blob: 1d688a8d57a5e54ac69f57b1651dac11e6d11c88 [file] [log] [blame]
<?xml version="1.0" encoding="utf-8"?>
<!DOCTYPE section PUBLIC "-//Boost//DTD BoostBook XML V1.0//EN"
"http://www.boost.org/tools/boostbook/dtd/boostbook.dtd">
<!--
Copyright Douglas Gregor 2001-2004
Copyright Frank Mori Hess 2007-2009
Distributed under the Boost Software License, Version 1.0. (See accompanying
file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
-->
<section last-revision="$Date: 2007-06-12 14:01:23 -0400 (Tue, 12 Jun 2007) $" id="signals2.rationale">
<title>Design Rationale</title>
<using-namespace name="boost::signals2"/>
<using-namespace name="boost"/>
<using-class name="boost::signals2::signal"/>
<section>
<title>User-level Connection Management</title>
<para> Users need to have fine control over the connection of
signals to slots and their eventual disconnection. The primary approach
taken by Boost.Signals2 is to return a
<code><classname>signals2::connection</classname></code> object that enables
connected/disconnected query, manual disconnection, and an
automatic disconnection on destruction mode (<classname>signals2::scoped_connection</classname>).
In addition, two other interfaces are supported by the
<methodname alt="signal::disconnect">signal::disconnect</methodname> overloaded method:</para>
<itemizedlist>
<listitem>
<para><emphasis role="bold">Pass slot to
disconnect</emphasis>: in this interface model, the
disconnection of a slot connected with
<code>sig.<methodname>connect</methodname>(typeof(sig)::slot_type(slot_func))</code> is
performed via
<code>sig.<methodname>disconnect</methodname>(slot_func)</code>. Internally,
a linear search using slot comparison is performed and the
slot, if found, is removed from the list. Unfortunately,
querying connectedness ends up as a
linear-time operation.</para>
</listitem>
<listitem>
<para><emphasis role="bold">Pass a token to
disconnect</emphasis>: this approach identifies slots with a
token that is easily comparable (e.g., a string), enabling
slots to be arbitrary function objects. While this approach is
essentially equivalent to the connection approach taken by Boost.Signals2,
it is possibly more error-prone for several reasons:</para>
<itemizedlist>
<listitem>
<para>Connections and disconnections must be paired, so
the problem becomes similar to the problems incurred when
pairing <code>new</code> and <code>delete</code> for
dynamic memory allocation. While errors of this sort would
not be catastrophic for a signals and slots
implementation, their detection is generally
nontrivial.</para>
</listitem>
<listitem>
<para>If tokens are not unique, two slots may have
the same name and be indistinguishable. In
environments where many connections will be made
dynamically, name generation becomes an additional task
for the user.</para>
</listitem>
</itemizedlist>
<para> This type of interface is supported in Boost.Signals2
via the slot grouping mechanism, and the overload of
<methodname alt="signal::disconnect">signal::disconnect</methodname>
which takes an argument of the signal's <code>Group</code> type.</para>
</listitem>
</itemizedlist>
</section>
<section>
<title>Automatic Connection Management</title>
<para>Automatic connection management in Signals2
depends on the use of <classname>boost::shared_ptr</classname> to
manage the lifetimes of tracked objects. This is differs from
the original Boost.Signals library, which instead relied on derivation
from the <code><classname>boost::signals::trackable</classname></code> class.
The library would be
notified of an object's destruction by the
<code><classname>boost::signals::trackable</classname></code> destructor.
</para>
<para>Unfortunately, the <code><classname>boost::signals::trackable</classname></code>
scheme cannot be made thread safe due
to destructor ordering. The destructor of an class derived from
<code><classname>boost::signals::trackable</classname></code> will always be
called before the destructor of the base <code><classname>boost::signals::trackable</classname></code>
class. However, for thread-safety the connection between the signal and object
needs to be disconnected before the object runs its destructors.
Otherwise, if an object being destroyed
in one thread is connected to a signal concurrently
invoking in another thread, the signal may call into
a partially destroyed object.
</para>
<para>We solve this problem by requiring that tracked objects be
managed by <classname>shared_ptr</classname>. Slots keep a
<classname>weak_ptr</classname> to every object the slot depends
on. Connections to a slot are disconnected when any of its tracked
<classname>weak_ptr</classname>s expire. Additionally, signals
create their own temporary <classname>shared_ptr</classname>s to
all of a slot's tracked objects prior to invoking the slot. This
insures none of the tracked objects destruct in mid-invocation.
</para>
<para>The new connection management scheme has the advantage of being
non-intrusive. Objects of any type may be tracked using the
<classname>shared_ptr</classname>/<classname>weak_ptr</classname> scheme. The old
<code><classname>boost::signals::trackable</classname></code>
scheme requires the tracked objects to be derived from the <code>trackable</code>
base class, which is not always practical when interacting
with classes from 3rd party libraries.
</para>
</section>
<section>
<title><code>optional_last_value</code> as the Default Combiner</title>
<para>
The default combiner for Boost.Signals2 has changed from the <code>last_value</code>
combiner used by default in the original Boost.Signals library.
This is because <code>last_value</code> requires that at least 1 slot be
connected to the signal when it is invoked (except for the <code>last_value&lt;void&gt;</code> specialization).
In a multi-threaded environment where signal invocations and slot connections
and disconnections may be happening concurrently, it is difficult
to fulfill this requirement. When using <classname>optional_last_value</classname>,
there is no requirement for slots to be connected when a signal
is invoked, since in that case the combiner may simply return an empty
<classname>boost::optional</classname>.
</para>
</section>
<section>
<title>Combiner Interface</title>
<para> The Combiner interface was chosen to mimic a call to an
algorithm in the C++ standard library. It is felt that by viewing
slot call results as merely a sequence of values accessed by input
iterators, the combiner interface would be most natural to a
proficient C++ programmer. Competing interface design generally
required the combiners to be constructed to conform to an
interface that would be customized for (and limited to) the
Signals2 library. While these interfaces are generally enable more
straighforward implementation of the signals &amp; slots
libraries, the combiners are unfortunately not reusable (either in
other signals &amp; slots libraries or within other generic
algorithms), and the learning curve is steepened slightly to learn
the specific combiner interface.</para>
<para> The Signals2 formulation of combiners is based on the
combiner using the "pull" mode of communication, instead of the
more complex "push" mechanism. With a "pull" mechanism, the
combiner's state can be kept on the stack and in the program
counter, because whenever new data is required (i.e., calling the
next slot to retrieve its return value), there is a simple
interface to retrieve that data immediately and without returning
from the combiner's code. Contrast this with the "push" mechanism,
where the combiner must keep all state in class members because
the combiner's routines will be invoked for each signal
called. Compare, for example, a combiner that returns the maximum
element from calling the slots. If the maximum element ever
exceeds 100, no more slots are to be called.</para>
<informaltable>
<tgroup cols="2" align="left">
<thead>
<row>
<entry><para>Pull</para></entry>
<entry><para>Push</para></entry>
</row>
</thead>
<tbody>
<row>
<entry>
<programlisting>
struct pull_max {
typedef int result_type;
template&lt;typename InputIterator&gt;
result_type operator()(InputIterator first,
InputIterator last)
{
if (first == last)
throw std::runtime_error("Empty!");
int max_value = *first++;
while(first != last &amp;&amp; *first &lt;= 100) {
if (*first &gt; max_value)
max_value = *first;
++first;
}
return max_value;
}
};
</programlisting>
</entry>
<entry>
<programlisting>
struct push_max {
typedef int result_type;
push_max() : max_value(), got_first(false) {}
// returns false when we want to stop
bool operator()(int result) {
if (result &gt; 100)
return false;
if (!got_first) {
got_first = true;
max_value = result;
return true;
}
if (result &gt; max_value)
max_value = result;
return true;
}
int get_value() const
{
if (!got_first)
throw std::runtime_error("Empty!");
return max_value;
}
private:
int max_value;
bool got_first;
};
</programlisting>
</entry>
</row>
</tbody>
</tgroup>
</informaltable>
<para>There are several points to note in these examples. The
"pull" version is a reusable function object that is based on an
input iterator sequence with an integer <code>value_type</code>,
and is very straightforward in design. The "push" model, on the
other hand, relies on an interface specific to the caller and is
not generally reusable. It also requires extra state values to
determine, for instance, if any elements have been
received. Though code quality and ease-of-use is generally
subjective, the "pull" model is clearly shorter and more reusable
and will often be construed as easier to write and understand,
even outside the context of a signals &amp; slots library.</para>
<para> The cost of the "pull" combiner interface is paid in the
implementation of the Signals2 library itself. To correctly handle
slot disconnections during calls (e.g., when the dereference
operator is invoked), one must construct the iterator to skip over
disconnected slots. Additionally, the iterator must carry with it
the set of arguments to pass to each slot (although a reference to
a structure containing those arguments suffices), and must cache
the result of calling the slot so that multiple dereferences don't
result in multiple calls. This apparently requires a large degree
of overhead, though if one considers the entire process of
invoking slots one sees that the overhead is nearly equivalent to
that in the "push" model, but we have inverted the control
structures to make iteration and dereference complex (instead of
making combiner state-finding complex).</para>
</section>
<section>
<title>Connection Interfaces: += operator</title>
<para> Boost.Signals2 supports a connection syntax with the form
<code>sig.<methodname>connect</methodname>(slot)</code>, but a
more terse syntax <code>sig += slot</code> has been suggested (and
has been used by other signals &amp; slots implementations). There
are several reasons as to why this syntax has been
rejected:</para>
<itemizedlist>
<listitem>
<para><emphasis role="bold">It's unnecessary</emphasis>: the
connection syntax supplied by Boost.Signals2 is no less
powerful that that supplied by the <code>+=</code>
operator. The savings in typing (<code>connect()</code>
vs. <code>+=</code>) is essentially negligible. Furthermore,
one could argue that calling <code>connect()</code> is more
readable than an overload of <code>+=</code>.</para>
</listitem>
<listitem>
<para><emphasis role="bold">Ambiguous return type</emphasis>:
there is an ambiguity concerning the return value of the
<code>+=</code> operation: should it be a reference to the
signal itself, to enable <code>sig += slot1 += slot2</code>,
or should it return a
<code><classname>signals2::connection</classname></code> for the
newly-created signal/slot connection?</para>
</listitem>
<listitem>
<para><emphasis role="bold">Gateway to operators -=,
+</emphasis>: when one has added a connection operator
<code>+=</code>, it seems natural to have a disconnection
operator <code>-=</code>. However, this presents problems when
the library allows arbitrary function objects to implicitly
become slots, because slots are no longer comparable. <!--
(see the discussion on this topic in User-level Connection
Management). --></para>
<para> The second obvious addition when one has
<code>operator+=</code> would be to add a <code>+</code>
operator that supports addition of multiple slots, followed by
assignment to a signal. However, this would require
implementing <code>+</code> such that it can accept any two
function objects, which is technically infeasible.</para>
</listitem>
</itemizedlist>
</section>
<section>
<title>Signals2 Mutex Classes</title>
<para>
The Boost.Signals2 library provides 2 mutex classes: <classname>boost::signals2::mutex</classname>,
and <classname>boost::signals2::dummy_mutex</classname>. The motivation for providing
<classname>boost::signals2::mutex</classname> is simply that the <classname>boost::mutex</classname>
class provided by the Boost.Thread library currently requires linking to libboost_thread.
The <classname>boost::signals2::mutex</classname> class allows Signals2 to remain
a header-only library. You may still choose to use <classname>boost::mutex</classname>
if you wish, by specifying it as the <code>Mutex</code> template type for your signals.
</para>
<para>
The <classname>boost::signals2::dummy_mutex</classname> class is provided to allow
performance sensitive single-threaded applications to minimize overhead by avoiding unneeded
mutex locking.
</para>
</section>
<section>
<title>Comparison with other Signal/Slot implementations</title>
<section>
<title>libsigc++</title>
<para> <ulink
url="http://libsigc.sourceforge.net">libsigc++</ulink> is a C++
signals &amp; slots library that originally started as part of
an initiative to wrap the C interfaces to <ulink
url="http://www.gtk.org">GTK</ulink> libraries in C++, and has
grown to be a separate library maintained by Karl Nelson. There
are many similarities between libsigc++ and Boost.Signals2, and
indeed the original Boost.Signals was strongly influenced by
Karl Nelson and libsigc++. A cursory inspection of each library will find a
similar syntax for the construction of signals and in the use of
connections. There
are some major differences in design that separate these
libraries:</para>
<itemizedlist>
<listitem>
<para><emphasis role="bold">Slot definitions</emphasis>:
slots in libsigc++ are created using a set of primitives
defined by the library. These primitives allow binding of
objects (as part of the library), explicit adaptation from
the argument and return types of the signal to the argument
and return types of the slot (libsigc++ is, by default, more
strict about types than Boost.Signals2).</para>
</listitem>
<listitem>
<para><emphasis role="bold">Combiner/Marshaller
interface</emphasis>: the equivalent to Boost.Signals2
combiners in libsigc++ are the marshallers. Marshallers are
similar to the "push" interface described in Combiner
Interface, and a proper treatment of the topic is given
there.</para>
</listitem>
</itemizedlist>
</section>
<section>
<title>.NET delegates</title>
<para> <ulink url="http://www.microsoft.com">Microsoft</ulink>
has introduced the .NET Framework and an associated set of
languages and language extensions, one of which is the
delegate. Delegates are similar to signals and slots, but they
are more limited than most C++ signals and slots implementations
in that they:</para>
<itemizedlist>
<listitem>
<para>Require exact type matches between a delegate and what
it is calling.</para>
</listitem>
<listitem><para>Only return the result of the last target called, with no option for customization.</para></listitem>
<listitem>
<para>Must call a method with <code>this</code> already
bound.</para>
</listitem>
</itemizedlist>
</section>
</section>
</section>