Copybook sprawl

Copybooks are the de facto data dictionary of a mainframe estate, and after forty years they no longer agree with each other. The same record exists as CUSTREC, CUSTREC2 and CUSTRECN with small differences. COPY REPLACING produces several prefixed copies of one layout inside a single program. Two libraries in the SYSLIB concatenation can both contain a member called CUSTREC, and the compiler uses whichever library comes first, so the layout depends on the compile JCL or the Endevor processor rather than the source. Some copybooks contain procedure division code, not data. A modernization that builds its canonical data model by reading the copybook library alphabetically ends up with the wrong field lengths for some programs and a model with three customer entities that are really one.

PYD ENTERPRISES resolves copybooks the way the compiler did: it reads the library concatenation for each program from the compile JCL, the Endevor processor output or the build configuration, then expands every COPY with its REPLACING operands, including partial-word replacement of tags such as :PFX:. Each expanded layout is fingerprinted by field offsets, lengths and usages. Layouts with matching fingerprints are clustered even when names differ, and members that shadow another member of the same name are reported with the programs affected. Field-level usage analysis shows which fields each program actually reads or writes.

Collapse each fingerprint cluster into one canonical schema with a version, published from a single schema registry or contracts package. Program-specific differences become explicit versions or optional fields, not separate entities. Fields that no program reads are candidates to drop, with the usage evidence attached for the data owner to review.