![]() |
Home | Libraries | People | FAQ | More |
This section lists some examples that use this library.
Combing local functions with the statement expression extension of GCC compilers, it is possible to implement lambda functions for GCC compilers even without C++11 support.
![]() |
Warning |
|---|---|
This code only works on compilers that support the statement expression GCC extension or that support C++11 lambda functions. |
For example (see also gcc_lambda.cpp
and gcc_lambda_cpp11.cpp):
|
With Local Functions (GCC only) |
C++11 Lambdas |
|---|---|
|
int val = 2; int nums[] = {1, 2, 3}; int* end = nums + 3; int* iter = std::find_if(nums, end, GCC_LAMBDA(const bind val, int num, return bool) { return num == val; } GCC_LAMBDA_END );
|
int val = 2; int nums[] = {1, 2, 3}; int* end = nums + 3; int* iter = std::find_if(nums, end, [val](int num) -> bool { return num == val; } );
|
Where the macros are defined in gcc_lambda.hpp.
This is possible because GCC statement expressions allow to use declaration
statements within expressions and therefore to declare a local function within
an expression. The macros automatically detect if the compiler supports
C++11
lambda functions in which case the implementation uses native lambdas
instead of local functions in statement expressions. However, C++11
lambda functions do not support constant binding so it is best to
only use const bind
variable (same as C++11
lambda =variable)
and bind&
variable (same as C++11
lambda &variable)
because these have the exact same semantic between the local function and
native lambda implementation. Unfortunately, the C++11
lambda short-hand binds &
and = (which automatically bind
all variables in scope either by reference or value) are not supported by
the macros because they are not supported by the local function implementation.
Finally, the result type return result-type
is optional and it is assumed void
when it is not specified (same as with C++11
lambda functions).
It is possible to use local functions to check assertions between variables that are made constant within the asserted expressions. This is advantageous because assertions are not supposed to change the state of the program and ideally the compiler will not compile assertions that modify variables.
For example, consider the following assertion where by mistake we programmed
operator=
instead of operator==:
int x = 1, y = 2; assert(x = y); // Mistakenly `=` instead of `==`.
Ideally this code will not compile instead this example not only compiles
but the assertion even passes the run-time check and no error is generated
at all. The [N1613]
paper introduces the idea of a const-block which could
be used to wrap the assertion above and catch the programming error at compile-time.
Similarly, the following code will generate a compile-time error when operator=
is mistakenly used instead of operator== because both x
and y are made constants
(using local functions) within the block of code performing the assertion
(see also const_block_error.cpp):
|
With Local Functions |
N1613 Const-Blocks |
|---|---|
|
int x = 1, y = 2; CONST_BLOCK(x, y) { // Constant block. assert(x = y); // Compiler error. } CONST_BLOCK_END
|
int x = 1, y = 2; const { // Constant block. assert(x = y); // Compiler error. }
|
Where the macros are defined in const_block.hpp.
The constant block macros are implemented using a local function which binds
by constant reference const bind&
all the specified variables (so the variables are constant within the code
block but they do not need to be CopyConstructible
and no extra copy is performed). The local function executes the assert instruction in its body which is
called immediately after it is defined. More in general, constant blocks
can be used to evaluate any instruction (not just assertions) within a block
were all specified variables are constant.
Unfortunately, constant blocks cannot be implemented with C++11
lambda functions because these do not support constant binding (of
course it is always possible to introduce extra constant variables const int& const_x
= x,
etc and use these variables in the assertion). Variables bound by value using
C++11
lambda functions (variable,
=variable,
and =) are constant but they
are required to be CopyConstructible
and they introduce potentially expensive copy operations. [27]
Scope exits allow to execute arbitrary code at the exit of the enclosing scope and they are provided by the Boost.ScopeExit library.
For curiosity, here we show how to re-implement scope exits using local functions.
One small advantage of scope exits that use local functions is that they
support constant binding. Boost.ScopeExit
does not directly support constant binding (however, it is always possible
to introduce an extra const
local variable, assign it to the value to bind, and then bind the const variable so to effectively have constant
binding with Boost.ScopeExit
as well). In general, the authors recommend to use Boost.ScopeExit
instead of the code listed here whenever possible.
The following example binds p
by constant reference so this variable cannot be modified within the scope
exit body but it is not copied and it will present the value it has at the
exit of the enclosing scope and not at the scope exit declaration (see also
scope_exit.cpp):
|
With Local Functions |
Boost.ScopeExit |
|---|---|
|
person& p = persons_.back(); person::evolution_t checkpoint = p.evolution_; SCOPE_EXIT(const bind checkpoint, const bind& p, bind this_) { if (checkpoint == p.evolution_) this_->persons_.pop_back(); } SCOPE_EXIT_END
|
person& p = persons_.back();
person::evolution_t checkpoint = p.evolution_;
BOOST_SCOPE_EXIT(checkpoint, &p, this_) { // Or extra variable
|
Where the macros are defined in scope_exit.hpp.
The scope exit macros are implemented by passing a local function when constructing an object of the following class:
struct scope_exit { scope_exit(boost::function<void (void)> f): f_(f) {} ~scope_exit(void) { f_(); } private: boost::function<void (void)> f_; };
A local variable within the enclosing scope is used to hold the object so
the destructor will be invoked at the exit of the enclosing scope and it
will in turn call the local function executing the scope exit instructions.
The scope exit local function has no parameter and void
result type but it supports binding and constant binding.
Local functions can be used to create Boost.Phoenix
functions. For example (see also phoenix_factorial_local.cpp
and phoenix_factorial.cpp):
|
Local Functions |
Global Functor |
|---|---|
|
BOOST_AUTO_TEST_CASE( test_phoenix_factorial_local ) { using boost::phoenix::arg_names::arg1; int BOOST_LOCAL_FUNCTION(int n) { // Unfortunately, monomorphic. return (n <= 0) ? 1 : n * factorial_impl(n - 1); } BOOST_LOCAL_FUNCTION_NAME(recursive factorial_impl) boost::phoenix::function< boost::function<int (int)> > factorial(factorial_impl); // Phoenix function from local function. int i = 4; BOOST_CHECK( factorial(i)() == 24 ); // Call. BOOST_CHECK( factorial(arg1)(i) == 24 ); // Lazy call. }
|
struct factorial_impl { // Phoenix function from global functor. template<typename Sig> struct result; template<typename This, typename Arg> struct result<This (Arg)> : result<This (Arg const&)> {}; template<typename This, typename Arg> struct result<This (Arg&)> { typedef Arg type; }; template<typename Arg> // Polymorphic. Arg operator()(Arg n) const { return (n <= 0) ? 1 : n * (*this)(n - 1); } }; BOOST_AUTO_TEST_CASE( test_phoenix_factorial ) { using boost::phoenix::arg_names::arg1; boost::phoenix::function<factorial_impl> factorial; int i = 4; BOOST_CHECK( factorial(i)() == 24 ); // Call. BOOST_CHECK( factorial(arg1)(i) == 24 ); // Lazy call. }
|
This is presented here mainly as a curiosity because Boost.Phoenix functions created from local functions have the important limitation that they cannot be polymorphic. [28] Therefore, in many cases creating the Boost.Phoenix function from global functors (possibly with the help of Boost.Phoenix adaptor macros) might be a more valuable option.
The following are examples of closures that illustrate how to return local functions to the calling scope (note how extra care is taken in order to ensure that all bound variables remain valid at the calling scope):
The GCC C compiler supports local functions under the name of nested functions. Nested functions are exclusively a C extension of the GCC compiler (they are not supported for C++ not even by the GCC compiler, and they are not part of any C or C++ standard, nor they are supported by other compilers like MSVC).
The following examples are taken form the GCC nested function documentation and programmed using this library:
The following examples are taken from a number of N-papers and programmed using this library.
|
Files |
Notes |
|---|---|
|
This example is adapted from [N2550]
(C++11
lambda functions): It passes a local function to the STL
algorithm |
|
|
This example is adapted from [N2529]
(C++11
lambda functions): It binds the object in scope |
[27]
Ideally, C++11
lambda functions would allow to bind variables also using const& variable (constant reference) and const&
(all variables by constant reference).