cpp-netlib includes and implements a number of HTTP clients that you can use and embed in your own applications. All of the HTTP client implementations:
- Cannot be copied. This means you may have to store instances of the clients in dynamic memory if you intend to use them as function parameters or pass them around in smart pointers or by reference.
- Assume that requests made are independent of each other. There currently is no cookie or session management system built-in to cpp-netlib’s HTTP client implementations.
The HTTP clients all share the same API, but the internals are documented in terms of what is different and what to expect with the different implementations.
As of 0.9.1 the default implementation for the cpp-netlib HTTP client is asynchronous.
As of 0.11 the Synchronous Clients are now DEPRECATED and will be removed in subsequent releases.
There is a single user-facing template class named basic_client which takes three template parameters:
- Tag - which static tag you choose that defines the behavior of the client.
- http_version_major - an unsigned int that defines the HTTP major version number, this directly affects the HTTP messages sent by the client.
- http_version_minor - an unsigned int that defines the HTTP minor version number.
There are two major different class of implementations of the basic_client template that depend on which tag you choose: Synchronous Clients and Asynchronous Clients. These two different classes are described in their own sections following this one. What follows is a table of all tags supported by the HTTP client implementation provided by cpp-netlib.
| Tag | Description |
|---|---|
| http_default_8bit_tcp_resolve | This is the default HTTP implementation tag that resolves addresses with a TCP resolver and provides a synchronous/blocking HTTP client interface. |
| http_default_8bit_udp_resolve | This is similar to the above tag except that it specifies the HTTP client to use a UDP resolver. It also provides a synchronous/ blocking HTTP client interface. |
| http_keepalive_8bit_tcp_resolve | This tag specifies that the HTTP client by default will keep connections to the server alive. It only makes sense if the version_major and version_minor are both 1, to indicate HTTP 1.1. This tag causes the HTTP client to resolve using a TCP resolver and provides a synchronous/ blocking HTTP client interface. |
| http_keepalive_8bit_udp_resolve | This is similar to the above tag except that it specifies the HTTP client to use a UDP resolver. It also provides a synchronous/ blocking HTTP client interface. |
| http_async_8bit_tcp_resolve | This tag provides an active HTTP client object implementation that uses a TCP resolver. Response objects returned will encapsulate a number of Boost.Thread shared futures to hold values. Users don’t have to see this as they are implementation details. |
| http_async_8bit_udp_resolve | This is similar to the above tag except that specifies the HTTP client to use a UDP resolver. |
In the above table the tags follow a pattern for describing the behavior introduced by the tags. This pattern is shown below:
<protocol>_<modifier>_<character-width>_<resolve-strategy>
For example, the tag http_default_8bit_tcp_resolve indicates the protocol http, a modifier default, a character width of 8bit, and a resolve strategy of tcp_resolve.
Of the client tags shown in the table, the following makes the basic_client behave as a fully synchronous client.
- http_default_8bit_tcp_resolve
- http_default_8bit_udp_resolve
- http_keepalive_8bit_tcp_resolve
- http_keepalive_8bit_udp_resolve
The synchronous client implements all the operations of the client underneath the interface all block to wait for I/O to finish. All the member methods are synchronous and will block until the response object is ready or throws if erros are encountered in the performance of the HTTP requests.
Warning
The synchronous clients are NOT thread safe. You will need to do external synchronization to use synchronous client implementations.
Note
As of version 0.11, all the synchronous client implementations are deprecated. They will be removed in the next version of the library.
The following tags specify the basic_client to behave in an asynchronous manner:
- http_async_8bit_tcp_resolve
- http_async_8bit_udp_resolve
An asynchronous client implementation means that``basic_client<...>`` is an Active Object. This means that the client has and manages its own lifetime thread, and returns values that are asynchronously filled in. The response object encapsulates Boost.Thread futures which get filled in once the values are available.
The asynchronous clients implement all operations asynchronously which are hidden from the user. The interface is still synchronous but the fetching of data happens on a different thread.
Note
The asynchronous clients are thread safe, and can be shared across many threads. Each request starts a sequence of asynchronous operations dedicated to that request. The client does not re-cycle connections and uses a one-request-one-connection model.
When an asynchronous client object is destroyed, it waits for all pending asynchronous operations to finish. Errors encountered during operations on retrieving data from the response objects cause exceptions to be thrown – therefore it is best that if a client object is constructed, it should outlive the response object or be outside the try-catch block handling the errors from operations on responses. In code, usage should look like the following:
http::client client;
try {
http::client::response response = client.get("http://www.example.com/");
std::cout << body(response);
} catch (std::exception& e) {
// deal with exceptions here
}
A common mistake is to declare the client inside the try block which invokes undefined behavior when errors arise from the handling of response objects.
In this section we assume that the following typedef is in effect:
typedef boost::network::http::basic_client<
boost::network::http::tags::http_default_8bit_udp_resolve
, 1
,1
>
client;
Also, that code using the HTTP client will have use the following header:
#include <boost/network/include/http/client.hpp>
Note
Starting version 0.9, cpp-netlib clients and server implementations by default now have an externally-linked component. This is a breaking change for code that used to rely on cpp-netlib being a header-only library, but can inhibited by defining the BOOST_NETWORK_NO_LIB preprocessor macro before including any cpp-netlib header.
Note
Starting version 0.11, cpp-netlib clients and server implementations no longer support the BOOST_NETWORK_NO_LIB option.
The client implementation can be default constructed, or customized at initialization.
| Parameter Name | Type | Description |
|---|---|---|
| follow_redirects | bool | Boolean to specify whether the client should follow HTTP redirects. Default is false. |
| cache_resolved | bool | Boolean to specify whether the client should cache resolved endpoints. The default is false. |
| io_service | shared_ptr<io_service> | Shared pointer to a Boost.Asio io_service. |
| openssl_certificate | string | The filename of the certificate to load for the SSL connection for verification. |
| openssl_verify_path | string | The directory from which the certificate authority files are located. |
| always_verify_peer | bool | Boolean to specify whether the client should always verify peers in SSL connections |
To use the above supported named parameters, you’ll have code that looks like the following:
using namespace boost::network::http; // parameters are in this namespace
client::options options;
options.follow_redirects(true)
.cache_resolved(true)
.io_service(boost::make_shared<boost::asio::io_service>())
.openssl_certificate("/tmp/my-cert")
.openssl_verify_path("/tmp/ca-certs");
client client_(options);
// use client_ as normal from here on out.
The client implementation supports various HTTP methods. The following constructs assume that a client has been properly constructed named client_ and that there is an appropriately constructed request object named request_ and that there is an appropriately constructed response object named response_ like the following:
using namespace boost::network::http; // parameters are here
client client_();
client::request request_("http://cpp-netib.github.com/");
client::response response_;
As of v0.9.1 the library now offers a way to support a streaming body callback function in all HTTP requests that expect a body part (GET, PUT, POST, DELETE). A convenience macro is also provided to make callback handlers easier to write. This macro is called BOOST_NETWORK_HTTP_BODY_CALLBACK which allows users to write the following code to easily create functions or function objects that are compatible with the callback function requirements.
An example of how to use the macro is shown below:
struct body_handler {
explicit body_handler(std::string & body)
: body(body) {}
BOOST_NETWORK_HTTP_BODY_CALLBACK(operator(), range, error) {
// in here, range is the Boost.Range iterator_range, and error is
// the Boost.System error code.
if (!error)
body.append(boost::begin(range), boost::end(range));
}
std::string & body;
};
// somewhere else
std::string some_string;
response_ = client_.get(request("http://cpp-netlib.github.com/"),
body_handler(some_string));
You can also use if for standalone functions instead if you don’t want or need to create a function object.
BOOST_NETWORK_HTTP_BODY_CALLBACK(print_body, range, error) {
if (!error)
std::cout << "Received " << boost::distance(range) << "bytes."
<< std::endl;
else
std::cout << "Error: " << error << std::endl;
}
// somewhere else
response_ = client_.get(request("http://cpp-netlib.github.com/"),
print_body);
The BOOST_NETWORK_HTTP_BODY_CALLBACK macro is defined in boost/network/protocol/http/client/macros.hpp.