A schema is the heart of any hierarchical database such as Novell Directory Services (NDS).  The schema is essentially a list of classes, attributes, and syntaxes used to instantiate objects in a database tree.  NDS is also a distributed database.  Because the data is spread across many servers, each server must have an identical copy of the schema.

If the schema on one server is different than another server, then all kinds of elusive bugs emerge.  In the past, Novell technical support has spent many wasted hours tracking down schema problems.  Comparing schemas is easy, like comparing two dictionaries.  Unfortunately, it's also laborious and boring (again, like comparing two dictionaries).  The comparison is time-consuming and error-prone.

Although the algorithm is the same in C++ and DemeterJava, the C++ solution took 4 man-weeks and DemeterJava 2.5 man-weeks.  In addition, Java gives us cross-platform independence and  DemeterJava gives us re-usability for non-Novell schema comparisons.  The application takes two reference files (schemas listings) and compares them for unique classes and attributes, and modified classes and attributes.  Using a batch file, the customer could (several have already) compare every server in their database.  The 5-hour manual process has been replaced by a 2-minute automation.
 

Phase 2
Compared classes, and printed out the unique classes of each schema.  (Nathan Jensen)

Phase 3
Compared class definitions of the non-unique classes, and printed out definition differences. (Nathan Jensen)

Phase 4
Compared attributes & syntaxes, and printed out the unique attributes and syntaxes of each schema.  (Scott Danielson)

Phase 5
Combined Phase 3 and Phase 4 (Nathan Jensen & Scott Danielson)
 

Phase 6
Read a schema live from the server.  Read a list of servers in a list file (.lst) and recursively compared each one.
 

Two weeks of public beta testing on the web.  One customer complained (correctly so) about some incorrect error messages.  The customer should have gotten the messages, but the messages didn't correctly explain the problem or how to fix it.  Re-released it.  No complaints.
 

• The parsing was HORRENDOUS in C++.  We should have used LEX or YACC.  Half the development time was writing the parsing.
• The list files are FANTASTIC.  You can put list files inside of list files.  You can even compare a list file to another list file.
• Code is solid.
• No improvements are recommended.
• The main code is generally NOT reusable.  We did make an NDS class that we've since reused without modification.

• Improvement:  Read from a live server (JNDI for NDS  is not completely ready for implementation in Java).
• Improvement:  Read a list file (not enough time).
• Code is solid.
• The code is so reusable, that it could be used on a non-Novell schema like Microsoft's Active Directory.  In fact, reusing the code would be trivial if the database is IEEE X.500 compliant.