Bt_compare_fcn_type: Second DBT corrupted?

Question

Folks-
I have a BTree db, opened as part of an env:
cEnvFlags = DB_CREATE|DB_INIT_LOCK|DB_INIT_LOG |DB_INIT_MPOOL|DB_INIT_TXN;
m_pDbEnv = new DbEnv(0);
m_pDbEnv->open("Env_Home", cEnvFlags, 0);
// db
m_pdb = new Db(m_pDbEnv, 0);
m_pdbWK->set_bt_compare(bt_compare_fcn_type_wk);
m_pdbWK->open(NULL, "MY_DB" NULL, DB_BTREE, DB_CREATE|DB_AUTO_COMMIT, 0);
I insert a new record with the key set to Windows GetTickCount() function (to get a unique number since the start of 'this' session)
DWORD dw = GetTickCount();
Dbt key;
key.set_data(&dw);
key.set_ulen(sizeof(DWORD));
Dbt data(pData, dataLen);
try
m_pdbWK->put(NULL, &key, &data, 0);
catch(DbException &e)
in my compare function I do the following:
int bt_compare_fcn_type_wk(DB db, const DBT dbt1, const DBT *dbt2)
DWORD dw1, dw2;
memcpy(&dw1, dbt1->data, sizeof(DWORD));
memcpy(&dw2, dbt2->data, sizeof(DWORD));
if (dw1 == dw2)
return 0;
else
return dw1 > dw2;
However dbt2 does not appear to point to any valid data. The size and ulen members are always 0 and the *(data) memory location does not have valid data.
I was expecting dbt2 to be the key from the 'last' insert. What am I doing wrong here?
I'm using MS VC++ 6.0 on XP Prof.
Any help is greatly appreciated.

Answer 1

The dbts passed to your callback are either the one passed in by the application, or a record that is in the database. So I suspect you have somehow inserted an empty key into the database (which Berkeley DB supports without complaining), and you're now seeing that record in the callback when Berkeley DB is finding the correct location for the newly-inserted key.
Take a look at the data in the database (the easiest way is to run the "db_dump" utility against your database and check for blank lines).
If you don't see an empty key in the db_dump output, can you please post a stack dump for when the callback hits this?
Regards,
Michael Cahill, Oracle.