It is of the utmost importance that Driver developers only use lowercase attribute names when defining private attributes. Private attribute names must be prefixed with the driver name or suitable abbreviation (e.g., "ora_" for Oracle, "ing_" for Ingres, etc).

Most DBI drivers require applications to use a dialect of SQL (Structured Query Language) to interact with the database engine. The "Standards Reference Information" section provides links to useful information about SQL.

The DBI itself does not mandate or require any particular language to be used; it is language independent. In ODBC terms, the DBI is in "pass-thru" mode, although individual drivers might not be. The only requirement is that queries and other statements must be expressed as a single string of characters passed as the first argument to the "prepare" or "do" methods.

For an interesting diversion on the real history of RDBMS and SQL, from the people who made it happen, see:

Follow the "Full Contents" then "Intergalactic dataspeak" links for the SQL history.

Some drivers support placeholders and bind values. Placeholders, also called parameter markers, are used to indicate values in a database statement that will be supplied later, before the prepared statement is executed. For example, an application might use the following to insert a row of data into the SALES table:

or the following, to select the description for a product:

The ? characters are the placeholders. The association of actual values with placeholders is known as binding, and the values are referred to as bind values. Note that the ? is not enclosed in quotation marks, even when the placeholder represents a string.

Some drivers also allow placeholders like :name and :N (e.g., :1, :2, and so on) in addition to ?, but their use is not portable.

If the :N form of placeholder is supported by the driver you're using, then you should be able to use either "bind_param" or "execute" to bind values. Check your driver documentation.

With most drivers, placeholders can't be used for any element of a statement that would prevent the database server from validating the statement and creating a query execution plan for it. For example:

Also, placeholders can only represent single scalar values. For example, the following statement won't work as expected for more than one value:

When using placeholders with the SQL LIKE qualifier, you must remember that the placeholder substitutes for the whole string. So you should use "... LIKE ? ..." and include any wildcard characters in the value that you bind to the placeholder.

NULL Values

Undefined values, or undef, are used to indicate NULL values. You can insert and update columns with a NULL value as you would a non-NULL value. These examples insert and update the column age with a NULL value:

However, care must be taken when trying to use NULL values in a WHERE clause. Consider:

Binding an undef (NULL) to the placeholder will not select rows which have a NULL age! At least for database engines that conform to the SQL standard. Refer to the SQL manual for your database engine or any SQL book for the reasons for this. To explicitly select NULLs you have to say "WHERE age IS NULL".

A common issue is to have a code fragment handle a value that could be either defined or undef (non-NULL or NULL) at runtime. A simple technique is to prepare the appropriate statement as needed, and substitute the placeholder for non-NULL cases:

The following technique illustrates qualifying a WHERE clause with several columns, whose associated values (defined or undef) are in a hash %h:

The techniques above call prepare for the SQL statement with each call to execute. Because calls to prepare() can be expensive, performance can suffer when an application iterates many times over statements like the above.

A better solution is a single WHERE clause that supports both NULL and non-NULL comparisons. Its SQL statement would need to be prepared only once for all cases, thus improving performance. Several examples of WHERE clauses that support this are presented below. But each example lacks portability, robustness, or simplicity. Whether an example is supported on your database engine depends on what SQL extensions it provides, and where it supports the ? placeholder in a statement.

Statements formed with the above WHERE clauses require execute statements as follows. The arguments are required, whether their values are defined or undef.

Example 0 should not work (as mentioned earlier), but may work on a few database engines anyway (e.g. Sybase). Example 0 is part of examples 4, 5, and 6, so if example 0 works, these other examples may work, even if the engine does not properly support the right hand side of the OR expression.

Examples 1 and 2 are not robust: they require that you provide a valid column value xx (e.g. '~') which is not present in any row. That means you must have some notion of what data won't be stored in the column, and expect clients to adhere to that.

Example 5 requires that you provide a stored procedure (SP_ISNULL in this example) that acts as a function: it checks whether a value is null, and returns 1 if it is, or 0 if not.

Example 6, the least simple, is probably the most portable, i.e., it should work with with most, if not all, database engines.

Here is a table that indicates which examples above are known to work on various database engines:

* Works only because Example 0 works.

DBI provides a sample perl script that will test the examples above on your database engine and tell you which ones work. It is located in the ex/ subdirectory of the DBI source distribution, or here: http://svn.perl.org/modules/dbi/trunk/ex/perl_dbi_nulls_test.pl Please use the script to help us fill-in and maintain this table.

Performance

Without using placeholders, the insert statement shown previously would have to contain the literal values to be inserted and would have to be re-prepared and re-executed for each row. With placeholders, the insert statement only needs to be prepared once. The bind values for each row can be given to the execute method each time it's called. By avoiding the need to re-prepare the statement for each row, the application typically runs many times faster. Here's an example:

See "execute" and "bind_param" for more details.

The q{...} style quoting used in this example avoids clashing with quotes that may be used in the SQL statement. Use the double-quote like qq{...} operator if you want to interpolate variables into the string. See "Quote and Quote-like Operators" in perlop for more details.

See also the "bind_columns" method, which is used to associate Perl variables with the output columns of a SELECT statement.

In this section, we cover the DBI class methods, utility functions, and the dynamic attributes associated with generic DBI handles.

Constants representing the values of the SQL standard types can be imported individually by name, or all together by importing the special :sql_types tag.

The names and values of all the defined SQL standard types can be produced like this:

These constants are defined by SQL/CLI, ODBC or both. SQL_BIGINT is (currently) omitted, because SQL/CLI and ODBC provide conflicting codes.

See the "type_info", "type_info_all", and "bind_param" methods for possible uses.

Note that just because the DBI defines a named constant for a given data type doesn't mean that drivers will support that data type.

The following methods are provided by the DBI class:

Breaks apart a DBI Data Source Name (DSN) and returns the individual parts. If $dsn doesn't contain a valid DSN then parse_dsn() returns an empty list.

$scheme is the first part of the DSN and is currently always 'dbi'. $driver is the driver name, possibly defaulted to $ENV{DBI_DRIVER}, and may be undefined. $attr_string is the contents of the optional attribute string, which may be undefined. If $attr_string is not empty then $attr_hash is a reference to a hash containing the parsed attribute names and values. $driver_dsn is the last part of the DBI DSN string. For example:

The parse_dsn() method was added in DBI 1.43.

Establishes a database connection, or session, to the requested $data_source. Returns a database handle object if the connection succeeds. Use $dbh->disconnect to terminate the connection.

If the connect fails (see below), it returns undef and sets both $DBI::err and $DBI::errstr. (It does not explicitly set $!.) You should generally test the return status of connect and print $DBI::errstr if it has failed.

Multiple simultaneous connections to multiple databases through multiple drivers can be made via the DBI. Simply make one connect call for each database and keep a copy of each returned database handle.

The $data_source value must begin with "dbi:driver_name:". The driver_name specifies the driver that will be used to make the connection. (Letter case is significant.)

As a convenience, if the $data_source parameter is undefined or empty, the DBI will substitute the value of the environment variable DBI_DSN. If just the driver_name part is empty (i.e., the $data_source prefix is "dbi::"), the environment variable DBI_DRIVER is used. If neither variable is set, then connect dies.

Examples of $data_source values are:

There is no standard for the text following the driver name. Each driver is free to use whatever syntax it wants. The only requirement the DBI makes is that all the information is supplied in a single string. You must consult the documentation for the drivers you are using for a description of the syntax they require.

It is recommended that drivers support the ODBC style, shown in the last example above. It is also recommended that that they support the three common names 'host', 'port', and 'database' (plus 'db' as an alias for database). This simplifies automatic construction of basic DSNs: "dbi:$driver:database=$db;host=$host;port=$port". Drivers should aim to 'do something reasonable' when given a DSN in this form, but if any part is meaningless for that driver (such as 'port' for Informix) it should generate an error if that part is not empty.

If the environment variable DBI_AUTOPROXY is defined (and the driver in $data_source is not "Proxy") then the connect request will automatically be changed to:

DBI_AUTOPROXY is typically set as "dbi:Proxy:hostname=...;port=...". If $ENV{DBI_AUTOPROXY} doesn't begin with 'dbi:' then "dbi:Proxy:" will be prepended to it first. See the DBD::Proxy documentation for more details.

If $username or $password are undefined (rather than just empty), then the DBI will substitute the values of the DBI_USER and DBI_PASS environment variables, respectively. The DBI will warn if the environment variables are not defined. However, the everyday use of these environment variables is not recommended for security reasons. The mechanism is primarily intended to simplify testing. See below for alternative way to specify the username and password.

DBI->connect automatically installs the driver if it has not been installed yet. Driver installation either returns a valid driver handle, or it dies with an error message that includes the string "install_driver" and the underlying problem. So DBI->connect will die on a driver installation failure and will only return undef on a connect failure, in which case $DBI::errstr will hold the error message. Use eval { ... } if you need to catch the "install_driver" error.

The $data_source argument (with the "dbi:...:" prefix removed) and the $username and $password arguments are then passed to the driver for processing. The DBI does not define any interpretation for the contents of these fields. The driver is free to interpret the $data_source, $username, and $password fields in any way, and supply whatever defaults are appropriate for the engine being accessed. (Oracle, for example, uses the ORACLE_SID and TWO_TASK environment variables if no $data_source is specified.)

The AutoCommit and PrintError attributes for each connection default to "on". (See "AutoCommit" and "PrintError" for more information.) However, it is strongly recommended that you explicitly define AutoCommit rather than rely on the default. The PrintWarn attribute defaults to on if $^W is true, i.e., perl is running with warnings enabled.

The \%attr parameter can be used to alter the default settings of PrintError, RaiseError, AutoCommit, and other attributes. For example:

The username and password can also be specified using the attributes Username and Password, in which case they take precedence over the $username and $password parameters.

You can also define connection attribute values within the $data_source parameter. For example:

Individual attributes values specified in this way take precedence over any conflicting values specified via the \%attr parameter to connect.

The dbi_connect_method attribute can be used to specify which driver method should be called to establish the connection. The only useful values are 'connect', 'connect_cached', or some specialized case like 'Apache::DBI::connect' (which is automatically the default when running within Apache).

Where possible, each session ($dbh) is independent from the transactions in other sessions. This is useful when you need to hold cursors open across transactions--for example, if you use one session for your long lifespan cursors (typically read-only) and another for your short update transactions.

For compatibility with old DBI scripts, the driver can be specified by passing its name as the fourth argument to connect (instead of \%attr):

In this "old-style" form of connect, the $data_source should not start with "dbi:driver_name:". (If it does, the embedded driver_name will be ignored). Also note that in this older form of connect, the $dbh->{AutoCommit} attribute is undefined, the $dbh->{PrintError} attribute is off, and the old DBI_DBNAME environment variable is checked if DBI_DSN is not defined. Beware that this "old-style" connect will soon be withdrawn in a future version of DBI.

connect_cached is like "connect", except that the database handle returned is also stored in a hash associated with the given parameters. If another call is made to connect_cached with the same parameter values, then the corresponding cached $dbh will be returned if it is still valid. The cached database handle is replaced with a new connection if it has been disconnected or if the ping method fails.

Note that the behaviour of this method differs in several respects from the behaviour of persistent connections implemented by Apache::DBI. However, if Apache::DBI is loaded then connect_cached will use it.

Caching connections can be useful in some applications, but it can also cause problems, such as too many connections, and so should be used with care. In particular, avoid changing the attributes of a database handle created via connect_cached() because it will affect other code that may be using the same handle. When connect_cached() returns a handle the attributes will be reset to their initial values. This can cause problems, especially with the AutoCommit attribute.

Where multiple separate parts of a program are using connect_cached() to connect to the same database with the same (initial) attributes it is a good idea to add a private attribute to the connect_cached() call to effectively limit the scope of the caching. For example:

Handles returned from that connect_cached() call will only be returned by other connect_cached() call elsewhere in the code if those other calls also pass in the same attribute values, including the private one. (I've used private_foo_cachekey here as an example, you can use any attribute name with a private_ prefix.)

Taking that one step further, you can limit a particular connect_cached() call to return handles unique to that one place in the code by setting the private attribute to a unique value for that place:

By using a private attribute you still get connection caching for the individual calls to connect_cached() but, by making separate database connections for separate parts of the code, the database handles are isolated from any attribute changes made to other handles.

The cache can be accessed (and cleared) via the "CachedKids" attribute:

Returns a list of all available drivers by searching for DBD::* modules through the directories in @INC. By default, a warning is given if some drivers are hidden by others of the same name in earlier directories. Passing a true value for $quiet will inhibit the warning.

Returns a list of driver name and driver handle pairs for all drivers 'installed' (loaded) into the current process. The driver name does not include the 'DBD::' prefix.

To get a list of all drivers available in your perl installation you can use "available_drivers".

Added in DBI 1.49.

Calls available_drivers() and attempts to load each of them in turn using install_driver(). For each load that succeeds the driver name and version number are added to a hash. When running under DBI::PurePerl drivers which appear not be pure-perl are ignored.

When called in array context the list of successfully loaded drivers is returned (without the 'DBD::' prefix).

When called in scalar context a reference to the hash is returned and the hash will also contain other entries for the DBI version, OS name, etc.

When called in a void context the installed_versions() method will print out a formatted list of the hash contents, one per line.

Due to the potentially high memory cost and unknown risks of loading in an unknown number of drivers that just happen to be installed on the system, this method is not recommended for general use. Use available_drivers() instead.

The installed_versions() method is primarily intended as a quick way to see from the command line what's installed. For example:

The installed_versions() method was added in DBI 1.38.

Returns a list of data sources (databases) available via the named driver. If $driver is empty or undef, then the value of the DBI_DRIVER environment variable is used.

The driver will be loaded if it hasn't been already. Note that if the driver loading fails then data_sources() dies with an error message that includes the string "install_driver" and the underlying problem.

Data sources are returned in a form suitable for passing to the "connect" method (that is, they will include the "dbi:$driver:" prefix).

Note that many drivers have no way of knowing what data sources might be available for it. These drivers return an empty or incomplete list or may require driver-specific attributes.

There is also a data_sources() method defined for database handles.

The DBI->trace method sets the global default trace settings and returns the previous trace settings. It can also be used to change where the trace output is sent.

There's a similar method, $h->trace, which sets the trace settings for the specific handle it's called on.

See the "TRACING" section for full details about the DBI's powerful tracing facilities.

Where $coderef is a reference to a subroutine and $info is an arbitrary value which, if undefined, defaults to a reference to an empty hash. Returns $info.

For each installed driver handle, if any, $coderef is invoked as:

If the execution of $coderef returns a true value then "visit_child_handles" is called on that child handle and passed the returned value as $info.

For example:

See also "visit_child_handles".

In addition to the DBI methods listed in the previous section, the DBI package also provides several utility functions.

These can be imported into your code by listing them in the use statement. For example:

Alternatively, all these utility functions (except hash) can be imported using the :utils import tag. For example:

Returns an informal description of the string. For example:

The initial UTF8 on/off refers to Perl's internal SvUTF8 flag. If $string has the SvUTF8 flag set but the sequence of bytes it contains are not a valid UTF-8 encoding then data_string_desc() will report UTF8 on but INVALID encoding.

The ASCII vs non-ASCII portion shows ASCII if all the characters in the string are ASCII (have code points <= 127).

The data_string_desc() function was added in DBI 1.46.

Returns an informal description of the first character difference between the strings. If both $a and $b contain the same sequence of characters then data_string_diff() returns an empty string. For example:

Unicode characters are reported in \x{XXXX} format. Unicode code points in the range U+0800 to U+08FF are unassigned and most likely to occur due to double-encoding. Characters in this range are reported as \x{08XX}='C' where C is the corresponding latin-1 character.

The data_string_diff() function only considers logical characters and not the underlying encoding. See "data_diff" for an alternative.

The data_string_diff() function was added in DBI 1.46.

Returns an informal description of the difference between two strings. It calls "data_string_desc" and "data_string_diff" and returns the combined results as a multi-line string.

For example, data_diff("abc", "ab\x{263a}") will return:

If $a and $b are identical in both the characters they contain and their physical encoding then data_diff() returns an empty string. If $logical is true then physical encoding differences are ignored (but are still reported if there is a difference in the characters).

The data_diff() function was added in DBI 1.46.

Return a string containing a neat (and tidy) representation of the supplied value.

Strings will be quoted, although internal quotes will not be escaped. Values known to be numeric will be unquoted. Undefined (NULL) values will be shown as undef (without quotes).

If the string is flagged internally as utf8 then double quotes will be used, otherwise single quotes are used and unprintable characters will be replaced by dot (.).

For result strings longer than $maxlen the result string will be truncated to $maxlen-4 and "...'" will be appended. If $maxlen is 0 or undef, it defaults to $DBI::neat_maxlen which, in turn, defaults to 400.

This function is designed to format values for human consumption. It is used internally by the DBI for "trace" output. It should typically not be used for formatting values for database use. (See also "quote".)

Calls neat on each element of the list and returns a string containing the results joined with $field_sep. $field_sep defaults to ", ".

Returns true for each element that looks like a number. Returns false for each element that does not look like a number. Returns undef for each element that is undefined or empty.

Return a 32-bit integer 'hash' value corresponding to the contents of $buffer. The $type parameter selects which kind of hash algorithm should be used.

For the technically curious, type 0 (which is the default if $type isn't specified) is based on the Perl 5.1 hash except that the value is forced to be negative (for obscure historical reasons). Type 1 is the better "Fowler / Noll / Vo" (FNV) hash. See http://www.isthe.com/chongo/tech/comp/fnv/ for more information. Both types are implemented in C and are very fast.

This function doesn't have much to do with databases, except that it can be handy to store hash values in a database.

sql_type_cast attempts to cast $sv to the SQL type (see "DBI Constants") specified in $sql_type. At present only the SQL types SQL_INTEGER, SQL_DOUBLE and SQL_NUMERIC are supported.

For SQL_INTEGER the effect is similar to using the value in an expression that requires an integer. It gives the perl scalar an 'integer aspect'. (Technically the value gains an IV, or possibly a UV or NV if the value is too large for an IV.)

For SQL_DOUBLE the effect is similar to using the value in an expression that requires a general numeric value. It gives the perl scalar a 'numeric aspect'. (Technically the value gains an NV.)

SQL_NUMERIC is similar to SQL_INTEGER or SQL_DOUBLE but more general and more cautious. It will look at the string first and if it looks like an integer (that will fit in an IV or UV) it will act like SQL_INTEGER, if it looks like a floating point value it will act like SQL_DOUBLE, if it looks like neither then it will do nothing - and thereby avoid the warnings that would be generated by SQL_INTEGER and SQL_DOUBLE when given non-numeric data.

$flags may be:

The returned $sts value is:

This method is exported by the :utils tag and was introduced in DBI 1.611.

Dynamic attributes are always associated with the last handle used (that handle is represented by $h in the descriptions below).

Where an attribute is equivalent to a method call, then refer to the method call for all related documentation.

Warning: these attributes are provided as a convenience but they do have limitations. Specifically, they have a short lifespan: because they are associated with the last handle used, they should only be used immediately after calling the method that "sets" them. If in any doubt, use the corresponding method call.

Equivalent to $h->err.

Equivalent to $h->errstr.

Equivalent to $h->state.

Equivalent to $h->rows. Please refer to the documentation for the "rows" method.

Returns the DBI object handle used for the most recent DBI method call. If the last DBI method call was a DESTROY then $DBI::lasth will return the handle of the parent of the destroyed handle, if there is one.

The following methods can be used by all types of DBI handles.

Returns the native database engine error code from the last driver method called. The code is typically an integer but you should not assume that.

The DBI resets $h->err to undef before almost all DBI method calls, so the value only has a short lifespan. Also, for most drivers, the statement handles share the same error variable as the parent database handle, so calling a method on one handle may reset the error on the related handles.

(Methods which don't reset err before being called include err() and errstr(), obviously, state(), rows(), func(), trace(), trace_msg(), ping(), and the tied hash attribute FETCH() and STORE() methods.)

If you need to test for specific error conditions and have your program be portable to different database engines, then you'll need to determine what the corresponding error codes are for all those engines and test for all of them.

The DBI uses the value of $DBI::stderr as the err value for internal errors. Drivers should also do likewise. The default value for $DBI::stderr is 2000000000.

A driver may return 0 from err() to indicate a warning condition after a method call. Similarly, a driver may return an empty string to indicate a 'success with information' condition. In both these cases the value is false but not undef. The errstr() and state() methods may be used to retrieve extra information in these cases.

See "set_err" for more information.

Returns the native database engine error message from the last DBI method called. This has the same lifespan issues as the "err" method described above.

The returned string may contain multiple messages separated by newline characters.

The errstr() method should not be used to test for errors, use err() for that, because drivers may return 'success with information' or warning messages via errstr() for methods that have not 'failed'.

See "set_err" for more information.

Returns a state code in the standard SQLSTATE five character format. Note that the specific success code 00000 is translated to any empty string (false). If the driver does not support SQLSTATE (and most don't), then state() will return S1000 (General Error) for all errors.

The driver is free to return any value via state, e.g., warning codes, even if it has not declared an error by returning a true value via the "err" method described above.

The state() method should not be used to test for errors, use err() for that, because drivers may return a 'success with information' or warning state code via state() for methods that have not 'failed'.

Set the err, errstr, and state values for the handle. This method is typically only used by DBI drivers and DBI subclasses.

If the "HandleSetErr" attribute holds a reference to a subroutine it is called first. The subroutine can alter the $err, $errstr, $state, and $method values. See "HandleSetErr" for full details. If the subroutine returns a true value then the handle err, errstr, and state values are not altered and set_err() returns an empty list (it normally returns $rv which defaults to undef, see below).

Setting err to a true value indicates an error and will trigger the normal DBI error handling mechanisms, such as RaiseError and HandleError, if they are enabled, when execution returns from the DBI back to the application.

Setting err to "" indicates an 'information' state, and setting it to "0" indicates a 'warning' state. Setting err to undef also sets errstr to undef, and state to "", irrespective of the values of the $errstr and $state parameters.

The $method parameter provides an alternate method name for the RaiseError/PrintError/PrintWarn error string instead of the fairly unhelpful 'set_err'.

The set_err method normally returns undef. The $rv parameter provides an alternate return value.

Some special rules apply if the err or errstr values for the handle are already set...

If errstr is true then: " [err was %s now %s]" is appended if $err is true and err is already true and the new err value differs from the original one. Similarly " [state was %s now %s]" is appended if $state is true and state is already true and the new state value differs from the original one. Finally "\n" and the new $errstr are appended if $errstr differs from the existing errstr value. Obviously the %s's above are replaced by the corresponding values.

The handle err value is set to $err if: $err is true; or handle err value is undef; or $err is defined and the length is greater than the handle err length. The effect is that an 'information' state only overrides undef; a 'warning' overrides undef or 'information', and an 'error' state overrides anything.

The handle state value is set to $state if $state is true and the handle err value was set (by the rules above).

Support for warning and information states was added in DBI 1.41.

The trace() method is used to alter the trace settings for a handle (and any future children of that handle). It can also be used to change where the trace output is sent.

There's a similar method, DBI->trace, which sets the global default trace settings.

See the "TRACING" section for full details about the DBI's powerful tracing facilities.

Writes $message_text to the trace file if the trace level is greater than or equal to $min_level (which defaults to 1). Can also be called as DBI->trace_msg($msg).

See "TRACING" for more details.

The func method can be used to call private non-standard and non-portable methods implemented by the driver. Note that the function name is given as the last argument.

It's also important to note that the func() method does not clear a previous error ($DBI::err etc.) and it does not trigger automatic error detection (RaiseError etc.) so you must check the return status and/or $h->err to detect errors.

(This method is not directly related to calling stored procedures. Calling stored procedures is currently not defined by the DBI. Some drivers, such as DBD::Oracle, support it in non-portable ways. See driver documentation for more details.)

See also install_method() in DBI::DBD for how you can avoid needing to use func() and gain direct access to driver-private methods.

Returns true if $method_name is implemented by the driver or a default method is provided by the DBI. It returns false where a driver hasn't implemented a method and the default method is provided by the DBI is just an empty stub.

Parses a string containing trace settings and returns the corresponding integer value used internally by the DBI and drivers.

The $trace_settings argument is a string containing a trace level between 0 and 15 and/or trace flag names separated by vertical bar ("|") or comma (",") characters. For example: "SQL|3|foo".

It uses the parse_trace_flag() method, described below, to process the individual trace flag names.

The parse_trace_flags() method was added in DBI 1.42.

Returns the bit flag corresponding to the trace flag name in $trace_flag_name. Drivers are expected to override this method and check if $trace_flag_name is a driver specific trace flags and, if not, then call the DBI's default parse_trace_flag().

The parse_trace_flag() method was added in DBI 1.42.

Returns a reference to a hash whose keys are the names of driver-private handle attributes available for the kind of handle (driver, database, statement) that the method was called on.

For example, the return value when called with a DBD::Sybase $dbh could look like this:

and when called with a DBD::Sybase $sth they could look like this:

The values should be undef. Meanings may be assigned to particular values in future.

Brain transplants for handles. You don't need to know about this unless you want to become a handle surgeon.

A DBI handle is a reference to a tied hash. A tied hash has an inner hash that actually holds the contents. The swap_inner_handle() method swaps the inner hashes between two handles. The $h1 and $h2 handles still point to the same tied hashes, but what those hashes are tied to has been swapped. In effect $h1 becomes $h2 and vice-versa. This is powerful stuff, expect problems. Use with care.

As a small safety measure, the two handles, $h1 and $h2, have to share the same parent unless $allow_reparent is true.

The swap_inner_handle() method was added in DBI 1.44.

Here's a quick kind of 'diagram' as a worked example to help think about what's happening:

Where $coderef is a reference to a subroutine and $info is an arbitrary value which, if undefined, defaults to a reference to an empty hash. Returns $info.

For each child handle of $h, if any, $coderef is invoked as:

If the execution of $coderef returns a true value then visit_child_handles is called on that child handle and passed the returned value as $info.

For example:

See also "visit_handles".

These attributes are common to all types of DBI handles.

Some attributes are inherited by child handles. That is, the value of an inherited attribute in a newly created statement handle is the same as the value in the parent database handle. Changes to attributes in the new statement handle do not affect the parent database handle and changes to the database handle do not affect existing statement handles, only future ones.

Attempting to set or get the value of an unknown attribute generates a warning, except for private driver specific attributes (which all have names starting with a lowercase letter).

Example:

The Warn attribute enables useful warnings for certain bad practices. It is enabled by default and should only be disabled in rare circumstances. Since warnings are generated using the Perl warn function, they can be intercepted using the Perl $SIG{__WARN__} hook.

The Warn attribute is not related to the PrintWarn attribute.

The Active attribute is true if the handle object is "active". This is rarely used in applications. The exact meaning of active is somewhat vague at the moment. For a database handle it typically means that the handle is connected to a database ($dbh->disconnect sets Active off). For a statement handle it typically means that the handle is a SELECT that may have more data to fetch. (Fetching all the data or calling $sth->finish sets Active off.)

The Executed attribute is true if the handle object has been "executed". Currently only the $dbh do() method and the $sth execute(), execute_array(), and execute_for_fetch() methods set the Executed attribute.

When it's set on a handle it is also set on the parent handle at the same time. So calling execute() on a $sth also sets the Executed attribute on the parent $dbh.

The Executed attribute for a database handle is cleared by the commit() and rollback() methods (even if they fail). The Executed attribute of a statement handle is not cleared by the DBI under any circumstances and so acts as a permanent record of whether the statement handle was ever used.

The Executed attribute was added in DBI 1.41.

For a driver handle, Kids is the number of currently existing database handles that were created from that driver handle. For a database handle, Kids is the number of currently existing statement handles that were created from that database handle. For a statement handle, the value is zero.

Like Kids, but only counting those that are Active (as above).

For a database handle, CachedKids returns a reference to the cache (hash) of statement handles created by the "prepare_cached" method. For a driver handle, returns a reference to the cache (hash) of database handles created by the "connect_cached" method.

The Type attribute identifies the type of a DBI handle. Returns "dr" for driver handles, "db" for database handles and "st" for statement handles.

The ChildHandles attribute contains a reference to an array of all the handles created by this handle which are still accessible. The contents of the array are weak-refs and will become undef when the handle goes out of scope.

ChildHandles returns undef if your perl version does not support weak references (check the Scalar::Util module). The referenced array returned should be treated as read-only.

For example, to enumerate all driver handles, database handles and statement handles:

The CompatMode attribute is used by emulation layers (such as Oraperl) to enable compatible behaviour in the underlying driver (e.g., DBD::Oracle) for this handle. Not normally set by application code.

It also has the effect of disabling the 'quick FETCH' of attribute values from the handles attribute cache. So all attribute values are handled by the drivers own FETCH method. This makes them slightly slower but is useful for special-purpose drivers like DBD::Multiplex.

The default value, false, means a handle will be fully destroyed as normal when the last reference to it is removed, just as you'd expect.

If set true then the handle will be treated by the DESTROY as if it was no longer Active, and so the database engine related effects of DESTROYing a handle will be skipped.

Think of the name as meaning 'treat the handle as not-Active in the DESTROY method'.

For a database handle, this attribute does not disable an explicit call to the disconnect method, only the implicit call from DESTROY that happens if the handle is still marked as Active.

This attribute is specifically designed for use in Unix applications that "fork" child processes. Either the parent or the child process, but not both, should set InactiveDestroy true on all their shared handles. (Note that some databases, including Oracle, don't support passing a database connection across a fork.)

To help tracing applications using fork the process id is shown in the trace log whenever a DBI or handle trace() method is called. The process id also shown for every method call if the DBI trace level (not handle trace level) is set high enough to show the trace from the DBI's method dispatcher, e.g. >= 9.

The PrintWarn attribute controls the printing of warnings recorded by the driver. When set to a true value the DBI will check method calls to see if a warning condition has been set. If so, the DBI will effectively do a warn("$class $method warning: $DBI::errstr") where $class is the driver class and $method is the name of the method which failed. E.g.,

By default, DBI->connect sets PrintWarn "on" if $^W is true, i.e., perl is running with warnings enabled.

If desired, the warnings can be caught and processed using a $SIG{__WARN__} handler or modules like CGI::Carp and CGI::ErrorWrap.

See also "set_err" for how warnings are recorded and "HandleSetErr" for how to influence it.

Fetching the full details of warnings can require an extra round-trip to the database server for some drivers. In which case the driver may opt to only fetch the full details of warnings if the PrintWarn attribute is true. If PrintWarn is false then these drivers should still indicate the fact that there were warnings by setting the warning string to, for example: "3 warnings".

The PrintError attribute can be used to force errors to generate warnings (using warn) in addition to returning error codes in the normal way. When set "on", any method which results in an error occurring will cause the DBI to effectively do a warn("$class $method failed: $DBI::errstr") where $class is the driver class and $method is the name of the method which failed. E.g.,

By default, DBI->connect sets PrintError "on".

If desired, the warnings can be caught and processed using a $SIG{__WARN__} handler or modules like CGI::Carp and CGI::ErrorWrap.

The RaiseError attribute can be used to force errors to raise exceptions rather than simply return error codes in the normal way. It is "off" by default. When set "on", any method which results in an error will cause the DBI to effectively do a die("$class $method failed: $DBI::errstr"), where $class is the driver class and $method is the name of the method that failed. E.g.,

If you turn RaiseError on then you'd normally turn PrintError off. If PrintError is also on, then the PrintError is done first (naturally).

Typically RaiseError is used in conjunction with eval { ... } to catch the exception that's been thrown and followed by an if ($@) { ... } block to handle the caught exception. For example:

In that eval block the $DBI::lasth variable can be useful for diagnosis and reporting if you can't be sure which handle triggered the error. For example, $DBI::lasth->{Type} and $DBI::lasth->{Statement}.

See also "Transactions".

If you want to temporarily turn RaiseError off (inside a library function that is likely to fail, for example), the recommended way is like this:

The original value will automatically and reliably be restored by Perl, regardless of how the block is exited. The same logic applies to other attributes, including PrintError.

The HandleError attribute can be used to provide your own alternative behaviour in case of errors. If set to a reference to a subroutine then that subroutine is called when an error is detected (at the same point that RaiseError and PrintError are handled).

The subroutine is called with three parameters: the error message string that RaiseError and PrintError would use, the DBI handle being used, and the first value being returned by the method that failed (typically undef).

If the subroutine returns a false value then the RaiseError and/or PrintError attributes are checked and acted upon as normal.

For example, to die with a full stack trace for any error:

Or to turn errors into exceptions:

It is possible to 'stack' multiple HandleError handlers by using closures:

Using a my inside a subroutine to store the previous HandleError value is important. See perlsub and perlref for more information about closures.

It is possible for HandleError to alter the error message that will be used by RaiseError and PrintError if it returns false. It can do that by altering the value of $_[0]. This example appends a stack trace to all errors and, unlike the previous example using Carp::confess, this will work PrintError as well as RaiseError:

It is also possible for HandleError to hide an error, to a limited degree, by using "set_err" to reset $DBI::err and $DBI::errstr, and altering the return value of the failed method. For example:

This only works for methods which return a single value and is hard to make reliable (avoiding infinite loops, for example) and so isn't recommended for general use! If you find a good use for it then please let me know.

The HandleSetErr attribute can be used to intercept the setting of handle err, errstr, and state values. If set to a reference to a subroutine then that subroutine is called whenever set_err() is called, typically by the driver or a subclass.

The subroutine is called with five arguments, the first five that were passed to set_err(): the handle, the err, errstr, and state values being set, and the method name. These can be altered by changing the values in the @_ array. The return value affects set_err() behaviour, see "set_err" for details.

It is possible to 'stack' multiple HandleSetErr handlers by using closures. See "HandleError" for an example.

The HandleSetErr and HandleError subroutines differ in subtle but significant ways. HandleError is only invoked at the point where the DBI is about to return to the application with err set true. It's not invoked by the failure of a method that's been called by another DBI method. HandleSetErr, on the other hand, is called whenever set_err() is called with a defined err value, even if false. So it's not just for errors, despite the name, but also warn and info states. The set_err() method, and thus HandleSetErr, may be called multiple times within a method and is usually invoked from deep within driver code.

In theory a driver can use the return value from HandleSetErr via set_err() to decide whether to continue or not. If set_err() returns an empty list, indicating that the HandleSetErr code has 'handled' the 'error', the driver could then continue instead of failing (if that's a reasonable thing to do). This isn't excepted to be common and any such cases should be clearly marked in the driver documentation and discussed on the dbi-dev mailing list.

The HandleSetErr attribute was added in DBI 1.41.

The ErrCount attribute is incremented whenever the set_err() method records an error. It isn't incremented by warnings or information states. It is not reset by the DBI at any time.

The ErrCount attribute was added in DBI 1.41. Older drivers may not have been updated to use set_err() to record errors and so this attribute may not be incremented when using them.

The ShowErrorStatement attribute can be used to cause the relevant Statement text to be appended to the error messages generated by the RaiseError, PrintError, and PrintWarn attributes. Only applies to errors on statement handles plus the prepare(), do(), and the various select*() database handle methods. (The exact format of the appended text is subject to change.)

If $h->{ParamValues} returns a hash reference of parameter (placeholder) values then those are formatted and appended to the end of the Statement text in the error message.

The TraceLevel attribute can be used as an alternative to the "trace" method to set the DBI trace level and trace flags for a specific handle. See "TRACING" for more details.

The TraceLevel attribute is especially useful combined with local to alter the trace settings for just a single block of code.

The FetchHashKeyName attribute is used to specify whether the fetchrow_hashref() method should perform case conversion on the field names used for the hash keys. For historical reasons it defaults to 'NAME' but it is recommended to set it to 'NAME_lc' (convert to lower case) or 'NAME_uc' (convert to upper case) according to your preference. It can only be set for driver and database handles. For statement handles the value is frozen when prepare() is called.

The ChopBlanks attribute can be used to control the trimming of trailing space characters from fixed width character (CHAR) fields. No other field types are affected, even where field values have trailing spaces.

The default is false (although it is possible that the default may change). Applications that need specific behaviour should set the attribute as needed.

Drivers are not required to support this attribute, but any driver which does not support it must arrange to return undef as the attribute value.

The LongReadLen attribute may be used to control the maximum length of 'long' type fields (LONG, BLOB, CLOB, MEMO, etc.) which the driver will read from the database automatically when it fetches each row of data.

The LongReadLen attribute only relates to fetching and reading long values; it is not involved in inserting or updating them.

A value of 0 means not to automatically fetch any long data. Drivers may return undef or an empty string for long fields when LongReadLen is 0.

The default is typically 0 (zero) or 80 bytes but may vary between drivers. Applications fetching long fields should set this value to slightly larger than the longest long field value to be fetched.

Some databases return some long types encoded as pairs of hex digits. For these types, LongReadLen relates to the underlying data length and not the doubled-up length of the encoded string.

Changing the value of LongReadLen for a statement handle after it has been prepare'd will typically have no effect, so it's common to set LongReadLen on the $dbh before calling prepare.

For most drivers the value used here has a direct effect on the memory used by the statement handle while it's active, so don't be too generous. If you can't be sure what value to use you could execute an extra select statement to determine the longest value. For example:

You may need to take extra care if the table can be modified between the first select and the second being executed. You may also need to use a different function if OCTET_LENGTH() does not work for long types in your database. For example, for Sybase use DATALENGTH() and for Oracle use LENGTHB().

See also "LongTruncOk" for information on truncation of long types.

The LongTruncOk attribute may be used to control the effect of fetching a long field value which has been truncated (typically because it's longer than the value of the LongReadLen attribute).

By default, LongTruncOk is false and so fetching a long value that needs to be truncated will cause the fetch to fail. (Applications should always be sure to check for errors after a fetch loop in case an error, such as a divide by zero or long field truncation, caused the fetch to terminate prematurely.)

If a fetch fails due to a long field truncation when LongTruncOk is false, many drivers will allow you to continue fetching further rows.

See also "LongReadLen".

If the TaintIn attribute is set to a true value and Perl is running in taint mode (e.g., started with the -T option), then all the arguments to most DBI method calls are checked for being tainted. This may change.

The attribute defaults to off, even if Perl is in taint mode. See perlsec for more about taint mode. If Perl is not running in taint mode, this attribute has no effect.

When fetching data that you trust you can turn off the TaintIn attribute, for that statement handle, for the duration of the fetch loop.

The TaintIn attribute was added in DBI 1.31.

If the TaintOut attribute is set to a true value and Perl is running in taint mode (e.g., started with the -T option), then most data fetched from the database is considered tainted. This may change.

The attribute defaults to off, even if Perl is in taint mode. See perlsec for more about taint mode. If Perl is not running in taint mode, this attribute has no effect.

When fetching data that you trust you can turn off the TaintOut attribute, for that statement handle, for the duration of the fetch loop.

Currently only fetched data is tainted. It is possible that the results of other DBI method calls, and the value of fetched attributes, may also be tainted in future versions. That change may well break your applications unless you take great care now. If you use DBI Taint mode, please report your experience and any suggestions for changes.

The TaintOut attribute was added in DBI 1.31.

The Taint attribute is a shortcut for "TaintIn" and "TaintOut" (it is also present for backwards compatibility).

Setting this attribute sets both "TaintIn" and "TaintOut", and retrieving it returns a true value if and only if "TaintIn" and "TaintOut" are both set to true values.

The Profile attribute enables the collection and reporting of method call timing statistics. See the DBI::Profile module documentation for much more detail.

The Profile attribute was added in DBI 1.24.

An application can set the ReadOnly attribute of a handle to a true value to indicate that it will not be attempting to make any changes using that handle or any children of it.

Note that the exact definition of 'read only' is rather fuzzy. For more details see the documentation for the driver you're using.

If the driver can make the handle truly read-only then it should (unless doing so would have unpleasant side effect, like changing the consistency level from per-statement to per-session). Otherwise the attribute is simply advisory.

A driver can set the ReadOnly attribute itself to indicate that the data it is connected to cannot be changed for some reason.

Library modules and proxy drivers can use the attribute to influence their behavior. For example, the DBD::Gofer driver considers the ReadOnly attribute when making a decision about whether to retry an operation that failed.

The attribute should be set to 1 or 0 (or undef). Other values are reserved.

The DBI callback mechanism lets you intercept, and optionally replace, any method call on a DBI handle. At the extreme, it lets you become a puppet master, deceiving the application in any way you want.

The Callbacks attribute is a hash reference where the keys are DBI method names and the values are code references. For each key naming a method, the DBI will execute the associated code reference before executing the method.

The arguments to the code reference will be the same as to the method, including the invocant (a database handle or statement handle). For example, say that to callback to some code on a call to prepare():

The callback would then be executed when you called the prepare() method:

And the output of course would be:

Because callbacks are executed before the methods they're associated with, you can modify the arguments before they're passed on to the method call. For example, to make sure that all calls to prepare() are immediately prepared by DBD::Pg, add a callback that makes sure that the pg_prepare_now attribute is always set:

Note that we are editing the contents of @_ directly. In this case we've created the attributes hash if it's not passed to the prepare call.

You can also prevent the associated method from ever executing. While a callback executes, $_ holds the method name. (This allows multiple callbacks to share the same code reference and still know what method was called.) To prevent the method from executing, simply undef $_. For example, if you wanted to disable calls to ping(), you could do this:

As with other attributes, Callbacks can be specified on a handle or via the attributes to connect(). Callbacks can also be applied to a statement methods on a statement handle. For example:

The Callbacks attribute of a database handle isn't copied to any statement handles it creates. So setting callbacks for a statement handle requires you to set the Callbacks attribute on the statement handle yourself, as in the example above, or use the special ChildCallbacks key described below.

Special Keys in Callbacks Attribute

In addition to DBI handle method names, the Callbacks hash reference supports three additional keys.

The first is the ChildCallbacks key. When a statement handle is created from a database handle the ChildCallbacks key of the database handle's Callbacks attribute, if any, becomes the new Callbacks attribute of the statement handle. This allows you to define callbacks for all statement handles created from a database handle. For example, if you wanted to count how many times execute was called in your application, you could write:

The other two special keys are connect_cached.new and connect_cached.reused. These keys define callbacks that are called when connect_cached() is called, but allow different behaviors depending on whether a new handle is created or a handle is returned. The callback is invoked with these arguments: $dbh, $dsn, $user, $auth, $attr.

For example, some applications uses connect_cached() to connect with AutoCommit enabled and then disable AutoCommit temporarily for transactions. If connect_cached() is called during a transaction, perhaps in a utility method, then it might select the same cached handle and then force AutoCommit on, forcing a commit of the transaction. See the "connect_cached" documentation for one way to deal with that. Here we'll describe an alternative approach using a callback.

Because the connect_cached.* callbacks are invoked before connect_cached() has applied the connect attributes you can use a callback to edit the attributes that will be applied. To prevent a cached handle from having its transactions committed before it's returned, you can eliminate the AutoCommit attribute in a connect_cached.reused callback, like so:

The upshot is that new database handles are created with AutoCommit enabled, while cached database handles are left in whatever transaction state they happened to be in when retrieved from the cache.

A more common application for callbacks is setting connection state only when a new connection is made (by connect() or connect_cached()). Adding a callback to the connected method makes this easy. This method is a no-op by default (unless you subclass the DBI and change it). The DBI calls it to indicate that a new connection has been made and the connection attributes have all been set. You can give it a bit of added functionality by applying a callback to it. For example, to make sure that MySQL understands your application's ANSI-compliant SQL, set it up like so:

One significant limitation with callbacks is that there can only be one per method per handle. This means it's easy for one use of callbacks to interfere with, or typically simply overwrite, another use of callbacks. For this reason modules using callbacks should document the fact clearly so application authors can tell if use of callbacks by the module will clash with use of callbacks by the application.

You might be able to work around this issue by taking a copy of the original callback and calling it within your own. For example:

The DBI provides a way to store extra information in a DBI handle as "private" attributes. The DBI will allow you to store and retrieve any attribute which has a name starting with "private_".

It is strongly recommended that you use just one private attribute (e.g., use a hash ref) and give it a long and unambiguous name that includes the module or application name that the attribute relates to (e.g., "private_YourFullModuleName_thingy").

Because of the way the Perl tie mechanism works you cannot reliably use the ||= operator directly to initialise the attribute, like this:

you should use a two step approach like this:

This attribute is primarily of interest to people sub-classing DBI, or for applications to piggy-back extra information onto DBI handles.

This section covers the methods and attributes associated with database handles.

The following methods are specified for DBI database handles:

The clone method duplicates the $dbh connection by connecting with the same parameters ($dsn, $user, $password) as originally used.

The attributes for the cloned connect are the same as those used for the original connect, with some other attributes merged over them depending on the \%attr parameter.

If \%attr is given then the attributes it contains are merged into the original attributes and override any with the same names. Effectively the same as doing:

If \%attr is not given then it defaults to a hash containing all the attributes in the attribute cache of $dbh excluding any non-code references, plus the main boolean attributes (RaiseError, PrintError, AutoCommit, etc.). This behaviour is subject to change.

The clone method can be used even if the database handle is disconnected.

The clone method was added in DBI 1.33. It is very new and likely to change.

Returns a list of data sources (databases) available via the $dbh driver's data_sources() method, plus any extra data sources that the driver can discover via the connected $dbh. Typically the extra data sources are other databases managed by the same server process that the $dbh is connected to.

Data sources are returned in a form suitable for passing to the "connect" method (that is, they will include the "dbi:$driver:" prefix).

The data_sources() method, for a $dbh, was added in DBI 1.38.

Prepare and execute a single statement. Returns the number of rows affected or undef on error. A return value of -1 means the number of rows is not known, not applicable, or not available.

This method is typically most useful for non-SELECT statements that either cannot be prepared in advance (due to a limitation of the driver) or do not need to be executed repeatedly. It should not be used for SELECT statements because it does not return a statement handle (so you can't fetch any data).

The default do method is logically similar to:

For example:

Using placeholders and @bind_values with the do method can be useful because it avoids the need to correctly quote any variables in the $statement. But if you'll be executing the statement many times then it's more efficient to prepare it once and call execute many times instead.

The q{...} style quoting used in this example avoids clashing with quotes that may be used in the SQL statement. Use the double-quote-like qq{...} operator if you want to interpolate variables into the string. See "Quote and Quote-like Operators" in perlop for more details.

Returns a value 'identifying' the row just inserted, if possible. Typically this would be a value assigned by the database server to a column with an auto_increment or serial type. Returns undef if the driver does not support the method or can't determine the value.

The $catalog, $schema, $table, and $field parameters may be required for some drivers (see below). If you don't know the parameter values and your driver does not need them, then use undef for each.

There are several caveats to be aware of with this method if you want to use it for portable applications:

* For some drivers the value may only available immediately after the insert statement has executed (e.g., mysql, Informix).

* For some drivers the $catalog, $schema, $table, and $field parameters are required, for others they are ignored (e.g., mysql).

* Drivers may return an indeterminate value if no insert has been performed yet.

* For some drivers the value may only be available if placeholders have not been used (e.g., Sybase, MS SQL). In this case the value returned would be from the last non-placeholder insert statement.

* Some drivers may need driver-specific hints about how to get the value. For example, being told the name of the database 'sequence' object that holds the value. Any such hints are passed as driver-specific attributes in the \%attr parameter.

* If the underlying database offers nothing better, then some drivers may attempt to implement this method by executing "select max($field) from $table". Drivers using any approach like this should issue a warning if AutoCommit is true because it is generally unsafe - another process may have modified the table between your insert and the select. For situations where you know it is safe, such as when you have locked the table, you can silence the warning by passing Warn => 0 in \%attr.

* If no insert has been performed yet, or the last insert failed, then the value is implementation defined.

Given all the caveats above, it's clear that this method must be used with care.

The last_insert_id method was added in DBI 1.38.

This utility method combines "prepare", "execute" and "fetchrow_array" into a single call. If called in a list context, it returns the first row of data from the statement. The $statement parameter can be a previously prepared statement handle, in which case the prepare is skipped.

If any method fails, and "RaiseError" is not set, selectrow_array will return an empty list.

If called in a scalar context for a statement handle that has more than one column, it is undefined whether the driver will return the value of the first column or the last. So don't do that. Also, in a scalar context, an undef is returned if there are no more rows or if an error occurred. That undef can't be distinguished from an undef returned because the first field value was NULL. For these reasons you should exercise some caution if you use selectrow_array in a scalar context, or just don't do that.

This utility method combines "prepare", "execute" and "fetchrow_arrayref" into a single call. It returns the first row of data from the statement. The $statement parameter can be a previously prepared statement handle, in which case the prepare is skipped.

If any method fails, and "RaiseError" is not set, selectrow_array will return undef.

This utility method combines "prepare", "execute" and "fetchrow_hashref" into a single call. It returns the first row of data from the statement. The $statement parameter can be a previously prepared statement handle, in which case the prepare is skipped.

If any method fails, and "RaiseError" is not set, selectrow_hashref will return undef.

This utility method combines "prepare", "execute" and "fetchall_arrayref" into a single call. It returns a reference to an array containing a reference to an array (or hash, see below) for each row of data fetched.

The $statement parameter can be a previously prepared statement handle, in which case the prepare is skipped. This is recommended if the statement is going to be executed many times.

If "RaiseError" is not set and any method except fetchall_arrayref fails then selectall_arrayref will return undef; if fetchall_arrayref fails then it will return with whatever data has been fetched thus far. You should check $sth->err afterwards (or use the RaiseError attribute) to discover if the data is complete or was truncated due to an error.

The "fetchall_arrayref" method called by selectall_arrayref supports a $max_rows parameter. You can specify a value for $max_rows by including a 'MaxRows' attribute in \%attr. In which case finish() is called for you after fetchall_arrayref() returns.

The "fetchall_arrayref" method called by selectall_arrayref also supports a $slice parameter. You can specify a value for $slice by including a 'Slice' or 'Columns' attribute in \%attr. The only difference between the two is that if Slice is not defined and Columns is an array ref, then the array is assumed to contain column index values (which count from 1), rather than perl array index values. In which case the array is copied and each value decremented before passing to /fetchall_arrayref.

You may often want to fetch an array of rows where each row is stored as a hash. That can be done simple using:

Or, to fetch into an array instead of an array ref:

See "fetchall_arrayref" method for more details.

This utility method combines "prepare", "execute" and "fetchall_hashref" into a single call. It returns a reference to a hash containing one entry, at most, for each row, as returned by fetchall_hashref().

The $statement parameter can be a previously prepared statement handle, in which case the prepare is skipped. This is recommended if the statement is going to be executed many times.

The $key_field parameter defines which column, or columns, are used as keys in the returned hash. It can either be the name of a single field, or a reference to an array containing multiple field names. Using multiple names yields a tree of nested hashes.

If a row has the same key as an earlier row then it replaces the earlier row.

If any method except fetchrow_hashref fails, and "RaiseError" is not set, selectall_hashref will return undef. If fetchrow_hashref fails and "RaiseError" is not set, then it will return with whatever data it has fetched thus far. $DBI::err should be checked to catch that.

See fetchall_hashref() for more details.

This utility method combines "prepare", "execute", and fetching one column from all the rows, into a single call. It returns a reference to an array containing the values of the first column from each row.

The $statement parameter can be a previously prepared statement handle, in which case the prepare is skipped. This is recommended if the statement is going to be executed many times.

If any method except fetch fails, and "RaiseError" is not set, selectcol_arrayref will return undef. If fetch fails and "RaiseError" is not set, then it will return with whatever data it has fetched thus far. $DBI::err should be checked to catch that.

The selectcol_arrayref method defaults to pushing a single column value (the first) from each row into the result array. However, it can also push another column, or even multiple columns per row, into the result array. This behaviour can be specified via a 'Columns' attribute which must be a ref to an array containing the column number or numbers to use. For example:

You can specify a maximum number of rows to fetch by including a 'MaxRows' attribute in \%attr.

Prepares a statement for later execution by the database engine and returns a reference to a statement handle object.

The returned statement handle can be used to get attributes of the statement and invoke the "execute" method. See "Statement Handle Methods".

Drivers for engines without the concept of preparing a statement will typically just store the statement in the returned handle and process it when $sth->execute is called. Such drivers are unlikely to give much useful information about the statement, such as $sth->{NUM_OF_FIELDS}, until after $sth->execute has been called. Portable applications should take this into account.

In general, DBI drivers do not parse the contents of the statement (other than simply counting any "Placeholders"). The statement is passed directly to the database engine, sometimes known as pass-thru mode. This has advantages and disadvantages. On the plus side, you can access all the functionality of the engine being used. On the downside, you're limited if you're using a simple engine, and you need to take extra care if writing applications intended to be portable between engines.

Portable applications should not assume that a new statement can be prepared and/or executed while still fetching results from a previous statement.

Some command-line SQL tools use statement terminators, like a semicolon, to indicate the end of a statement. Such terminators should not normally be used with the DBI.

Like "prepare" except that the statement handle returned will be stored in a hash associated with the $dbh. If another call is made to prepare_cached with the same $statement and %attr parameter values, then the corresponding cached $sth will be returned without contacting the database server.

The $if_active parameter lets you adjust the behaviour if an already cached statement handle is still Active. There are several alternatives:

Here are some examples of prepare_cached:

Caveat emptor: This caching can be useful in some applications, but it can also cause problems and should be used with care. Here is a contrived case where caching would cause a significant problem:

In this example, since both handles are preparing the exact same statement, $sth2 will not be its own statement handle, but a duplicate of $sth returned from the cache. The results will certainly not be what you expect. Typically the inner fetch loop will work normally, fetching all the records and terminating when there are no more, but now that $sth is the same as $sth2 the outer fetch loop will also terminate.

You'll know if you run into this problem because prepare_cached() will generate a warning by default (when $if_active is false).

The cache used by prepare_cached() is keyed by both the statement and any attributes so you can also avoid this issue by doing something like:

which will ensure that prepare_cached only returns statements cached by that line of code in that source file.

If you'd like the cache to managed intelligently, you can tie the hashref returned by CachedKids to an appropriate caching module, such as Tie::Cache::LRU:

Commit (make permanent) the most recent series of database changes if the database supports transactions and AutoCommit is off.

If AutoCommit is on, then calling commit will issue a "commit ineffective with AutoCommit" warning.

See also "Transactions" in the "FURTHER INFORMATION" section below.

Rollback (undo) the most recent series of uncommitted database changes if the database supports transactions and AutoCommit is off.

If AutoCommit is on, then calling rollback will issue a "rollback ineffective with AutoCommit" warning.

See also "Transactions" in the "FURTHER INFORMATION" section below.

Enable transactions (by turning AutoCommit off) until the next call to commit or rollback. After the next commit or rollback, AutoCommit will automatically be turned on again.

If AutoCommit is already off when begin_work is called then it does nothing except return an error. If the driver does not support transactions then when begin_work attempts to set AutoCommit off the driver will trigger a fatal error.

See also "Transactions" in the "FURTHER INFORMATION" section below.

Disconnects the database from the database handle. disconnect is typically only used before exiting the program. The handle is of little use after disconnecting.

The transaction behaviour of the disconnect method is, sadly, undefined. Some database systems (such as Oracle and Ingres) will automatically commit any outstanding changes, but others (such as Informix) will rollback any outstanding changes. Applications not using AutoCommit should explicitly call commit or rollback before calling disconnect.

The database is automatically disconnected by the DESTROY method if still connected when there are no longer any references to the handle. The DESTROY method for each driver should implicitly call rollback to undo any uncommitted changes. This is vital behaviour to ensure that incomplete transactions don't get committed simply because Perl calls DESTROY on every object before exiting. Also, do not rely on the order of object destruction during "global destruction", as it is undefined.

Generally, if you want your changes to be committed or rolled back when you disconnect, then you should explicitly call "commit" or "rollback" before disconnecting.

If you disconnect from a database while you still have active statement handles (e.g., SELECT statement handles that may have more data to fetch), you will get a warning. The warning may indicate that a fetch loop terminated early, perhaps due to an uncaught error. To avoid the warning call the finish method on the active handles.

Attempts to determine, in a reasonably efficient way, if the database server is still running and the connection to it is still working. Individual drivers should implement this function in the most suitable manner for their database engine.

The current default implementation always returns true without actually doing anything. Actually, it returns "0 but true" which is true but zero. That way you can tell if the return value is genuine or just the default. Drivers should override this method with one that does the right thing for their type of database.

Few applications would have direct use for this method. See the specialized Apache::DBI module for one example usage.

Returns information about the implementation, i.e. driver and data source capabilities, restrictions etc. It returns undef for unknown or unimplemented information types. For example:

See "Standards Reference Information" for more detailed information about the information types and their meanings and possible return values.

The DBI::Const::GetInfoType module exports a %GetInfoType hash that can be used to map info type names to numbers. For example:

The names are a merging of the ANSI and ODBC standards (which differ in some cases). See DBI::Const::GetInfoType for more details.

Because some DBI methods make use of get_info(), drivers are strongly encouraged to support at least the following very minimal set of information types to ensure the DBI itself works properly:

Returns an active statement handle that can be used to fetch information about tables and views that exist in the database.

The arguments $catalog, $schema and $table may accept search patterns according to the database/driver, for example: $table = '%FOO%'; Remember that the underscore character ('_') is a search pattern that means match any character, so 'FOO_%' is the same as 'FOO%' and 'FOO_BAR%' will match names like 'FOO1BAR'.

The value of $type is a comma-separated list of one or more types of tables to be returned in the result set. Each value may optionally be quoted, e.g.:

In addition the following special cases may also be supported by some drivers:

If your driver doesn't support one or more of the selection filter parameters then you may get back more than you asked for and can do the filtering yourself.

This method can be expensive, and can return a large amount of data. (For example, small Oracle installation returns over 2000 rows.) So it's a good idea to use the filters to limit the data as much as possible.

The statement handle returned has at least the following fields in the order show below. Other fields, after these, may also be present.

TABLE_CAT: Table catalog identifier. This field is NULL (undef) if not applicable to the data source, which is usually the case. This field is empty if not applicable to the table.

TABLE_SCHEM: The name of the schema containing the TABLE_NAME value. This field is NULL (undef) if not applicable to data source, and empty if not applicable to the table.

TABLE_NAME: Name of the table (or view, synonym, etc).

TABLE_TYPE: One of the following: "TABLE", "VIEW", "SYSTEM TABLE", "GLOBAL TEMPORARY", "LOCAL TEMPORARY", "ALIAS", "SYNONYM" or a type identifier that is specific to the data source.

REMARKS: A description of the table. May be NULL (undef).

Note that table_info might not return records for all tables. Applications can use any valid table regardless of whether it's returned by table_info.

See also "tables", "Catalog Methods" and "Standards Reference Information".

Returns an active statement handle that can be used to fetch information about columns in specified tables.

The arguments $schema, $table and $column may accept search patterns according to the database/driver, for example: $table = '%FOO%';

Note: The support for the selection criteria is driver specific. If the driver doesn't support one or more of them then you may get back more than you asked for and can do the filtering yourself.

Note: If your driver does not support column_info an undef is returned. This is distinct from asking for something which does not exist in a driver which supports column_info as a valid statement handle to an empty result-set will be returned in this case.

If the arguments don't match any tables then you'll still get a statement handle, it'll just return no rows.

The statement handle returned has at least the following fields in the order shown below. Other fields, after these, may also be present.

TABLE_CAT: The catalog identifier. This field is NULL (undef) if not applicable to the data source, which is often the case. This field is empty if not applicable to the table.

TABLE_SCHEM: The schema identifier. This field is NULL (undef) if not applicable to the data source, and empty if not applicable to the table.

TABLE_NAME: The table identifier. Note: A driver may provide column metadata not only for base tables, but also for derived objects like SYNONYMS etc.

COLUMN_NAME: The column identifier.

DATA_TYPE: The concise data type code.

TYPE_NAME: A data source dependent data type name.

COLUMN_SIZE: The column size. This is the maximum length in characters for character data types, the number of digits or bits for numeric data types or the length in the representation of temporal types. See the relevant specifications for detailed information.

BUFFER_LENGTH: The length in bytes of transferred data.

DECIMAL_DIGITS: The total number of significant digits to the right of the decimal point.

NUM_PREC_RADIX: The radix for numeric precision. The value is 10 or 2 for numeric data types and NULL (undef) if not applicable.

NULLABLE: Indicates if a column can accept NULLs. The following values are defined:

REMARKS: A description of the column.

COLUMN_DEF: The default value of the column, in a format that can be used directly in an SQL statement.

Note that this may be an expression and not simply the text used for the default value in the original CREATE TABLE statement. For example, given:

where "current_user" is the name of a function, the corresponding COLUMN_DEF values would be:

SQL_DATA_TYPE: The SQL data type.

SQL_DATETIME_SUB: The subtype code for datetime and interval data types.

CHAR_OCTET_LENGTH: The maximum length in bytes of a character or binary data type column.

ORDINAL_POSITION: The column sequence number (starting with 1).

IS_NULLABLE: Indicates if the column can accept NULLs. Possible values are: 'NO', 'YES' and ''.

SQL/CLI defines the following additional columns:

Drivers capable of supplying any of those values should do so in the corresponding column and supply undef values for the others.

Drivers wishing to provide extra database/driver specific information should do so in extra columns beyond all those listed above, and use lowercase field names with the driver-specific prefix (i.e., 'ora_...'). Applications accessing such fields should do so by name and not by column number.

The result set is ordered by TABLE_CAT, TABLE_SCHEM, TABLE_NAME and ORDINAL_POSITION.

Note: There is some overlap with statement handle attributes (in perl) and SQLDescribeCol (in ODBC). However, SQLColumns provides more metadata.

See also "Catalog Methods" and "Standards Reference Information".

Returns an active statement handle that can be used to fetch information about columns that make up the primary key for a table. The arguments don't accept search patterns (unlike table_info()).

The statement handle will return one row per column, ordered by TABLE_CAT, TABLE_SCHEM, TABLE_NAME, and KEY_SEQ. If there is no primary key then the statement handle will fetch no rows.

Note: The support for the selection criteria, such as $catalog, is driver specific. If the driver doesn't support catalogs and/or schemas, it may ignore these criteria.

The statement handle returned has at least the following fields in the order shown below. Other fields, after these, may also be present.

TABLE_CAT: The catalog identifier. This field is NULL (undef) if not applicable to the data source, which is often the case. This field is empty if not applicable to the table.

TABLE_SCHEM: The schema identifier. This field is NULL (undef) if not applicable to the data source, and empty if not applicable to the table.

TABLE_NAME: The table identifier.

COLUMN_NAME: The column identifier.

KEY_SEQ: The column sequence number (starting with 1). Note: This field is named ORDINAL_POSITION in SQL/CLI.

PK_NAME: The primary key constraint identifier. This field is NULL (undef) if not applicable to the data source.

See also "Catalog Methods" and "Standards Reference Information".

Simple interface to the primary_key_info() method. Returns a list of the column names that comprise the primary key of the specified table. The list is in primary key column sequence order. If there is no primary key then an empty list is returned.

Returns an active statement handle that can be used to fetch information about foreign keys in and/or referencing the specified table(s). The arguments don't accept search patterns (unlike table_info()).

$pk_catalog, $pk_schema, $pk_table identify the primary (unique) key table (PKT).

$fk_catalog, $fk_schema, $fk_table identify the foreign key table (FKT).

If both PKT and FKT are given, the function returns the foreign key, if any, in table FKT that refers to the primary (unique) key of table PKT. (Note: In SQL/CLI, the result is implementation-defined.)

If only PKT is given, then the result set contains the primary key of that table and all foreign keys that refer to it.

If only FKT is given, then the result set contains all foreign keys in that table and the primary keys to which they refer. (Note: In SQL/CLI, the result includes unique keys too.)

For example:

Note: The support for the selection criteria, such as $catalog, is driver specific. If the driver doesn't support catalogs and/or schemas, it may ignore these criteria.

The statement handle returned has the following fields in the order shown below. Because ODBC never includes unique keys, they define different columns in the result set than SQL/CLI. SQL/CLI column names are shown in parentheses.

PKTABLE_CAT ( UK_TABLE_CAT ): The primary (unique) key table catalog identifier. This field is NULL (undef) if not applicable to the data source, which is often the case. This field is empty if not applicable to the table.

PKTABLE_SCHEM ( UK_TABLE_SCHEM ): The primary (unique) key table schema identifier. This field is NULL (undef) if not applicable to the data source, and empty if not applicable to the table.

PKTABLE_NAME ( UK_TABLE_NAME ): The primary (unique) key table identifier.

PKCOLUMN_NAME (UK_COLUMN_NAME ): The primary (unique) key column identifier.

FKTABLE_CAT ( FK_TABLE_CAT ): The foreign key table catalog identifier. This field is NULL (undef) if not applicable to the data source, which is often the case. This field is empty if not applicable to the table.

FKTABLE_SCHEM ( FK_TABLE_SCHEM ): The foreign key table schema identifier. This field is NULL (undef) if not applicable to the data source, and empty if not applicable to the table.

FKTABLE_NAME ( FK_TABLE_NAME ): The foreign key table identifier.

FKCOLUMN_NAME ( FK_COLUMN_NAME ): The foreign key column identifier.

KEY_SEQ ( ORDINAL_POSITION ): The column sequence number (starting with 1).

UPDATE_RULE ( UPDATE_RULE ): The referential action for the UPDATE rule. The following codes are defined:

DELETE_RULE ( DELETE_RULE ): The referential action for the DELETE rule. The codes are the same as for UPDATE_RULE.

FK_NAME ( FK_NAME ): The foreign key name.

PK_NAME ( UK_NAME ): The primary (unique) key name.

DEFERRABILITY ( DEFERABILITY ): The deferrability of the foreign key constraint. The following codes are defined:

( UNIQUE_OR_PRIMARY ): This column is necessary if a driver includes all candidate (i.e. primary and alternate) keys in the result set (as specified by SQL/CLI). The value of this column is UNIQUE if the foreign key references an alternate key and PRIMARY if the foreign key references a primary key, or it may be undefined if the driver doesn't have access to the information.

See also "Catalog Methods" and "Standards Reference Information".

Warning: This method is experimental and may change.

Returns an active statement handle that can be used to fetch statistical information about a table and its indexes.

The arguments don't accept search patterns (unlike "table_info").

If the boolean argument $unique_only is true, only UNIQUE indexes will be returned in the result set, otherwise all indexes will be returned.

If the boolean argument $quick is set, the actual statistical information columns (CARDINALITY and PAGES) will only be returned if they are readily available from the server, and might not be current. Some databases may return stale statistics or no statistics at all with this flag set.

The statement handle will return at most one row per column name per index, plus at most one row for the entire table itself, ordered by NON_UNIQUE, TYPE, INDEX_QUALIFIER, INDEX_NAME, and ORDINAL_POSITION.

Note: The support for the selection criteria, such as $catalog, is driver specific. If the driver doesn't support catalogs and/or schemas, it may ignore these criteria.

The statement handle returned has at least the following fields in the order shown below. Other fields, after these, may also be present.

TABLE_CAT: The catalog identifier. This field is NULL (undef) if not applicable to the data source, which is often the case. This field is empty if not applicable to the table.

TABLE_SCHEM: The schema identifier. This field is NULL (undef) if not applicable to the data source, and empty if not applicable to the table.

TABLE_NAME: The table identifier.

NON_UNIQUE: Unique index indicator. Returns 0 for unique indexes, 1 for non-unique indexes

INDEX_QUALIFIER: Index qualifier identifier. The identifier that is used to qualify the index name when doing a DROP INDEX; NULL (undef) is returned if an index qualifier is not supported by the data source. If a non-NULL (defined) value is returned in this column, it must be used to qualify the index name on a DROP INDEX statement; otherwise, the TABLE_SCHEM should be used to qualify the index name.

INDEX_NAME: The index identifier.

TYPE: The type of information being returned. Can be any of the following values: 'table', 'btree', 'clustered', 'content', 'hashed', or 'other'.

In the case that this field is 'table', all fields other than TABLE_CAT, TABLE_SCHEM, TABLE_NAME, TYPE, CARDINALITY, and PAGES will be NULL (undef).

ORDINAL_POSITION: Column sequence number (starting with 1).

COLUMN_NAME: The column identifier.

ASC_OR_DESC: Column sort sequence. A for Ascending, D for Descending, or NULL (undef) if not supported for this index.

CARDINALITY: Cardinality of the table or index. For indexes, this is the number of unique values in the index. For tables, this is the number of rows in the table. If not supported, the value will be NULL (undef).

PAGES: Number of storage pages used by this table or index. If not supported, the value will be NULL (undef).

FILTER_CONDITION: The index filter condition as a string. If the index is not a filtered index, or it cannot be determined whether the index is a filtered index, this value is NULL (undef). If the index is a filtered index, but the filter condition cannot be determined, this value is the empty string ''. Otherwise it will be the literal filter condition as a string, such as SALARY <= 4500.

See also "Catalog Methods" and "Standards Reference Information".

Simple interface to table_info(). Returns a list of matching table names, possibly including a catalog/schema prefix.

See "table_info" for a description of the parameters.

If $dbh->get_info(29) returns true (29 is SQL_IDENTIFIER_QUOTE_CHAR) then the table names are constructed and quoted by "quote_identifier" to ensure they are usable even if they contain whitespace or reserved words etc. This means that the table names returned will include quote characters.

Returns a reference to an array which holds information about each data type variant supported by the database and driver. The array and its contents should be treated as read-only.

The first item is a reference to an 'index' hash of Name => Index pairs. The items following that are references to arrays, one per supported data type variant. The leading index hash defines the names and order of the fields within the arrays that follow it. For example:

More than one row may have the same value in the DATA_TYPE field if there are different ways to spell the type name and/or there are variants of the type with different attributes (e.g., with and without AUTO_UNIQUE_VALUE set, with and without UNSIGNED_ATTRIBUTE, etc).

The rows are ordered by DATA_TYPE first and then by how closely each type maps to the corresponding ODBC SQL data type, closest first.

The meaning of the fields is described in the documentation for the "type_info" method.

An 'index' hash is provided so you don't need to rely on index values defined above. However, using DBD::ODBC with some old ODBC drivers may return older names, shown as comments in the example above. Another issue with the index hash is that the lettercase of the keys is not defined. It is usually uppercase, as show here, but drivers may return names with any lettercase.

Drivers are also free to return extra driver-specific columns of information - though it's recommended that they start at column index 50 to leave room for expansion of the DBI/ODBC specification.

The type_info_all() method is not normally used directly. The "type_info" method provides a more usable and useful interface to the data.

Returns a list of hash references holding information about one or more variants of $data_type. The list is ordered by DATA_TYPE first and then by how closely each type maps to the corresponding ODBC SQL data type, closest first. If called in a scalar context then only the first (best) element is returned.

If $data_type is undefined or SQL_ALL_TYPES, then the list will contain hashes for all data type variants supported by the database and driver.

If $data_type is an array reference then type_info returns the information for the first type in the array that has any matches.

The keys of the hash follow the same letter case conventions as the rest of the DBI (see "Naming Conventions and Name Space"). The following uppercase items should always exist, though may be undef: