Check if the embedding fills up when we are writing the same object twice when generating

[skinkie]

This would happen when in for example reprojection or infer directions. It would (again) generate the same objects. Because of dupSort it won't overwrite. Think about how to remove the objects from the inwards and inverse index.

[skinkie]

The following might be a solution. Also notice how iternext_dup is used, I think this may be a better way than the prefix option I am currently using.

import lmdb

env = lmdb.open("mydb", subdir=False, map_size=2**30, max_dbs=2)
db1 = env.open_db(b"by_parent", dupsort=True)
db2 = env.open_db(b"by_id", dupsort=True)

def delete_inverse_entries(txn, parent_id, parent_version, parent_class):
    """Remove entries from db2 based on parent_id, parent_version, parent_class."""
    key = f"{parent_id}:{parent_version}:{parent_class}".encode()
    
    with txn.cursor(db1) as cursor1, txn.cursor(db2) as cursor2:
        if cursor1.set_key(key):
            # Iterate over all values that match the key in db1
            for value in cursor1.iternext_dup():
                id_, version, cls = value.split(b":")
                inv_key = f"{id_.decode()}:{version.decode()}:{cls.decode()}".encode()

                # Use MDB_GET_BOTH_RANGE to find the exact key-value pair in db2
                if cursor2.set_range(inv_key) and cursor2.value() == key:
                    cursor2.delete()

def insert_entry(txn, parent_id, parent_version, parent_class, id_, version, cls):
    """Insert a new entry, replacing the old one."""
    delete_inverse_entries(txn, parent_id, parent_version, parent_class)

    key1 = f"{parent_id}:{parent_version}:{parent_class}".encode()
    value1 = f"{id_}:{version}:{cls}".encode()
    
    key2 = f"{id_}:{version}:{cls}".encode()
    value2 = f"{parent_id}:{parent_version}:{parent_class}".encode()
    
    txn.put(key1, value1, db=db1, dupdata=True)
    txn.put(key2, value2, db=db2, dupdata=True)

# Example usage
with env.begin(write=True) as txn:
    insert_entry(txn, "P1", "V1", "C1", "I1", "V1", "C2")
    insert_entry(txn, "P1", "V1", "C1", "I2", "V2", "C3")  # Overwrites existing