tr>
<tr>
<td><code>is_signed</code></td>
<td><code>is_specialized == true</code></td>
<td>Flag indicating whether the signed-variant (<code>true</code>) or
the unsigned-variant (<code>false</code>) of the meta-function is
used. This is controlled by the <code>Signedness</code> template
parameter:
<table border="1" cellpadding="3" align="center">
<caption>Effect of <code>Signedness</code> Setting</caption>
<tr>
<th><code>Signedness</code> Type</th>
<th><code>is_signed</code></th>
</tr>
<tr>
<td><code>signed</code></td>
<td><code>true</code></td>
</tr>
<tr>
<td><code>unsigned</code></td>
<td><code>false</code></td>
</tr>
<tr>
<td>anything else</td>
<td><em>not defined</em></td>
</tr>
</table>
The type used is a programmer mnemonic; the compiler cannot prevent
someone from using <code>int</code> or <code>signed int</code>
instead of <code>signed</code>, or <code>unsigned int</code> instead
of <code>unsigned</code>.</td>
</tr>
<tr>
<td><code><var>switch_id</var></code> (Actual name is template-specific.)</td>
<td>Always</td>
<td>The value of the main control parameter, accessible even if the
template class instantiation is aliased.</td>
</tr>
<tr>
<td><code>type</code></td>
<td><code>is_specialized == true</code></td>
<td>The meta-function's result. It appears only if the input parameters
satisfy the template's requirements. Its presence, or lack thereof,
enables &quot;Substitution Failure Is Not An Error&quot; (SFINAE)
techniques, instead of a hard compiler diagnostic.</td>
</tr>
</table>
<p>The exceptions are the extreme-value class templates
(<code>maximum_signed_integral</code>, <code>minimum_signed_integral</code>, and
<code>maximum_unsigned_integral</code>), which do not take a <var>Signedness</var>
template parameter because the meta-functions already inherently have signedness.
<p>The following table describes each template's criteria. The classic signed
and unsigned equivalents are the sized-type class templates that each
MPL-compatible class template emulates. (The setting of <var>Signedness</var>
controls the appropriate emulation.)</p>
<table border="2" cellpadding="5">
<caption>Criteria for the MPL-Compatible Class Templates</caption>
<tr>
<th rowspan="2">Class Template (all in name-space <code>boost</code>)</th>
<th rowspan="2">Parameter Type (in name-space <code>boost</code> as needed)</th>
<th rowspan="2">Parameter Member ID</th>
<th colspan="2">Classic Equivalent</th>
<th rowspan="2" colspan="2">Template Parameter Mapping (when <code>type</code> is defined)</th>
</tr>
<tr>
<th>Signed</th>
<th>Unsigned</th>
</tr>
<tr>
<td><code>sized_integral</code></td>
<td><code>int</code></td>
<td><code>bit_count</code></td>
<td><code>int_t</code></td>
<td><code>uint_t</code></td>
<td colspan="2">The smallest built-in integral type with at least
<code>bit_count</code> bits (including the sign bit when
<var>Signedness</var> is <code>signed</code>). Not present if no
type qualifies.</td>
</tr>
<tr>
<td><code>exact_integral</code></td>
<td><code>int</code></td>
<td><code>bit_count</code></td>
<td><code>int_exact_t</code></td>
<td><code>uint_exact_t</code></td>
<td colspan="2">The smallest built-in integral type with exactly
<code>bit_count</code> bits (including the sign bit when
<var>Signedness</var> is <code>signed</code>). Not present if no
type qualifies.</td>
</tr>
<tr>
<td><code>maximum_signed_integral</code></td>
<td><code>intmax_t</code></td>
<td><code>bound</code></td>
<td colspan="2"><code>int_max_value_t</code></td>
<td>The smallest built-in integral type that can perserve the value in
<code>bound</code>. Not present if <code>bound</code> is non-positive.</td>
<td rowspan="3">It is possible for a <code>type</code> to be absent if
a platform supports really-extended integral types (beyond <code>long
long</code> or <code>__int64</code>), support for those types goes
into &lt;<a href="../../boost/cstdint.hpp">boost/cstdint.hpp</a>&gt;,
but said support hadn't yet been added to &lt;<a
href="../../boost/integer.hpp">boost/integer.hpp</a>&gt;</td>
</tr>
<tr>
<td><code>minimum_signed_integral</code></td>
<td><code>intmax_t</code></td>
<td><code>bound</code></td>
<td colspan="2"><code>int_min_value_t</code></td>
<td>The smallest built-in integral type that can perserve the value in
<code>bound</code>. Not present if <code>bound</code> is non-negative.</td>
</tr>
<tr>
<td><code>maximum_unsigned_integral</code></td>
<td><code>uintmax_t</code></td>
<td><code>bound</code></td>
<td colspan="2"><code>uint_value_t</code></td>
<td>The smallest built-in integral type that can perserve the value in
<code>bound</code>. Should always be present.</td>
</tr>
</table>
<h2><a name="example">Example</a></h2>
<blockquote><pre>
#include &lt;<a href="../../boost/integer.hpp">boost/integer.hpp</a>&gt;
#include &lt;<a href="../../boost/mpl/int.hpp">boost/mpl/int.hpp</a>&gt;
#include &lt;iostream&gt;
#include &lt;ostream&gt;
//...
template &lt; int Bits &gt;
bool
fit_exactly( boost::mpl::int_&lt;Bits&gt; const &amp;x,
typename boost::exact_integral&lt;Bits, signed&gt;::type *unused = 0 )
{
return true;
}
template &lt; typename T &gt;
bool
fit_exactly( T const &amp;x )
{
return false;
}
//...
int main()
{
typedef boost::mpl::int_&lt;24&gt; twenty_four;
boost::int_t&lt;twenty_four::value&gt;::least my_var;
//...
std::cout &lt;&lt; &quot;my_var &quot; &lt;&lt; ( fit_exactly(twenty_four()) ? &quot;does&quot; :
&quot;does not&quot; ) &lt;&lt; &quot; fit its type exactly.&quot; &lt;&lt; std::endl;
//...
}
</pre></blockquote>
<h2><a name="demo">Demonstration Program</a></h2>
<p>The program <a href="integer_test.cpp">integer_test.cpp</a> is a
simplistic demonstration of the results from instantiating various
examples of the sized type class templates.</p>
<h2><a name="rationale">Rationale</a></h2>
<p>The rationale for the design of the templates in this header includes:</p>
<ul>
<li>Avoid recursion because of concern about C++'s limited
guaranteed recursion depth (17).</li>
<li>Avoid macros on general principles.</li>
<li>Try to keep the design as simple as possible.</li>
</ul>
<h2><a name="alternative">Alternative</a></h2>
<p>If the number of bits required is known beforehand, it may be more
appropriate to use the types supplied in <cite><a
href="../../boost/cstdint.hpp">&lt;boost/cstdint.hpp&gt;</a></cite>.</p>
<h2><a name="credits">Credits</a></h2>
<p>The author of most of the Boost integer type choosing templates is <a
href="http://www.boost.org/people/beman_dawes.html">Beman Dawes</a>. He gives thanks
to Valentin Bonnard and
<a href="http://www.boost.org/people/kevlin_henney.htm"> Kevlin Henney</a> for sharing
their designs for similar templates. <a
href="http://www.boost.org/people/daryle_walker.html">Daryle Walker</a> designed the
exact and value-based sized templates, and the MPL-compatible templates.</p>
<hr>
<p>Revised July 16, 2008</p>
<p>&copy; Copyright Beman Dawes 1999. Use, modification, and distribution are
subject to the Boost Software License, Version 1.0. (See accompanying file <a
href="../../LICENSE_1_0.txt">LICENSE_1_0.txt</a> or a copy at &lt;<a
href="http://www.boost.org/LICENSE_1_0.txt">http://www.boost.org/LICENSE_1_0.txt</a>&gt;.)</p>
</body>
</html>