...one of the most highly
regarded and expertly designed C++ library projects in the
world.
— Herb Sutter and Andrei
Alexandrescu, C++
Coding Standards
This tutorial is not meant to be read linearly. Its top-level structure roughly separates different concepts in the library (e.g., handling calling multiple slots, passing values to and from slots) and in each of these concepts the basic ideas are presented first and then more complex uses of the library are described later. Each of the sections is marked Beginner, Intermediate, or Advanced to help guide the reader. The Beginner sections include information that all library users should know; one can make good use of the Signals2 library after having read only the Beginner sections. The Intermediate sections build on the Beginner sections with slightly more complex uses of the library. Finally, the Advanced sections detail very advanced uses of the Signals2 library, that often require a solid working knowledge of the Beginner and Intermediate topics; most users will not need to read the Advanced sections.
The following example writes "Hello, World!" using signals and
slots. First, we create a signal sig
, a signal that
takes no arguments and has a void return value. Next, we connect
the hello
function object to the signal using the
connect
method. Finally, use the signal
sig
like a function to call the slots, which in turns
invokes HelloWorld::operator()
to print "Hello,
World!".
struct HelloWorld
{
void operator()() const
{
std::cout << "Hello, World!" << std::endl;
}
};
// Signal with no arguments and a void return value
boost::signals2::signal<void ()> sig;
// Connect a HelloWorld slot
HelloWorld hello;
sig.connect(hello);
// Call all of the slots
sig();
Calling a single slot from a signal isn't very interesting, so we can make the Hello, World program more interesting by splitting the work of printing "Hello, World!" into two completely separate slots. The first slot will print "Hello" and may look like this:
struct Hello
{
void operator()() const
{
std::cout << "Hello";
}
};
The second slot will print ", World!" and a newline, to complete the program. The second slot may look like this:
struct World
{
void operator()() const
{
std::cout << ", World!" << std::endl;
}
};
Like in our previous example, we can create a signal
sig
that takes no arguments and has a
void
return value. This time, we connect both a
hello
and a world
slot to the same
signal, and when we call the signal both slots will be called.
boost::signals2::signal<void ()> sig;
sig.connect(Hello());
sig.connect(World());
sig();
By default, slots are pushed onto the back of the slot list, so the output of this program will be as expected:
Hello, World!
Slots are free to have side effects, and that can mean that some
slots will have to be called before others even if they are not connected in that order. The Boost.Signals2
library allows slots to be placed into groups that are ordered in
some way. For our Hello, World program, we want "Hello" to be
printed before ", World!", so we put "Hello" into a group that must
be executed before the group that ", World!" is in. To do this, we
can supply an extra parameter at the beginning of the
connect
call that specifies the group. Group values
are, by default, int
s, and are ordered by the integer
< relation. Here's how we construct Hello, World:
boost::signals2::signal<void ()> sig;
sig.connect(1, World()); // connect with group 1
sig.connect(0, Hello()); // connect with group 0
Invoking the signal will correctly print "Hello, World!", because the
Hello
object is in group 0, which precedes group 1 where
the World
object resides. The group
parameter is, in fact, optional. We omitted it in the first Hello,
World example because it was unnecessary when all of the slots are
independent. So what happens if we mix calls to connect that use the
group parameter and those that don't? The "unnamed" slots (i.e., those
that have been connected without specifying a group name) can be
placed at the front or back of the slot list (by passing
boost::signals2::at_front
or boost::signals2::at_back
as the last parameter to connect
, respectively),
and default to the end of the list. When
a group is specified, the final at_front
or at_back
parameter describes where the slot
will be placed within the group ordering. Ungrouped slots connected with
at_front
will always precede all grouped slots. Ungrouped
slots connected with at_back
will always succeed all
grouped slots.
If we add a new slot to our example like this:
struct GoodMorning
{
void operator()() const
{
std::cout << "... and good morning!" << std::endl;
}
};
// by default slots are connected at the end of the slot list
sig.connect(GoodMorning());
// slots are invoked this order:
// 1) ungrouped slots connected with boost::signals2::at_front
// 2) grouped slots according to ordering of their groups
// 3) ungrouped slots connected with boost::signals2::at_back
sig();
... we will get the result we wanted:
Hello, World! ... and good morning!
Signals can propagate arguments to each of the slots they call. For instance, a signal that propagates mouse motion events might want to pass along the new mouse coordinates and whether the mouse buttons are pressed.
As an example, we'll create a signal that passes two
float
arguments to its slots. Then we'll create a few
slots that print the results of various arithmetic operations on
these values.
void print_args(float x, float y)
{
std::cout << "The arguments are " << x << " and " << y << std::endl;
}
void print_sum(float x, float y)
{
std::cout << "The sum is " << x + y << std::endl;
}
void print_product(float x, float y)
{
std::cout << "The product is " << x * y << std::endl;
}
void print_difference(float x, float y)
{
std::cout << "The difference is " << x - y << std::endl;
}
void print_quotient(float x, float y)
{
std::cout << "The quotient is " << x / y << std::endl;
}
boost::signals2::signal<void (float, float)> sig;
sig.connect(&print_args);
sig.connect(&print_sum);
sig.connect(&print_product);
sig.connect(&print_difference);
sig.connect(&print_quotient);
sig(5., 3.);
This program will print out the following:
The arguments are 5 and 3 The sum is 8 The product is 15 The difference is 2 The quotient is 1.66667
So any values that are given to sig
when it is
called like a function are passed to each of the slots. We have to
declare the types of these values up front when we create the
signal. The type boost::signals2::signal<void (float,
float)>
means that the signal has a void
return value and takes two float
values. Any slot
connected to sig
must therefore be able to take two
float
values.
Just as slots can receive arguments, they can also return values. These values can then be returned back to the caller of the signal through a combiner. The combiner is a mechanism that can take the results of calling slots (there may be no results or a hundred; we don't know until the program runs) and coalesces them into a single result to be returned to the caller. The single result is often a simple function of the results of the slot calls: the result of the last slot call, the maximum value returned by any slot, or a container of all of the results are some possibilities.
We can modify our previous arithmetic operations example slightly so that the slots all return the results of computing the product, quotient, sum, or difference. Then the signal itself can return a value based on these results to be printed:
float product(float x, float y) { return x * y; }
float quotient(float x, float y) { return x / y; }
float sum(float x, float y) { return x + y; }
float difference(float x, float y) { return x - y; }
boost::signals2::signal<float (float, float)> sig;
sig.connect(&product);
sig.connect("ient);
sig.connect(&sum);
sig.connect(&difference);
// The default combiner returns a boost::optional containing the return
// value of the last slot in the slot list, in this case the
// difference function.
std::cout << *sig(5, 3) << std::endl;
This example program will output 2
. This is because the
default behavior of a signal that has a return type
(float
, the first template argument given to the
boost::signals2::signal
class template) is to call all slots and
then return a boost::optional
containing
the result returned by the last slot called. This
behavior is admittedly silly for this example, because slots have
no side effects and the result is the last slot connected.
A more interesting signal result would be the maximum of the values returned by any slot. To do this, we create a custom combiner that looks like this:
// combiner which returns the maximum value returned by all slots
template<typename T>
struct maximum
{
typedef T result_type;
template<typename InputIterator>
T operator()(InputIterator first, InputIterator last) const
{
// If there are no slots to call, just return the
// default-constructed value
if(first == last ) return T();
T max_value = *first++;
while (first != last) {
if (max_value < *first)
max_value = *first;
++first;
}
return max_value;
}
};
The maximum
class template acts as a function
object. Its result type is given by its template parameter, and
this is the type it expects to be computing the maximum based on
(e.g., maximum<float>
would find the maximum
float
in a sequence of float
s). When a
maximum
object is invoked, it is given an input
iterator sequence [first, last)
that includes the
results of calling all of the slots. maximum
uses this
input iterator sequence to calculate the maximum element, and
returns that maximum value.
We actually use this new function object type by installing it as a combiner for our signal. The combiner template argument follows the signal's calling signature:
boost::signals2::signal
<float (float x, float y),
maximum<float> > sig;
Now we can connect slots that perform arithmetic functions and use the signal:
sig.connect(&product);
sig.connect("ient);
sig.connect(&sum);
sig.connect(&difference);
// Outputs the maximum value returned by the connected slots, in this case
// 15 from the product function.
std::cout << "maximum: " << sig(5, 3) << std::endl;
The output of this program will be 15
, because
regardless of the order in which the slots are connected, the product
of 5 and 3 will be larger than the quotient, sum, or
difference.
In other cases we might want to return all of the values computed by the slots together, in one large data structure. This is easily done with a different combiner:
// aggregate_values is a combiner which places all the values returned
// from slots into a container
template<typename Container>
struct aggregate_values
{
typedef Container result_type;
template<typename InputIterator>
Container operator()(InputIterator first, InputIterator last) const
{
Container values;
while(first != last) {
values.push_back(*first);
++first;
}
return values;
}
};
Again, we can create a signal with this new combiner:
boost::signals2::signal
<float (float, float),
aggregate_values<std::vector<float> > > sig;
sig.connect("ient);
sig.connect(&product);
sig.connect(&sum);
sig.connect(&difference);
std::vector<float> results = sig(5, 3);
std::cout << "aggregate values: ";
std::copy(results.begin(), results.end(),
std::ostream_iterator<float>(std::cout, " "));
std::cout << "\n";
The output of this program will contain 15, 8, 1.6667, and 2. It
is interesting here that
the first template argument for the signal
class,
float
, is not actually the return type of the signal.
Instead, it is the return type used by the connected slots and will
also be the value_type
of the input iterators passed
to the combiner. The combiner itself is a function object and its
result_type
member type becomes the return type of the
signal.
The input iterators passed to the combiner transform dereference
operations into slot calls. Combiners therefore have the option to
invoke only some slots until some particular criterion is met. For
instance, in a distributed computing system, the combiner may ask
each remote system whether it will handle the request. Only one
remote system needs to handle a particular request, so after a
remote system accepts the work we do not want to ask any other
remote systems to perform the same task. Such a combiner need only
check the value returned when dereferencing the iterator, and
return when the value is acceptable. The following combiner returns
the first non-NULL pointer to a FulfilledRequest
data
structure, without asking any later slots to fulfill the
request:
struct DistributeRequest { typedef FulfilledRequest* result_type; template<typename InputIterator> result_type operator()(InputIterator first, InputIterator last) const { while (first != last) { if (result_type fulfilled = *first) return fulfilled; ++first; } return 0; } };
Slots aren't expected to exist indefinitely after they are connected. Often slots are only used to receive a few events and are then disconnected, and the programmer needs control to decide when a slot should no longer be connected.
The entry point for managing connections explicitly is the
boost::signals2::connection
class. The
connection
class uniquely represents the connection
between a particular signal and a particular slot. The
connected()
method checks if the signal and slot are
still connected, and the disconnect()
method
disconnects the signal and slot if they are connected before it is
called. Each call to the signal's connect()
method
returns a connection object, which can be used to determine if the
connection still exists or to disconnect the signal and slot.
boost::signals2::connection c = sig.connect(HelloWorld());
std::cout << "c is connected\n";
sig(); // Prints "Hello, World!"
c.disconnect(); // Disconnect the HelloWorld object
std::cout << "c is disconnected\n";
sig(); // Does nothing: there are no connected slots
Slots can be temporarily "blocked", meaning that they will be
ignored when the signal is invoked but have not been permanently disconnected.
This is typically used to prevent infinite recursion in cases where
otherwise running a slot would cause the signal it is connected to to be
invoked again. A
boost::signals2::shared_connection_block
object will
temporarily block a slot. The connection is unblocked by either
destroying or calling
unblock
on all the
shared_connection_block
objects that reference the connection.
Here is an example of
blocking/unblocking slots:
boost::signals2::connection c = sig.connect(HelloWorld());
std::cout << "c is not blocked.\n";
sig(); // Prints "Hello, World!"
{
boost::signals2::shared_connection_block block(c); // block the slot
std::cout << "c is blocked.\n";
sig(); // No output: the slot is blocked
} // shared_connection_block going out of scope unblocks the slot
std::cout << "c is not blocked.\n";
sig(); // Prints "Hello, World!"}
The boost::signals2::scoped_connection
class
references a signal/slot connection that will be disconnected when
the scoped_connection
class goes out of scope. This
ability is useful when a connection need only be temporary,
e.g.,
{
boost::signals2::scoped_connection c(sig.connect(ShortLived()));
sig(); // will call ShortLived function object
} // scoped_connection goes out of scope and disconnects
sig(); // ShortLived function object no longer connected to sig
Note, attempts to initialize a scoped_connection with the assignment syntax
will fail due to it being noncopyable. Either the explicit initialization syntax
or default construction followed by assignment from a signals2::connection
will work:
// doesn't compile due to compiler attempting to copy a temporary scoped_connection object // boost::signals2::scoped_connection c0 = sig.connect
(ShortLived()); // okay boost::signals2::scoped_connection c1(sig.connect
(ShortLived())); // also okay boost::signals2::scoped_connection c2; c2 = sig.connect
(ShortLived());
One can disconnect slots that are equivalent to a given function
object using a form of the
signal::disconnect
method, so long as
the type of the function object has an accessible ==
operator. For instance:
void foo() { std::cout << "foo"; }
void bar() { std::cout << "bar\n"; }
boost::signals2::signal
<void ()> sig;
sig.connect(&foo);
sig.connect(&bar);
sig();
// disconnects foo, but not bar
sig.disconnect(&foo);
sig();
Boost.Signals2 can automatically track the lifetime of objects involved in signal/slot connections, including automatic disconnection of slots when objects involved in the slot call are destroyed. For instance, consider a simple news delivery service, where clients connect to a news provider that then sends news to all connected clients as information arrives. The news delivery service may be constructed like this:
class NewsItem { /* ... */ }; typedef boost::signals2::signal<void (const NewsItem&)> signal_type; signal_type deliverNews;
Clients that wish to receive news updates need only connect a
function object that can receive news items to the
deliverNews
signal. For instance, we may have a
special message area in our application specifically for news,
e.g.,:
struct NewsMessageArea : public MessageArea
{
public:
// ...
void displayNews(const NewsItem& news) const
{
messageText = news.text();
update();
}
};
// ...
NewsMessageArea *newsMessageArea = new NewsMessageArea(/* ... */);
// ...
deliverNews.connect
(boost::bind(&NewsMessageArea::displayNews,
newsMessageArea, _1));
However, what if the user closes the news message area,
destroying the newsMessageArea
object that
deliverNews
knows about? Most likely, a segmentation
fault will occur. However, with Boost.Signals2 one may track any object
which is managed by a shared_ptr, by using
slot::track
. A slot will automatically
disconnect when any of its tracked objects expire. In
addition, Boost.Signals2 will ensure that no tracked object expires
while the slot it is associated with is in mid-execution. It does so by creating
temporary shared_ptr copies of the slot's tracked objects before executing it.
To track NewsMessageArea
, we use a shared_ptr to manage
its lifetime, and pass the shared_ptr to the slot via its
slot::track
method before connecting it,
e.g.:
// ...
boost::shared_ptr<NewsMessageArea> newsMessageArea(new NewsMessageArea(/* ... */));
// ...
deliverNews.connect
(signal_type::slot_type(&NewsMessageArea::displayNews,
newsMessageArea.get(), _1).track(newsMessageArea));
Note there is no explicit call to bind() needed in the above example. If the
signals2::slot
constructor is passed more than one
argument, it will automatically pass all the arguments to bind
and use the
returned function object.
Also note, we pass an ordinary pointer as the
second argument to the slot constructor, using newsMessageArea.get()
instead of passing the shared_ptr
itself. If we had passed the
newsMessageArea
itself, a copy of the shared_ptr
would
have been bound into the slot function, preventing the shared_ptr
from expiring. However, the use of
slot::track
implies we wish to allow the tracked object to expire, and automatically
disconnect the connection when this occurs.
shared_ptr
classes other than boost::shared_ptr
(such as std::shared_ptr
) may also be tracked for connection management
purposes. They are supported by the slot::track_foreign
method.
One limitation of using shared_ptr
for tracking is that
an object cannot setup tracking of itself in its constructor. However, it is
possible to set up tracking in a post-constructor which is called after the
object has been created and passed to a shared_ptr
.
The Boost.Signals2
library provides support for post-constructors and pre-destructors
via the deconstruct()
factory function.
For most cases, the simplest and most robust way to setup postconstructors
for a class is to define an associated adl_postconstruct
function
which can be found by deconstruct()
,
make the class' constructors private, and give deconstruct
access to the private constructors by declaring deconstruct_access
a friend. This will ensure that objects of the class may only be created
through the deconstruct()
function, and their
associated adl_postconstruct()
function will always be called.
The examples section
contains several examples of defining classes with postconstructors and
predestructors, and creating objects of these classes using
deconstruct()
Be aware that the postconstructor/predestructor support in Boost.Signals2
is in no way essential to the use of the library. The use of
deconstruct
is purely optional. One alternative is to
define static factory functions for your classes. The
factory function can create an object, pass ownership of the object to
a shared_ptr
, setup tracking for the object,
then return the shared_ptr
.
Signal/slot disconnections occur when any of these conditions occur:
The connection is explicitly disconnected via the connection's
disconnect
method directly, or indirectly via the
signal's disconnect
method, or
scoped_connection
's destructor.
An object tracked by the slot is destroyed.
The signal is destroyed.
These events can occur at any time without disrupting a signal's calling sequence. If a signal/slot connection is disconnected at any time during a signal's calling sequence, the calling sequence will still continue but will not invoke the disconnected slot. Additionally, a signal may be destroyed while it is in a calling sequence, and which case it will complete its slot call sequence but may not be accessed directly.
Signals may be invoked recursively (e.g., a signal A calls a slot B that invokes signal A...). The disconnection behavior does not change in the recursive case, except that the slot calling sequence includes slot calls for all nested invocations of the signal.
Note, even after a connection is disconnected, its's associated slot may still be in the process of executing. In other words, disconnection does not block waiting for the connection's associated slot to complete execution. This situation may occur in a multi-threaded environment if the disconnection occurs concurrently with signal invocation, or in a single-threaded environment if a slot disconnects itself.
Slots in the Boost.Signals2 library are created from arbitrary
function objects, and therefore have no fixed type. However, it is
commonplace to require that slots be passed through interfaces that
cannot be templates. Slots can be passed via the
slot_type
for each particular signal type and any
function object compatible with the signature of the signal can be
passed to a slot_type
parameter. For instance:
// a pretend GUI button
class Button
{
typedef boost::signals2::signal<void (int x, int y)> OnClick;
public:
typedef OnClick::slot_type OnClickSlotType;
// forward slots through Button interface to its private signal
boost::signals2::connection doOnClick(const OnClickSlotType & slot);
// simulate user clicking on GUI button at coordinates 52, 38
void simulateClick();
private:
OnClick onClick;
};
boost::signals2::connection Button::doOnClick(const OnClickSlotType & slot)
{
return onClick.connect(slot);
}
void Button::simulateClick()
{
onClick(52, 38);
}
void printCoordinates(long x, long y)
{
std::cout << "(" << x << ", " << y << ")\n";
}
Button button;
button.doOnClick(&printCoordinates);
button.simulateClick();
The doOnClick
method is now functionally equivalent
to the connect
method of the onClick
signal, but the details of the doOnClick
method can be
hidden in an implementation detail file.
Signals can be used to implement flexible Document-View
architectures. The document will contain a signal to which each of
the views can connect. The following Document
class
defines a simple text document that supports mulitple views. Note
that it stores a single signal to which all of the views will be
connected.
class Document
{
public:
typedef boost::signals2::signal<void ()> signal_t;
public:
Document()
{}
/* Connect a slot to the signal which will be emitted whenever
text is appended to the document. */
boost::signals2::connection connect(const signal_t::slot_type &subscriber)
{
return m_sig.connect(subscriber);
}
void append(const char* s)
{
m_text += s;
m_sig();
}
const std::string& getText() const
{
return m_text;
}
private:
signal_t m_sig;
std::string m_text;
};
Next, we can begin to define views. The
following TextView
class provides a simple view of the
document text.
class TextView
{
public:
TextView(Document& doc): m_document(doc)
{
m_connection = m_document.connect(boost::bind(&TextView::refresh, this));
}
~TextView()
{
m_connection.disconnect();
}
void refresh() const
{
std::cout << "TextView: " << m_document.getText() << std::endl;
}
private:
Document& m_document;
boost::signals2::connection m_connection;
};
Alternatively, we can provide a view of the document
translated into hex values using the HexView
view:
class HexView
{
public:
HexView(Document& doc): m_document(doc)
{
m_connection = m_document.connect(boost::bind(&HexView::refresh, this));
}
~HexView()
{
m_connection.disconnect();
}
void refresh() const
{
const std::string& s = m_document.getText();
std::cout << "HexView:";
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it)
std::cout << ' ' << std::hex << static_cast<int>(*it);
std::cout << std::endl;
}
private:
Document& m_document;
boost::signals2::connection m_connection;
};
To tie the example together, here is a
simple main
function that sets up two views and then
modifies the document:
int main(int argc, char* argv[])
{
Document doc;
TextView v1(doc);
HexView v2(doc);
doc.append(argc == 2 ? argv[1] : "Hello world!");
return 0;
}
The complete example source, contributed by Keith MacDonald, is available in the examples section. We also provide variations on the program which employ automatic connection management to disconnect views on their destruction.
You may encounter situations where you wish to disconnect or block a slot's connection from within the slot itself. For example, suppose you have a group of asynchronous tasks, each of which emits a signal when it completes. You wish to connect a slot to all the tasks to retrieve their results as each completes. Once a given task completes and the slot is run, the slot no longer needs to be connected to the completed task. Therefore, you may wish to clean up old connections by having the slot disconnect its invoking connection when it runs.
For a slot to disconnect (or block) its invoking connection, it must have
access to a signals2::connection
object which references
the invoking signal-slot connection. The difficulty is,
the connection
object is returned by the
signal::connect
method, and therefore is not available until after the slot is
already connected to the signal. This can be particularly troublesome
in a multi-threaded environment where the signal may be invoked
concurrently by a different thread while the slot is being connected.
Therefore, the signal classes provide
signal::connect_extended
methods, which allow slots which take an extra argument to be connected to a signal.
The extra argument is a signals2::connection
object which refers
to the signal-slot connection currently invoking the slot.
signal::connect_extended
uses slots of the type given by the
signal::extended_slot_type
typedef.
The examples section includes an
extended_slot
program which demonstrates the syntax for using
signal::connect_extended
.
For most cases the default type of boost::signals2::mutex
for
a signals2::signal
's Mutex
template type parameter should
be fine. If you wish to use an alternate mutex type, it must be default-constructible
and fulfill the Lockable
concept defined by the Boost.Thread library.
That is, it must have lock()
and unlock()
methods
(the Lockable
concept also includes a try_lock()
method
but this library does not require try locking).
The Boost.Signals2 library provides one alternate mutex class for use with signal
:
boost::signals2::dummy_mutex
. This is a fake mutex for
use in single-threaded programs, where locking a real mutex would be useless
overhead. Other mutex types you could use with signal
include
boost::mutex
, or the std::mutex
from
C++0x.
Changing a signal's Mutex
template type parameter can be tedious, due to
the large number of template parameters which precede it. The
signal_type
metafunction is particularly useful in this case,
since it enables named template type parameters for the signals2::signal
class. For example, to declare a signal which takes an int
as
an argument and uses a boost::signals2::dummy_mutex
for its Mutex
types, you could write:
namespace bs2 = boost::signals2; using bs2::keywords; bs2::signal_type<void (int), mutex_type<bs2::dummy_mutex> >::type sig;
Last revised: June 12, 2007 at 14:01:23 -0400 |