enum IndexKind { LHash, BTree, UniqueBTree, RTree, RDTree };
struct IndexId {
lrid_t obj;
int i;
};
typedef IndexId IndexId;
VASResult shore_vas::addIndex(
const IndexId &iid,
IndexKind indexKind
);
VASResult shore_vas::dropIndex(
const IndexId &iid,
);
VASResult shore_vas::statIndex(
const IndexId &iid,
);
VASResult shore_vas::insertIndexElem(
const IndexId &iid,
const vec_t &key,
const vec_t &value
);
VASResult shore_vas::removeIndexElem(
const IndexId &iid,
const vec_t &key,
const vec_t &value
);
VASResult shore_vas::removeIndexElem(
const IndexId &iid,
const vec_t &key,
int *numremoved // # elems removed
);
VASResult shore_vas::findIndexElem(
const IndexId &iid,
const vec_t &key,
const vec_t &value,
ObjectSize *vallen,
bool *found
);
The language binding may call insertIndexElem and removeIndexElem to update indexes. When keys are known, the language binding layer locates values associated with the keys by calling findIndexElem. Indexes are scanned with a set of methods that are described in indexscan(svas).
The Shore Value-Added Server maintains an array of the logical identifiers of the indexes that are "contained" in an object. The array has origin 0, and by repeatedly calling indexId, the identifiers of the indexes in an object can be discovered.
The disk space used for an index and its entries is not reflected in the system properties of the containing object, but it can be computed by calling statIndex on each index contained in an object.
When an index is created, its type must be given in indexKind. There is no default index type. TODO add references for index types (btree, rtree, etc).
An index is part of an object, and when it is manipulated, the system properties of the "owning" object are checked for access permission. The object is thereby locked.
To insert a key-value pair into an index, both the the key and value are treated as byte-strings. They need not be located in contiguous memory the caller's address space, since vectors are used to describe both the key and value. InsertIndexElem does not change the vectors or the values identified by the vectors.
To remove one key-value pair, use the first form of removeIndexElem, in which the caller provides both the key and value. To remove all the values associated with a key in a non-unique index, use the second form of removeIndexElem, in which only the key is given. The SVAS returns the number of key-value pairs removed in *numremoved. Numremoved may not be a null pointer.
To find a value associated with a known key, use the method findIndexElem. The caller passes in the key and the value is returned. The value is placed in caller's address space in the place or places identified by the argument value. The caller is responsible for allocating this space, and it must be large enough to accommodate any of the values associated with the given key. The SVAS writes the length of the value returned into vallen. The vector value is passed as a const reference because the vector is not modified, even though the memory areas that are identified by this vector are updated. If eof is not null, findIndexElement writes TRUE(1) into *eof if there is a value associated with the given key. If there is no value for the key, it writes FALSE(0) in *eof.
Deadlocks can occur while locks are being acquired. See transaction(svas) for information about deadlocks.
A complete list of errors is in errors(svas).