Attached is a tar ball that demonstrates the problem.
Unpack it, and execute via:
The script calls a Test program which creates and
populates a database with 500K entries, each unique,
starting with key0000000=val0000000. It then checks to
see if each of the created entries actually exists, and
fails on the first miss.
The program continues on to run a binary search
algorithm, finding the point at which the database
first looses data.
I have verified that this behaviour does not exist in
Perl's SDBM_File, and I have tested this on linux
Windows 2000 Server, and Solaris, each under JDK1.4.
If you have any questions, you can contact me at