more/generic_programming.html

449 lines
18 KiB
HTML
Raw Normal View History

2001-02-12 11:17:54 +08:00
<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 3.2//EN">
<meta name="generator" content="HTML Tidy, see www.w3.org">
<meta http-equiv="Content-Type" content="text/html; charset=windows-1252">
<meta name="GENERATOR" content="Microsoft FrontPage 4.0">
<meta name="ProgId" content="FrontPage.Editor.Document">
<title>Generic Programming Techniques</title>
<img src="../c++boost.gif" alt="c++boost.gif (8819 bytes)" align="center"
width="277" height="86">
<body bgcolor="#FFFFFF" text="#000000">
2001-02-12 11:17:54 +08:00
<h1>Generic Programming Techniques</h1>
<p>This is an incomplete survey of some of the generic programming
techniques used in the <a href="../index.htm">boost</a> libraries.
<h2>Table of Contents</h2>
<ul>
<li><a href="#introduction">Introduction</a>
<li><a href="#concept">The Anatomy of a Concept</a>
2001-02-12 11:17:54 +08:00
<li><a href="#traits">Traits</a>
<li><a href="#tag_dispatching">Tag Dispatching</a>
2001-02-14 01:50:49 +08:00
<li><a href="#adaptors">Adaptors</a>
2001-02-12 11:17:54 +08:00
<li><a href="#type_generator">Type Generators</a>
<li><a href="#object_generator">Object Generators</a>
2001-02-12 11:17:54 +08:00
<li><a href="#policy">Policy Classes</a>
2001-02-12 11:17:54 +08:00
</ul>
<h2><a name="introduction">Introduction</a></h2>
<p>Generic programming is about generalizing software components so that
they can be easily reused in a wide variety of situations. In C++, class
and function templates are particularly effective mechanisms for generic
programming because they make the generalization possible without
sacrificing efficiency.
<p>As a simple example of generic programming, we will look at how one
might generalize the <tt>memcpy()</tt> function of the C standard library.
An implementation of <tt>memcpy()</tt> might look like the following:
<br>
<br>
<blockquote>
<pre>
void* memcpy(void* region1, const void* region2, size_t n)
{
const char* first = (const char*)region2;
const char* last = ((const char*)region2) + n;
char* result = (char*)region1;
while (first != last)
*result++ = *first++;
return result;
}
</pre>
</blockquote>
The <tt>memcpy()</tt> function is already generalized to some extent by the
use of <tt>void*</tt> so that the function can be used to copy arrays of
different kinds of data. But what if the data we would like to copy is not
in an array? Perhaps it is in a linked list. Can we generalize the notion
of copy to any sequence of elements? Looking at the body of
<tt>memcpy()</tt>, the function's <b><i>minimal requirements</i></b> are
that it needs to to <i>traverse</i> through the sequence using some sort of
pointer, <i>access</i> elements pointed to, <i>write</i> the elements to
the destination, and <i>compare</i> pointers to know when to stop. The C++
standard library groups requirements such as these into
<b><i>concepts</i></b>, in this case the <a href=
"http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a> concept
(for <tt>region2</tt>) and the <a href=
"http://www.sgi.com/tech/stl/OutputIterator.html">Output Iterator</a>
concept (for <tt>region1</tt>).
<p>If we rewrite the <tt>memcpy()</tt> as a function template, and use the
<a href="http://www.sgi.com/tech/stl/InputIterator.html">Input Iterator</a>
and <a href="http://www.sgi.com/tech/stl/OutputIterator.html">Output
Iterator</a> concepts to describe the requirements on the template
parameters, we can implement a highly reusable <tt>copy()</tt> function in
the following way:
<br>
<br>
<blockquote>
<pre>
template &lt;typename InputIterator, typename OutputIterator&gt;
OutputIterator
copy(InputIterator first, InputIterator last, OutputIterator result)
{
while (first != last)
*result++ = *first++;
return result;
}
</pre>
</blockquote>
<p>Using the generic <tt>copy()</tt> function, we can now copy elements
from any kind of sequence, including a linked list that exports iterators
such as <tt>std::<a href=
"http://www.sgi.com/tech/stl/List.html">list</a></tt>.
<br>
<br>
<blockquote>
<pre>
#include &lt;list&gt;
#include &lt;vector&gt;
#include &lt;iostream&gt;
int main()
{
const int N = 3;
std::vector&lt;int&gt; region1(N);
std::list&lt;int&gt; region2;
region2.push_back(1);
region2.push_back(0);
region2.push_back(3);
std::copy(region2.begin(), region2.end(), region1.begin());
for (int i = 0; i &lt; N; ++i)
std::cout &lt;&lt; region1[i] &lt;&lt; " ";
std::cout &lt;&lt; std::endl;
}
</pre>
</blockquote>
<h2><a name="concept">Anatomy of a Concept</a></h2>
A <b><i>concept</i></b> is a set requirements, where the requirements
consist of valid expressions, associated types, invariants, and complexity
guarantees. A type that satisfies the set of requirements is said to
<b><i>model</i></b> the concept. A concept can extend the requirements of
another concept, which is called <b><i>refinement</i></b>.
<ul>
<li><a name="valid_expression"><b>Valid Expressions</b></a> are C++
expressions which must compile successfully for the objects involved in
the expression to be considered <i>models</i> of the concept.
<li><a name="associated_type"><b>Associated Types</b></a> are types that
are related to the modeling type in that they participate in one or more
of the valid expressions. Typically associated types can be accessed
either through typedefs nested within a class definition for the modeling
type, or they are accessed through a <a href="#traits">traits class</a>.
<li><b>Invariants</b> are run-time characteristics of the objects that
must always be true, that is, the functions involving the objects must
preserve these characteristics. The invariants often take the form of
pre-conditions and post-conditions.
<li><b>Complexity Guarantees</b> are maximum limits on how long the
execution of one of the valid expressions will take, or how much of
various resources its computation will use.
</ul>
<p>The concepts used in the C++ Standard Library are documented at the <a
href="http://www.sgi.com/tech/stl/table_of_contents.html">SGI STL site</a>.
2001-02-12 11:17:54 +08:00
<h2><a name="traits">Traits</a></h2>
<p>A traits class provides a way of associating information with a
compile-time entity (a type, integral constant, or address). For example,
the class template <tt><a href=
2001-02-12 11:17:54 +08:00
"http://www.sgi.com/tech/stl/iterator_traits.html">std::iterator_traits&lt;T&gt;</a></tt>
looks something like this:
<blockquote>
<pre>
template &lt;class Iterator&gt;
struct iterator_traits {
typedef ... iterator_category;
typedef ... value_type;
typedef ... difference_type;
typedef ... pointer;
typedef ... reference;
};
</pre>
</blockquote>
The traits' <tt>value_type</tt> gives generic code the type which the
iterator is "pointing at", while the <tt>iterator_category</tt> can be used
to select more efficient algorithms depending on the iterator's
capabilities.
<p>A key feature of traits templates is that they're <i>non-intrusive</i>:
they allow us to associate information with arbitrary types, including
built-in types and types defined in third-party libraries, Normally, traits
are specified for a particular type by (partially) specializing the traits
template.
<p>For an in-depth description of <tt>std::iterator_traits</tt>, see <a
href="http://www.sgi.com/tech/stl/iterator_traits.html">this page</a>
provided by SGI. Another very different expression of the traits idiom in
the standard is <tt>std::numeric_limits&lt;T&gt;</tt> which provides
constants describing the range and capabilities of numeric types.
2001-02-12 11:17:54 +08:00
<h2><a name="tag_dispatching">Tag Dispatching</a></h2>
<p>A technique that often goes hand in hand with traits classes is tag
dispatching, which is a way of using function overloading to dispatch based
on properties of a type. A good example of this is the implementation of the
<a href=
"http://www.sgi.com/tech/stl/advance.html"><tt>std::advance()</tt></a>
function in the C++ Standard Library, which increments an iterator
<tt>n</tt> times. Depending on the kind of iterator, there are different
optimizations that can be applied in the implementation. If the iterator is
<a href="http://www.sgi.com/tech/stl/RandomAccessIterator.html">random
access</a> (can jump forward and backward arbitrary distances), then the
<tt>advance()</tt> function can simply be implemented with <tt>i += n</tt>,
and is very efficient: constant time. Other iterators must be
<tt>advance</tt>d in steps, making the operation linear in n. If the
iterator is <a href=
"http://www.sgi.com/tech/stl/BidirectionalIterator.html">bidirectional</a>,
then it makes sense for <tt>n</tt> to be negative, so we must decide whether
to increment or decrement the iterator.
<p>The relation between tag dispatching and traits classes is that the
property used for dispatching (in this case the <tt>iterator_category</tt>)
is often accessed through a traits class. The main <tt>advance()</tt> function
uses the <a href=
"http://www.sgi.com/tech/stl/iterator_traits.html"><tt>iterator_traits</tt></a>
class to get the <tt>iterator_category</tt>. It then makes a call the the
overloaded <tt>advance_dispatch()</tt> function. The appropriate
<tt>advance_dispatch()</tt> is selected by the compiler based on whatever
type the <tt>iterator_category</tt> resolves to, either <a href=
"http://www.sgi.com/tech/stl/input_iterator_tag.html"><tt>input_iterator_tag</tt></a>,
<a href=
"http://www.sgi.com/tech/stl/bidirectional_iterator_tag.html"><tt>bidirectional_iterator_tag</tt></a>,
or <a href=
"http://www.sgi.com/tech/stl/random_access_iterator_tag.html"><tt>random_access_iterator_tag</tt></a>.
A <b><i>tag</i></b> is simply a class whose only purpose is to convey some
property for use in tag dispatching and similar techniques. Refer to <a
href="http://www.sgi.com/tech/stl/iterator_tags.html">this page</a> for a
more detailed description of iterator tags.
<blockquote>
<pre>
namespace std {
struct input_iterator_tag { };
struct bidirectional_iterator_tag { };
struct random_access_iterator_tag { };
namespace detail {
template &lt;class InputIterator, class Distance&gt;
void advance_dispatch(InputIterator&amp; i, Distance n, <b>input_iterator_tag</b>) {
while (n--) ++i;
}
template &lt;class BidirectionalIterator, class Distance&gt;
void advance_dispatch(BidirectionalIterator&amp; i, Distance n,
<b>bidirectional_iterator_tag</b>) {
if (n &gt;= 0)
while (n--) ++i;
else
while (n++) --i;
}
template &lt;class RandomAccessIterator, class Distance&gt;
void advance_dispatch(RandomAccessIterator&amp; i, Distance n,
<b>random_access_iterator_tag</b>) {
i += n;
}
}
template &lt;class InputIterator, class Distance&gt;
void advance(InputIterator&amp; i, Distance n) {
typename <b>iterator_traits&lt;InputIterator&gt;::iterator_category</b> category;
detail::advance_dispatch(i, n, <b>category</b>);
}
}
</pre>
</blockquote>
2001-02-14 01:50:49 +08:00
<h2><a name="adaptors">Adaptors</a></h2>
<p>An <i>adaptor</i> is a class template which builds on another type or
types to provide a new interface or behavioral variant. Examples of
standard adaptors are <a href=
"http://www.sgi.com/tech/stl/ReverseIterator.html">std::reverse_iterator</a>,
which adapts an iterator type by reversing its motion upon
increment/decrement, and <a href=
"http://www.sgi.com/tech/stl/stack.html">std::stack</a>, which adapts a
container to provide a simple stack interface.
<p>A more comprehensive review of the adaptors in the standard can be found
<a href=
"http://www.cs.rpi.edu/~wiseb/xrds/ovp2-3b.html#SECTION00015000000000000000">
here</a>.
2001-02-12 11:17:54 +08:00
<h2><a name="type_generator">Type Generators</a></h2>
<p>A <i>type generator</i> is a template whose only purpose is to
synthesize a new type or types based on its template argument(s)<a href=
"#1">[1]</a>. The generated type is usually expressed as a nested typedef
named, appropriately <tt>type</tt>. A type generator is usually used to
consolidate a complicated type expression into a simple one, as in
<tt>boost::<a href=
"../libs/utility/filter_iterator.htm">filter_iterator_generator</a></tt>,
2001-02-12 11:17:54 +08:00
which looks something like this:
<blockquote>
<pre>
template &lt;class Predicate, class Iterator,
class Value = <i>complicated default</i>,
class Reference = <i>complicated default</i>,
class Pointer = <i>complicated default</i>,
class Category = <i>complicated default</i>,
class Distance = <i>complicated default</i>
&gt;
struct filter_iterator_generator {
typedef iterator_adaptor&lt;
Iterator,filter_iterator_policies&lt;Predicate,Iterator&gt;,
Value,Reference,Pointer,Category,Distance&gt; <b>type</b>;
};
</pre>
</blockquote>
<p>Now, that's complicated, but producing an adapted filter iterator is
much easier. You can usually just write:
<blockquote>
<pre>
boost::filter_iterator_generator&lt;my_predicate,my_base_iterator&gt;::type
</pre>
</blockquote>
<h2><a name="object_generator">Object Generators</a></h2>
<p>An <i>object generator</i> is a function template whose only purpose is
to construct a new object out of its arguments. Think of it as a kind of
generic constructor. An object generator may be more useful than a plain
constructor when the exact type to be generated is difficult or impossible
to express and the result of the generator can be passed directly to a
function rather than stored in a variable. Most Boost object generators are
named with the prefix "<tt>make_</tt>", after <tt>std::<a href=
2001-02-12 11:17:54 +08:00
"http://www.sgi.com/tech/stl/pair.html">make_pair</a>(const<73>T&amp;,<2C>const<73>U&amp;)</tt>.
<p>For example, given:
2001-02-12 11:17:54 +08:00
<blockquote>
<pre>
struct widget {
void tweak(int);
};
std::vector&lt;widget *&gt; widget_ptrs;
</pre>
</blockquote>
By chaining two standard object generators, <tt>std::<a href=
"http://www.dinkumware.com/htm_cpl/functio2.html#bind2nd">bind2nd</a>()</tt>
and <tt>std::<a href=
"http://www.dinkumware.com/htm_cpl/functio2.html#mem_fun">mem_fun</a>()</tt>,
we can easily tweak all widgets:
<blockquote>
<pre>
void tweak_all_widgets1(int arg)
2001-02-12 11:17:54 +08:00
{
for_each(widget_ptrs.begin(), widget_ptrs.end(),
<b>bind2nd</b>(std::<b>mem_fun</b>(&amp;widget::tweak), arg));
2001-02-12 11:17:54 +08:00
}
</pre>
</blockquote>
<p>Without using object generators the example above would look like this:
2001-02-12 11:17:54 +08:00
<blockquote>
<pre>
void tweak_all_widgets2(int arg)
2001-02-12 11:17:54 +08:00
{
for_each(struct_ptrs.begin(), struct_ptrs.end(),
<b>std::binder2nd&lt;std::mem_fun1_t&lt;void, widget, int&gt; &gt;</b>(
std::<b>mem_fun1_t&lt;void, widget, int&gt;</b>(&amp;widget::tweak), arg));
2001-02-12 11:17:54 +08:00
}
</pre>
</blockquote>
<p>As expressions get more complicated the need to reduce the verbosity of
type specification gets more compelling.
<h2><a name="policy">Policy Classes</a></h2>
2001-02-12 11:17:54 +08:00
<p>A policy class is a template parameter used to transmit behavior. An
example from the standard library is <tt>std::<a href=
"http://www.dinkumware.com/htm_cpl/memory.html#allocator">allocator</a></tt>,
which supplies memory management behaviors to standard <a href=
"http://www.sgi.com/tech/stl/Container.html">containers</a>.
<p>Policy classes have been explored in detail by <a href=
"mailto:andrewalex@hotmail.com">Andrei Alexandrescu</a> in <a href=
2001-02-12 11:17:54 +08:00
"http://www.cs.ualberta.ca/~hoover/cmput401/XP-Notes/xp-conf/Papers/7_3_Alexandrescu.pdf">
this paper</a>. He writes:
<blockquote>
<p>Policy classes are implementations of punctual design choices. They
are inherited from, or contained within, other classes. They provide
different strategies under the same syntactic interface. A class using
policies is templated having one template parameter for each policy it
uses. This allows the user to select the policies needed.
<p>The power of policy classes comes from their ability to combine
freely. By combining several policy classes in a template class with
multiple parameters, one achieves combinatorial behaviors with a linear
amount of code.
</blockquote>
<p>Andrei's description of policy classes describe their power as being
derived from their granularity and orthogonality. Boost has probably
diluted the distinction in the <a href=
"../libs/utility/iterator_adaptors.htm">Iterator Adaptors</a> library,
where we transmit all of an adapted iterator's behavior in a single policy
class. There is precedent for this, however: <tt><a href=
"http://www.dinkumware.com/htm_cpl/string2.html#char_traits">std::char_traits</a></tt>,
despite its name, acts as a policies class that determines the behaviors of
<a href=
"http://www.dinkumware.com/htm_cpl/string2.html#basic_string">std::basic_string</a>.
2001-02-12 11:17:54 +08:00
<h2>Notes</h2>
<a name="1">[1]</a> Type generators are a workaround for the lack of
``templated typedefs'' in C++.
<hr>
<p>Revised
<!--webbot bot="Timestamp" s-type="EDITED" s-format="%d %b %Y" startspan -->14 Mar 2001<!--webbot bot="Timestamp" endspan i-checksum="14885" -->
<p>&copy; Copyright David Abrahams 2001. Permission to copy, use, modify,
sell and distribute this document is granted provided this copyright notice
appears in all copies. This document is provided "as is" without express or
implied warranty, and with no claim as to its suitability for any purpose.
<!-- LocalWords: HTML html charset gif alt htm struct SGI namespace std libs
-->
<!-- LocalWords: InputIterator BidirectionalIterator RandomAccessIterator pdf
-->
<!-- LocalWords: typename Alexandrescu templated Andrei's Abrahams memcpy int
-->
<!-- LocalWords: const OutputIterator iostream pre cpl
-->
</body>