Docs Menu
Docs Home
/ / /
C++ Driver
/

Working with BSON

On this page

  • Document Builders
  • List Builder
  • "One-off" Builder Functions
  • Basic Builder
  • Building Arrays in Loops
  • Owning BSON Documents
  • Non-owning BSON Documents (views)
  • Optionally-owning BSON Documents (view_or_value)
  • BSON Document Lifetime
  • Printing BSON Documents
  • Getting Fields out of BSON Documents
  • BSON Types

The mongocxx driver ships with a new library, bsoncxx. This article will go over some of the different types in this library, and how and when to use each. For more information and example code, see our examples.

  1. Document Builders

  2. Owning BSON Documents (values)

  3. Non-owning BSON Documents (views)

  4. Optionally-owning BSON Documents(view_or_value)

  5. BSON Document Lifetime

  6. Printing BSON Documents

  7. Getting Fields out of BSON Documents

  8. BSON Types

The bsoncxx library offers four interfaces for building BSON: one-off functions, a basic builder, a list builder and a stream-based builder.

bsoncxx::builder::basic::document bsoncxx::builder::stream::document

The various methods of creating BSON documents and arrays are all equivalent. All interfaces will provide the same results, the choice of which to use is entirely aesthetic.

The simplest way to create a BSON document or array is to use the JSON-like list builder:

// { "hello": "world" }
bsoncxx::builder::list list_builder = {"hello", "world"};
bsoncxx::document::view document = list_builder.view().get_document();

More advanced uses of the list builder are shown in this example.

The "One-off" builder creates documents and arrays in a single call. These can be used when no additional logic (such as conditionals or loops) needs to be used to create the object:

using bsoncxx::builder::basic::kvp;
// { "hello": "world" }
bsoncxx::document::value document = bsoncxx::builder::basic::make_document(kvp("hello", "world"));
using bsoncxx::builder::basic::kvp;
// { "hello" : "world" }
bsoncxx::builder::basic::document basic_builder{};
basic_builder.append(kvp("hello", "world"));
bsoncxx::document::value document = basic_builder.extract();

More advanced uses of the basic builder are shown in this example.

// { "hello" : "world" }
using bsoncxx::builder::stream;
bsoncxx::document::value document = stream::document{} << "hello" << "world" << stream::finalize;

More advanced uses of the stream builder are shown in this example.

Note

In order to properly append each new value, a stream builder needs to keep track of the state of the current document, including the nesting level and the type of the most recent value appended to the builder. The initial stream builder must not be reused after this state changes, which means that intermediate values must be stored in new variables if a document is being built with the stream builder across multiple statements. Because doing this properly is difficult and the compiler error messages can be confusing, using the stream builder is discouraged. We recommend instead using the basic builder or the one-off builder functions.

Sometimes it's necessary to build an array using a loop. With the basic builder, a top-level array can be built by simply calling append inside a loop:

// [ 1, 2, 3 ]
const auto elements = {1, 2, 3};
auto array_builder = bsoncxx::builder::basic::array{};
for (const auto& element : elements) {
array_builder.append(element);
}

To build a subarray in a loop, pass a lambda to append (or as the second argument of kvp if the subarray is contained by a document rather than an array):

// { "foo" : [ 1, 2, 3 ] }
using bsoncxx::builder::basic::kvp;
using bsoncxx::builder::basic::sub_array;
const auto elements = {1, 2, 3};
auto doc = bsoncxx::builder::basic::document{};
doc.append(kvp("foo", [&elements](sub_array child) {
for (const auto& element : elements) {
child.append(element);
}
}));

When building an array with the stream builder, it's important to be aware that the return type of using the << operator on a stream builder is not uniform. To build an array in a loop properly, intermediate values returned by the stream builder should be stored in variables when the type changes. One attempt to build an array from a stream builder using a loop might look like the following:

// { "subdocs" : [ { "key" : 1 }, { "key" : 2 }, { "key" : 3 } ], "another_key" : 42 }
using namespace bsoncxx;
builder::stream::document builder{};
auto in_array = builder << "subdocs" << builder::stream::open_array;
for (auto&& e : {1, 2, 3}) {
in_array = in_array << builder::stream::open_document << "key" << e
<< builder::stream::close_document;
}
auto after_array = in_array << builder::stream::close_array;
after_array << "another_key" << 42;
document::value doc = after_array << builder::stream::finalize;
std::cout << to_json(doc) << std::endl;

Note

The result of any stream operation should be captured, so if you want to split the single statement within the for loop above into multiple statements, you must capture each intermediate result. Additionally, the last statement within the loop body should assign its result back to the in_array object, so that the loop restarts in a consistent state:

for (auto && e : {1, 2, 3}) {
auto open_state = in_array << builder::stream::open_document;
auto temp_state = open_state << "key" << e;
in_array = temp_state << builder::stream::close_document;
}

bsoncxx::document::value

This type represents an actual BSON document, one that owns its buffer of data. These documents can be constructed from a builder by calling extract():

bsoncxx::document::value basic_doc{basic_builder.extract()};
bsoncxx::document::value stream_doc{stream_builder.extract()};

After calling extract() the builder is in a moved-from state, and should not be used.

It's possible to create a bsoncxx::document::value in a single line using the stream builder interface and the finalize token. finalize returns a document::value from a temporary stream builder:

// { "finalize" : "is nifty" }
bsoncxx::document::value one_line = bsoncxx::builder::stream::document{} << "finalize" << "is nifty" << bsoncxx::builder::stream::finalize;

bsoncxx::document::view

This type is a view into an owning bsoncxx::document::value.

bsoncxx::document::view document_view{document_value.view()};

A document::value also implicitly converts to a document::view:

bsoncxx::document::view document_view{document_value};

In performance-critical code, passing views around is preferable to using values because we can avoid excess copying. Also, passing a view of a document allows us to use the document multiple times:

// { "copies" : { "$gt" : 100 } }
auto query_value = document{} << "copies" << open_document << "$gt" << 100 << close_document << finalize;
// Run the same query across different collections
auto collection1 = db["science_fiction"];
auto cursor1 = collection1.find(query_value.view());
auto collection2 = db["cookbooks"];
auto cursor2 = collection2.find(query_value.view());

Many driver methods take a document::view_or_value parameter, for example, run_command:

bsoncxx::document::value run_command(bsoncxx::document::view_or_value command);

Such methods can take either a document::view or a document::value. If a document::value is passed in, it must be passed by r-value reference, so ownership of the document is transferred to the method.

document::value ping = document{} << "ping" << 1 << finalize;
// You can pass a document::view into run_command()
db.run_command(ping.view());
// Or you can move in a document::value
db.run_command(std::move(ping));

You shouldn't need to create view_or_value types directly in order to use the driver. They are offered as a convenience method to allow driver methods to take documents in either an owning or non-owning way. The view_or_value type also helps mitigate some of the lifetime issues discussed in the next section.

It is imperative that document::value types outlive any document::view types that use them. If the underlying value gets cleaned up, the view will be left with a dangling pointer. Consider a method that returns a view of a newly-created document:

bsoncxx::document::view make_a_dangling_view() {
bsoncxx::builder::basic::document builder{};
builder.append(kvp("hello", "world"));
// This creates a document::value on the stack that will disappear when we return.
bsoncxx::document::value stack_value{builder.extract()};
// We're returning a view of the local value
return stack_value.view(); // Bad!!
}

This method returns a dangling view that should not be used:

// This view contains a dangling pointer
bsoncxx::document::view dangling_view = make_a_dangling_view(); // Warning!!

Attempting to create a view off of a builder will similarly create a dangerous view object, because the temporary value returned from extract() isn't captured:

bsoncxx::builder::stream::document temp_builder{};
temp_builder << "oh" << "no";
bsoncxx::document::view dangling_view = temp_builder.extract().view(); // Bad!!

bsoncxx::to_json()

The bsoncxx library comes with a convenience method to convert BSON documents to strings for easy inspection:

bsoncxx::document::value = document{} << "I am" << "a BSON document" << finalize;
std::cout << bsoncxx::to_json(doc.view()) << std::endl;

There is an analogous method, from_json(), to build document::values out of existing JSON strings.

The [ ] operator reaches into a BSON document to retrieve values:

// doc_view = { "store" : "Key Foods", "fruits" : [ "apple", "banana" ] }
auto store = doc_view["store"];
auto first_fruit = doc_view["fruits"][0];

This returns a bsoncxx::document::element, which holds the actual desired value:

document::element store_ele{doc_view["store"]};
if (store_ele) {
// this block will only execute if "store" was found in the document
std::cout << "Examining inventory at " << to_json(store_ele.get_value()) << std::endl;
}

This feature is shown in more detail in this example and this example.

The BSON specification provides a list of supported types. These are represented in C++ using the b_xxx type wrappers.

Some BSON types don't necessarily have a native representation to wrap and are implemented via special classes.

The bsoncxx::decimal128 class represents a 128-bit IEEE 754-2008 decimal floating point value. We expect users to convert these to and from strings, but provide access to the low and high 64-bit values if users need to convert to a native decimal128 type.

You can see how to work with bsoncxx::decimal128 in this example.

Back

Time Series Data