The system has been designed to use algorithms so an
implementation can introduce their own logic in a way that's
100% upgradeable (without the need to retrofit logic). The
following points describe strong recommendations about how to
construct new algorithm type programs so that you won't have to
make program changes during future upgrades:
- Do not alter an algorithm type's hard parameters. For
example, you might be tempted to redefine or initialize
parameters defined in an algorithm type's linkage section. Do
not do this.
- Follow the naming conventions for the new algorithm type
code and your source code, i.e., both the source code and the
algorithm type should be prefixed with "CM". The reason for
this naming convention is to make it impossible for a new,
base-package algorithm type from overwriting your source code
or algorithm type meta-data (we will never develop a program
or introduce meta-data beginning with CM).
- Avoid using inline SQL to perform insert/update/delete.
Rather, invoke the base-package's object routines or common
routines.
- Avoid using base messages (outside of common messages,
i.e., those with a message number < 1000) as we may
deprecate or change these messages in future releases. The
most common problem is caused when an implementation clones a
base package algorithm type program because they need to
change a few lines of logic. Technically, to be 100%
upgradeable, you should add new messages in the "90000" or
greater category (i.e., the category reserved for
implementation-specific messages) for every message in your
new program even though these messages may be duplicates of
those in the base package.
Copyright © 2011, Oracle and/or its affiliates. All rights reserved.