m4_comment([$Id: db_associate.so,v 10.47 2007/10/24 16:06:06 bostic Exp $])

define(M4PAGELOCAL,
    [dbh_associate, DB_DBT_APPMALLOC, DB_DBT_MULTIPLE, DB_DONOTINDEX,
    DB_IMMUTABLE_KEY])
include(m4/m4.seealso)

m4_pf_header(m4_ref(dbh_associate),
ifelse(M4API, C_API, [dnl
int
DB-__GT__associate(DB *primary, DB_TXN *txnid, DB *secondary,
    int (*callback)(DB *secondary,
    const DBT *key, const DBT *data, DBT *result), u_int32_t flags);])
ifelse(M4API, CXX_API, [dnl
int
Db::associate(DbTxn *txnid, Db *secondary,
    int (*callback)(Db *secondary,
    const Dbt *key, const Dbt *data, Dbt *result), u_int32_t flags);])
)

m4_p([dnl
The m4_ref(dbh_associate) function is used to declare one database a
secondary index for a primary database.  After a secondary database has
been "associated" with a primary database, all updates to the primary
will be automatically reflected in the secondary and all reads from the
secondary will return corresponding data from the primary.  Note that
as primary keys must be unique for secondary indices to work, the
primary database must be configured without support for duplicate data
items.  See m4_link(M4RELDIR/ref/am/second, [Secondary indices]) for
more information.])

m4_return(dbh_associate, std)

m4_parambegin
m4_param(callback, [dnl
The m4_arg(callback) parameter is a callback function that creates the
set of secondary keys corresponding to a given primary key and data
pair.
m4_p([dnl
The callback parameter may be NULL if both the primary and secondary
database handles were opened with the m4_ref(DB_RDONLY) flag.])

m4_p([dnl
The callback takes four arguments:])
m4_tagbegin
m4_tag(m4_arg(secondary), [dnl
The m4_arg(secondary) parameter is the database handle for the secondary.])
m4_tag(m4_arg(key), [dnl
The m4_arg(key) parameter is a m4_ref(Dbt) referencing the primary key.])
m4_tag(m4_arg(data), [dnl
The m4_arg(data) parameter is a m4_ref(Dbt) referencing the primary data
item.])
m4_tag(m4_arg(result), [dnl
The m4_arg(result) parameter is a zeroed m4_ref(Dbt) in which the callback
function should fill in m4_arg(data) and m4_arg(size) fields that describe
the secondary key or keys.])
m4_tagend

m4_p([dnl
The callback optionally returns some special values:])
m4_tagbegin
m4_tag(m4_idef(DB_DBT_APPMALLOC), [dnl
If the callback function needs to allocate memory for the m4_arg(result)
data field (rather than simply pointing into the primary key or datum),
m4_ref(DB_DBT_APPMALLOC) should be set in the m4_arg(flags) field of the
m4_arg(result) m4_ref(Dbt), which indicates that m4_db should free the
memory when it is done with it.])

m4_tag(m4_idef(DB_DBT_MULTIPLE), [dnl
To return multiple secondary keys, m4_ref(DB_DBT_MULTIPLE) should be set
in the m4_arg(flags) field of the m4_arg(result) m4_ref(Dbt), which
indicates m4_db should treat the m4_arg(size) field as the number of
secondary keys (zero or more), and the m4_arg(data) field as a pointer
to an array of that number of m4_ref(Dbt)s describing the set of
secondary keys.
m4_p([dnl
m4_bold([When multiple secondary keys are returned, keys may not be repeated]).
In other words, there must be no repeated record numbers in the array
for Recno and Queue databases, and keys must not compare equally using
the secondary database's comparison function for Btree and Hash
databases.  If keys are repeated, operations may fail and the secondary
may become inconsistent with the primary.])
m4_p([dnl
The m4_ref(DB_DBT_APPMALLOC) flag may be set for any m4_ref(Dbt) in the
array of returned m4_ref(Dbt)'s to indicate that m4_db should free the
memory referenced by that particular m4_ref(Dbt)'s data field when it
is done with it.])
m4_p([dnl
The m4_ref(DB_DBT_APPMALLOC) flag may be combined with
m4_ref(DB_DBT_MULTIPLE) in the m4_arg(result) m4_ref(Dbt)'s m4_arg(flag)
field to indicate that m4_db should free the array once it is done with
all of the returned keys.])])

m4_tag(m4_idef(DB_DONOTINDEX), [dnl
If any key/data pair in the primary yields a null secondary key and
should be left out of the secondary index, the callback function may
optionally return m4_ref(DB_DONOTINDEX).  Otherwise, the callback
function should return 0 in case of success or an error outside of the
m4_db name space in case of failure; the error code will be returned
from the m4_db call that initiated the callback.
m4_p([dnl
If the callback function returns m4_ref(DB_DONOTINDEX) for any key/data
pairs in the primary database, the secondary index will not contain any
reference to those key/data pairs, and such operations as cursor
iterations and range queries will reflect only the corresponding subset
of the database.  If this is not desirable, the application should
ensure that the callback function is well-defined for all possible
values and never returns m4_ref(DB_DONOTINDEX).])
m4_p([dnl
Returning m4_ref(DB_DONOTINDEX) is equivalent to setting
m4_ref(DB_DBT_MULTIPLE) on the m4_arg(result) m4_ref(Dbt) and setting
the m4_arg(size) field to zero.])])
m4_tagend

m4_not_reentrant])

m4_param(flags, [dnl
m4_sf_or_may
m4_tagbegin
m4_tag(m4_idef(DB_CREATE), [dnl
If the secondary database is empty, walk through the primary and create
an index to it in the empty secondary.  This operation is potentially
very expensive.
m4_p([dnl
If the secondary database has been opened in an environment configured
with transactions, each put necessary for its creation will be done in
the context of a transaction created for the purpose.])
m4_p([dnl
Care should be taken not to use a newly-populated secondary database in
another thread of control until the m4_ref(dbh_associate) call has
returned successfully in the first thread.])
m4_p([dnl
If transactions are not being used, care should be taken not to modify
a primary database being used to populate a secondary database, in
another thread of control, until the m4_ref(dbh_associate) call has
returned successfully in the first thread.  If transactions are being
used, m4_db will perform appropriate locking and the application need
not do any special operation ordering.])])

m4_tag(m4_idef(DB_IMMUTABLE_KEY), [dnl
Specifies the secondary key is immutable.
m4_p([dnl
This flag can be used to optimize updates when the secondary key in a
primary record will never be changed after the primary record is
inserted.  For immutable secondary keys, a best effort is made to avoid
calling the secondary callback function when primary records are
updated.  This optimization may reduce the overhead of update operations
significantly if the callback function is expensive.])
m4_p([dnl
Be sure to specify this flag only if the secondary key in the primary
record is never changed.  If this rule is violated, the secondary index
will become corrupted, that is, it will become out of sync with the
primary.])])
m4_tagend])

m4_param(primary, [dnl
ifelse(M4API, C_API, [dnl
The m4_arg(primary) parameter should be a database handle for the primary
database that is to be indexed.],[dnl
The associate method called should be a method of a database handle for
the primary database that is to be indexed.])])

m4_param(secondary, [dnl
The m4_arg(secondary) parameter should be an open database handle of
either a newly created and empty database that is to be used to store
a secondary index, or of a database that was previously associated with
the same primary and contains a secondary index.  Note that it is not
safe to associate as a secondary database a handle that is in use by
another thread of control or has open cursors.  If the handle was opened
with the m4_ref(DB_THREAD) flag it is safe to use it in multiple threads
of control after the m4_refT(dbh_associate) has returned.  Note also
that either secondary keys must be unique or the secondary database must
be configured with support for duplicate data items.])

m4_param_txn(dbh_associate)

m4_paramend

m4_err(dbh_associate, rephandle, replockout, einval,
[the secondary database handle has already been associated with this or
another database handle; the secondary database handle is not open; the
primary database has been configured to allow duplicates])

m4_seealso(Db)
m4_page_footer