Roadmap

v0.1 — Compiler credibility

Theme: deterministic compiler skeleton.

Deliverables:

• BankTS parser subset
• typed AST
• decimal type metadata
• minimal IR
• readable COBOL emitter
• source map
• golden tests
• bankc check
• bankc emit cobol
• account-transfer pure-function demo
• audit report skeleton

v0.2 — Financial arithmetic

Theme: safe decimal and money behaviour.

Deliverables:

• decimal operations
• rounding policies
• overflow diagnostics
• currency types
• decimal property tests
• COBOL decimal mapping report

v0.3 — Copybook foundation

Theme: data-layout credibility.

Deliverables:

• BankTS record to copybook generation
• subset copybook parser
• layout inspector
• copybook diff
• fixture generator
• copybook roundtrip example

v0.4 — Banking diagnostics

Theme: domain safety.

Deliverables:

• diagnostic catalogue
• audit-event checker
• idempotency checker
• double-entry checker
• file-status checker
• source-map coverage checker

v0.5 — Batch and file programs

Theme: batch workload credibility.

Deliverables:

• batch job declaration
• sequential file IO
• generated file sections
• generated JCL example
• batch interest accrual example

v0.6 — Db2 profile

Theme: embedded SQL credibility.

Deliverables:

• SQL declarations
• host variable mapping
• SQLCA
• SQLCODE diagnostics
• CRUD lowering
• cursor support

v0.7 — CICS profile

Theme: online transaction credibility.

Deliverables:

• transaction declaration
• COMMAREA support
• CICS command generation
• response-code diagnostics
• syncpoint/rollback model

v0.8 — Migration analysis

Theme: legacy-estate relevance.

Deliverables:

• COBOL inventory tools — done, bankc analyse
• SQL extractor — done
• CICS extractor — done
• paragraph graph — done, as Mermaid
• copybook dependency graph — copybooks are named, not followed
• skeleton migration output — not started, and the least certain of the six: a skeleton that is wrong in a way a reader trusts is worse than none

See migration-analysis.md.

v0.9 — The depth each subsystem is missing

Theme: the second program of each kind, rather than the first.

Db2, in the order a batch meets them:

• WITH HOLD cursors — done, cursor ... hold, with BANK-SQL-008
• multi-row FETCH — done, cursor ... rowset n
• isolation level on a statement — never missing: the SQL is passed through
• savepoints — the same, with BANK-SQL-009 keeping commit out of raw SQL
• LOCK TABLE — the same
• GET DIAGNOSTICS — never missing either: an ordinary statement in a sql declaration, with host variables resolved like any other
• scrollable cursors — done 2026-08-07, and it was the one that needed syntax rather than pass-through. cursor ... scroll emits INSENSITIVE SCROLL CURSOR, and for each ... from n backward walks the result set from a chosen row with FETCH ABSOLUTE :position

zUnit test generation — done, test <name> for <entry transaction> and bankc zunit. It was blocked on not having IBM's schema, and what unblocked it was test cases IBM's own generator produced, published in public repositories: every shape in the three artifacts is copied from one of those and cited in zunit.md. Two values are inferred rather than observed and say so, in D20 and D21. No generated case has been run — that is what the z/OS kit is for.

v1.0

What has to be true before the interfaces stop moving:

• a stable CLI
• a documented language subset
• stable audit schemas
• a deterministic output guarantee
• a security policy
• an SBOM
• a signed release target
• demos that run something end to end
• a compatibility matrix
• a contributor process

Researched, not built

Four integrations were researched during planning and never started. They are here rather than as pages of their own, because a page describing something that does not exist reads as though it does. What is worth keeping is the reading.

IBM Dependency Based Build. DBB builds COBOL, PL/I and Assembler as part of a DevOps pipeline, and a generated program has to reach a real z/OS build rather than a script invented here. What BankLang would have to emit is the dependency graph it already knows: source, copybooks, Db2 precompile and CICS translation metadata, and the compiler options bankc already writes onto the CBL statement. Overview · With IDz

z/OS Connect. Contract-first OpenAPI 3.0 over CICS and IMS. The interesting half for a compiler is copybook-to-OpenAPI and back, which is the same layout problem bankc copybook import already solves in one direction. Designer · Calling APIs

Galasa, for deep integration tests against real CICS and IMS, once anything runs on z/OS at all. COBOL Check was the other candidate and is marked Emeritus by the Open Mainframe Project, so it would need evaluating before anything depended on it. Galasa · COBOL Check

IBM Z Open Editor. The editor work that did happen — the LSP and the VS Code extension — went its own way. What is still unbuilt is the part that would make a generated program navigable from the editor a z/OS developer already uses: copybook preview, generated-COBOL preview, and source-to-COBOL navigation. Z Open Editor · LSP

Read this page as Markdown on GitHub →