Welcome to EXTRAS-SOCI!

Backends reference

This part of the documentation is provided for those who want to write (and contribute!) their own backends. It is anyway recommended that authors of new backend see the code of some existing backend for hints on how things are really done.

The backend interface is a set of base classes that the actual backends are supposed to specialize. The main SOCI interface uses only the interface and respecting the protocol (for example, the order of function calls) described here. Note that both the interface and the protocol were initially designed with the Oracle database in mind, which means that whereas it is quite natural with respect to the way Oracle API (OCI) works, it might impose some implementation burden on other backends, where things are done differently and therefore have to be adjusted, cached, converted, etc.

The interface to the common SOCI interface is defined in the core/soci-backend.h header file. This file is dissected below.

All names are defined in either soci or soci::details namespace.

// data types, as seen by the user
enum data_type
{
    dt_string, dt_date, dt_double, dt_integer, dt_unsigned_long, dt_long_long
};

// the enum type for indicator variables
enum indicator { i_ok, i_null, i_truncated };

// data types, as used to describe exchange format
enum exchange_type
{
    x_char, x_cstring, x_stdstring, x_short, x_integer,
    x_unsigned_long, x_long_long, x_double, x_stdtm, x_statement,
    x_rowid, x_blob
};

struct cstring_descriptor
{
    cstring_descriptor(char * str, std::size_t bufSize)
        : str_(str), bufSize_(bufSize) {}

    char * str_;
    std::size_t bufSize_;
};

// actually in error.h:
class soci_error : public std::runtime_error
{
public:
    soci_error(std::string const & msg);
};

The data_type enumeration type defines all types that form the core type support for SOCI. The enum itself can be used by clients when dealing with dynamic rowset description.

The indicator enumeration type defines all recognized states of data. The i_truncated state is provided for the case where the string is retrieved from the database into the char buffer that is not long enough to hold the whole value.

The exchange_type enumeration type defines all possible types that can be used with the into and use elements.

The cstring_descriptor is a helper class that allows to store the address of char buffer together with its size. The objects of this class are passed to the backend when the x_cstring type is involved.

The soci_error class is an exception type used for database-related (and also usage-related) errors. The backends should throw exceptions of this or derived type only.

class standard_into_type_backend
{
public:
    virtual ~standard_into_type_backend() {}

    virtual void define_by_pos(int & position,
        void * data, exchange_type type) = 0;

    virtual void pre_fetch() = 0;
    virtual void post_fetch(bool gotData, bool calledFromFetch,
        indicator * ind) = 0;

    virtual void clean_up() = 0;
};

The standard_into_type_back_end class implements the dynamic interactions with the simple (non-bulk) into elements. The objects of this class (or, rather, of the derived class implemented by the actual backend) are created by the statement object when the into element is bound - in terms of lifetime management, statement is the master of this class.

define_by_pos - Called when the into element is bound, once and before the statement is executed. The data pointer points to the variable used for into element (or to the cstring_descriptor object, which is artificially created when the plain char buffer is used for data exchange). The position parameter is a "column number", assigned by the library. The backend should increase this parameter, according to the number of fields actually taken (usually 1).
pre_fetch - Called before each row is fetched.
post_fetch - Called after each row is fetched. The gotData parameter is true if the fetch operation really retrieved some data and false otherwise; calledFromFetch is true when the call is from the fetch operation and false if it is from the execute operation (this is also the case for simple, one-time queries). In particular, (calledFromFetch && !gotData) indicates that there is an end-of-rowset condition. ind points to the indicator provided by the user, or is NULL, if there is no indicator.
clean_up - Called once when the statement is destroyed.

The intended use of pre_fetch and post_fetch functions is to manage any internal buffer and/or data conversion for each value retrieved from the database. If the given server supports binary data transmission and the data format for the given type agrees with what is used on the client machine, then these two functions need not do anything; otherwise buffer management and data conversions should go there.

class vector_into_type_backend
{
public:
    virtual ~vector_into_type_backend() {}

    virtual void define_by_pos(int & position,
        void * data, exchange_type type) = 0;

    virtual void pre_fetch() = 0;
    virtual void post_fetch(bool gotData, indicator * ind) = 0;

    virtual void resize(std::size_t sz) = 0;
    virtual std::size_t size() = 0;

    virtual void clean_up() = 0;
};

The vector_into_type_back_end has similar structure and purpose as the previous one, but is used for vectors (bulk data retrieval).

The data pointer points to the variable of type std::vector<T> (and not to its internal buffer), resize is supposed to really resize the user-provided vector and size is supposed to return the current size of this vector. The important difference with regard to the previous class is that ind points (if not NULL) to the beginning of the array of indicators. The backend should fill this array according to the actual state of the retrieved data.

class standard_use_type_backend
{
public:
    virtual ~standard_use_type_backend() {}

    virtual void bind_by_pos(int & position,
        void * data, exchange_type type, bool readOnly) = 0;
    virtual void bind_by_name(std::string const & name,
        void * data, exchange_type type, bool readOnly) = 0;

    virtual void pre_use(indicator const * ind) = 0;
    virtual void post_use(bool gotData, indicator * ind) = 0;

    virtual void clean_up() = 0;
};

The standard_use_type_backend implements the interactions with the simple (non-bulk) use elements, created and destroyed by the statement object.

bind_by_pos - Called for each use element, once and before the statement is executed - for those use elements that do not provide explicit names for parameter binding. The meaning of parameters is same as in previous classes.
bind_by_name - Called for those use elements that provide the explicit name.
pre_use - Called before the data is transmitted to the server (this means before the statement is executed, which can happen many times for the prepared statement). ind points to the indicator provided by the user (or is NULL).
post_use - Called after statement execution. gotData and ind have the same meaning as in standard_into_type_back_end::post_fetch, and this can be used by those backends whose respective servers support two-way data exchange (like in/out parameters in stored procedures).

The intended use for pre_use and post_use methods is to manage any internal buffers and/or data conversion. They can be called many times with the same statement.

class vector_use_type_backend
{
public:
    virtual ~vector_use_type_backend() {}

    virtual void bind_by_pos(int & position,
        void * data, exchange_type type) = 0;
    virtual void bind_by_name(std::string const & name,
        void * data, exchange_type type) = 0;

    virtual void pre_use(indicator const * ind) = 0;

    virtual std::size_t size() = 0;

    virtual void clean_up() = 0;
};

Objects of this type (or rather of type derived from this one) are used to implement interactions with user-provided vector (bulk) use elements and are managed by the statement object. The data pointer points to the whole vector object provided by the user (and not to its internal buffer); ind points to the beginning of the array of indicators (or is NULL). The meaning of this interface is analogous to those presented above.

class statement_backend
{
public:
    virtual ~statement_backend() {}

    virtual void alloc() = 0;
    virtual void clean_up() = 0;

    virtual void prepare(std::string const & query, statement_type eType) = 0;

    enum exec_fetch_result { ef_success, ef_no_data };
    virtual exec_fetch_result execute(int number) = 0;
    virtual exec_fetch_result fetch(int number) = 0;

    virtual int get_number_of_rows() = 0;

    virtual std::string rewrite_for_procedure_call(std::string const & query) = 0;

    virtual int prepare_for_describe() = 0;
    virtual void describe_column(int colNum, data_type & dtype,
        std::string & column_name) = 0;

    virtual standard_into_type_backend * make_into_type_backend() = 0;
    virtual standard_use_type_backend * make_use_type_backend() = 0;
    virtual vector_into_type_backend * make_vector_into_type_backend() = 0;
    virtual vector_use_type_backend * make_vector_use_type_backend() = 0;
};

The statement_backend type implements the internals of the statement objects. The objects of this class are created by the session object.

alloc - Called once to allocate everything that is needed for the statement to work correctly.
clean_up - Supposed to clean up the resources, called once.
prepare - Called once with the text of the SQL query. For servers that support explicit query preparation, this is the place to do it.
execute - Called to execute the query; if number is zero, the intent is not to exchange data with the user-provided objects (into and use elements); positive values mean the number of rows to exchange (more than 1 is used only for bulk operations).
fetch - Called to fetch next bunch of rows; number is positive and determines the requested number of rows (more than 1 is used only for bulk operations).
get_number_of_rows - Called to determine the actual number of rows retrieved by the previous call to execute or fetch.
rewrite_for_procedure_call - Used when the procedure is used instead of statement, to call the stored procedure. This function should rewrite the SQL query (if necessary) to the form that will allow to execute the given procedure.
prepare_for_describe - Called once when the into element is used with the row type, which means that dynamic rowset description should be performed. It is supposed to do whatever is needed to later describe the column properties and should return the number of columns.
describe_column - Called once for each column (column numbers - colNum - start from 1), should fill its parameters according to the column properties.
make_into_type_backend, make_use_type_backend, make_vector_into_type_backend, make_vector_use_type_backend - Called once for each into or use element, to create the objects of appropriate classes (described above).

Notes:

Whether the query is executed using the simple one-time syntax or is prepared, the alloc, prepare and execute functions are always called, in this order.
All into and use elements are bound (their define_by_pos or bind_by_pos/bind_by_name functions are called) between statement preparation and execution.

class rowid_backend
{
public:
    virtual ~rowid_backend() {}
};

The rowid_backend class is a hook for the backends to provide their own state for the row identifier. It has no functions, since the only portable interaction with the row identifier object is to use it with into and use elements.

class blob_backend
{
public:
    virtual ~blob_backend() {}

    virtual std::size_t get_len() = 0;
    virtual std::size_t read(std::size_t offset, char * buf,
        std::size_t toRead) = 0;
    virtual std::size_t write(std::size_t offset, char const * buf,
        std::size_t toWrite) = 0;
    virtual std::size_t append(char const * buf, std::size_t toWrite) = 0;
    virtual void trim(std::size_t newLen) = 0;
};

The blob_backend interface provides the entry points for the blob methods.

class session_backend
{
public:
    virtual ~session_backend() {}

    virtual void begin() = 0;
    virtual void commit() = 0;
    virtual void rollback() = 0;

    virtual std::string get_backend_name() const = 0;

    virtual statement_backend * make_statement_backend() = 0;
    virtual rowid_backend * make_rowid_backend() = 0;
    virtual blob_backend * make_blob_backend() = 0;
};

The object of the class derived from session_backend implements the internals of the session object.

begin, commit, rollback - Forward-called when the same functions of session are called by user.
make_statement_backend, make_rowid_backend, make_blob_backend - Called to create respective implementations for the statement, rowid and blob classes.

struct backend_factory
{
    virtual ~backend_factory() {}

    virtual details::session_backend * make_session(
        std::string const & connectString) const = 0;
};

The backend_factory is a base class for backend-provided factory class that is able to create valid sessions. The connectString parameter passed to make_session is provided here by the session constructor and contains only the backend-related parameters, without the backend name (if the dynamic backend loading is used).

The actual backend factory object is supposed to be provided by the backend implementation and declared in its header file. In addition to this, the factory_ABC function with the "C" calling convention and returning the pointer to concrete factory object should be provided, where ABC is the backend name.

The following example is taken from soci-postgresql.h, which declares entities of the PostgreSQL backend:

struct postgresql_backend_factory : backend_factory
{
    virtual postgresql_session_backend * make_session(
        std::string const & connectString) const;
};

extern postgresql_backend_factory const postgresql;

extern "C"
{

// for dynamic backend loading
backend_factory const * factory_postgresql();

} // extern "C"

With the above declarations, it is enough to pass the postgresql factory name to the constructor of the session object, which will use this factory to create concrete implementations for any other objects that are needed, with the help of appropriate make_XYZ functions. Alternatively, the factory_postgresql function will be called automatically by the backend loader if the backend name is provided at run-time instead.

Note that the backend source code is placed in the backends/name directory (for example, backends/oracle) and the test driver is in backends/name/test. There is also backends/empty directory provided as a skeleton for development of new backends and their tests. It is recommended that all backends respect naming conventions by just appending their name to the base-class names. The backend name used for the global factory object should clearly identify the given database engine, like oracle, postgresql, mysql, and so on.

Previous (Client reference)

Next (Rationale FAQ)