Deprecation Migration

Deprecation and Migration

Code is a liability, not an asset. Every line carries maintenance cost — bugs, dependency updates, security patches, onboarding overhead. Deprecation is the discipline of removing code that no longer earns its keep; migration is the process of moving users safely from old to new. Most organizations build things well; few remove them well.

Hyrum's Law makes removal hard: with enough users, every observable behavior becomes depended on — including bugs, timing quirks, and undocumented side effects. Deprecation therefore requires active migration, not just announcement. Plan removal at design time — systems with clean interfaces, feature flags, and minimal surface area are far easier to sunset.

When to use

• Replacing an old system, API, or library with a new one.
• Sunsetting a feature that's no longer needed, or consolidating duplicate implementations.
• Removing dead code that nobody owns but everybody depends on.
• Planning the lifecycle of a new system (deprecation planning starts at design time).
• Deciding whether to maintain a legacy system or invest in migration.
• Upgrading a dependency or framework version, including major/breaking upgrades (see references/dependency-upgrade.md).
• Moving data between systems (dual-write, backfill, CDC, cutover) — see references/data-migration.md.

Not for: patching a security vulnerability in place (use security-review); routine commits (use git-workflow).

Steps

1. Make the deprecation decision

Before deprecating anything, answer:

1. Does this system still provide unique value? If yes, maintain it.
2. How many consumers depend on it? Quantify the migration scope.
3. Does a replacement exist? If no, build the replacement first — never deprecate without an alternative.
4. What's the migration cost for each consumer? Trivially automated → do it; manual and high-effort → weigh against maintenance cost.
5. What's the ongoing cost of not deprecating? Security risk, engineer time, opportunity cost of complexity.

Zombie-code diagnostic — if the target shows these signs, it's zombie code: nobody owns it, but everybody depends on it.

• No commits in 6+ months; no assigned maintainer or team
• Failing tests that nobody fixes
• Dependencies with known vulnerabilities that nobody updates
• Documentation referencing systems that no longer exist

Response: assign an owner and maintain it properly, or deprecate it with a concrete migration plan. Zombie code cannot stay in limbo — it either gets investment or removal.

2. Choose advisory vs compulsory

| Type | When | Mechanism | |------|------|-----------| | Advisory | Migration optional; old system stable | Warnings, docs, nudges. Users migrate on their own timeline. | | Compulsory | Security issues, blocks progress, or maintenance unsustainable | Hard deadline; removal by date X. Migration tooling must be provided. |

Default to advisory. Use compulsory only when the cost or risk justifies forcing migration — and never just announce a deadline; provide tooling, documentation, and support (the Churn Rule: if you own the infrastructure being deprecated, you migrate your users, or provide backward-compatible updates that require no migration).

3. Migrate incrementally

1. Build the replacement — covers all critical use cases, documented, proven in production (not "theoretically better").
2. Announce and document — deprecation notice with status, replacement, removal date, reason, and a migration guide with concrete steps and examples.
3. Migrate consumers one at a time — identify touchpoints, update to the replacement, verify behavior matches (tests, integration checks), remove old references, confirm no regressions.
4. Remove the old system — only after all consumers migrated. Verify zero active usage (metrics, logs, dependency analysis), then remove code, tests, docs, config, and the deprecation notices. Removing code is an achievement.

4. Pick a migration pattern

Strangler — run old and new in parallel; route traffic incrementally (0% → 10% canary → 50% → 100%); remove the old system when it handles 0%.

Adapter — translate calls from the old interface to the new implementation. Consumers keep using the old interface while you migrate the backend.

class LegacyTaskService implements OldTaskAPI {
  constructor(private newService: NewTaskService) {}
  getTask(id: number): OldTask {
    return this.toOldFormat(this.newService.findById(String(id)));
  }
}

Feature flag — switch consumers one at a time:

function getTaskService(userId: string): TaskService {
  if (featureFlags.isEnabled('new-task-service', { userId })) {
    return new NewTaskService();
  }
  return new LegacyTaskService();
}

Concrete example — migrating test helpers to @total-typescript/shoehorn. When the migration target is a library whose API replaces a problematic pattern (e.g. as type assertions in tests), the same incremental process applies:

1. Install the replacement: npm i @total-typescript/shoehorn.
2. Find call sites: grep -rE ' as [A-Z]' --include='*.test.ts' --include='*.spec.ts'.
3. Replace as Type → fromPartial(...) (partial data that still type-checks); replace as unknown as Type → fromAny(...) (intentionally wrong data for error tests). Use fromExact() to force a full object when you plan to swap to fromPartial later.
4. Add imports from @total-typescript/shoehorn, run typecheck, verify.

Test code only — never use shoehorn in production code.

Database schema (expand/contract) — the riskiest migration because data is the one thing you can't roll back by reverting a deploy. Never change a column in place. Migrate in additive phases so old and new code are both valid at every step:

EXPAND ──────→ MIGRATE ──────→ CONTRACT
add new       backfill rows,   once no code reads
column        dual-write       the old column,
(nullable)    old+new          drop it in a later,
              from app         separate deploy

Worked example — renaming name to full_name:

1. Expand. Add full_name nullable. Deploy. (Old code ignores it.)
2. Dual-write. App writes both name and full_name on every insert/update. Deploy.
3. Backfill. Copy name → full_name for existing rows, in throttled batches (don't lock the table).
4. Switch reads. Point the app at full_name, keep writing both. Deploy and bake.
5. Contract. Stop writing name; in a separate, later deploy, drop the column.

Rules: additive first, destructive last and alone; every migration has a tested down path; backfill in batches off the hot path; build large indexes without blocking writes (e.g. Postgres CREATE INDEX CONCURRENTLY); decouple cutover from code by feature flag when risky.

> Warning: If CONCURRENTLY fails mid-build, it leaves an INVALID index that silently does nothing — check \di or pg_indexes for validity; drop and recreate before relying on it.

Verify

After completing a deprecation:

• [ ] Replacement is production-proven and covers all critical use cases
• [ ] Migration guide exists with concrete steps and examples
• [ ] All active consumers migrated (verified by metrics/logs)
• [ ] Old code, tests, documentation, and config fully removed
• [ ] No references to the deprecated system remain in the codebase
• [ ] Deprecation notices removed (they served their purpose)

After a database schema migration:

• [ ] Change ships in additive phases (expand → backfill → contract), not a single in-place edit
• [ ] Old and new code are both valid against the schema at every deploy step
• [ ] Each migration has a tested down path; backfills run in throttled batches
• [ ] Destructive steps (drop/rename) ship in their own deploy after no code references the old shape

References

• [${CLAUDEPLUGINROOT}/references/engineering-principles.md](${CLAUDEPLUGINROOT}/references/engineering-principles.md) — shared discipline (verify don't assume, surgical scope, simplicity)
• [references/dependency-upgrade.md](references/dependency-upgrade.md) — changelog/migration-guide reading, breaking-change triage, branch-based upgrade, root-cause fix per break
• [references/data-migration.md](references/data-migration.md) — moving data between systems: dual-write + backfill, CDC, cutover, verification, failure modes

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: int2t05
• Source: int2t05/engineering-skills
• License: MIT

Install and usage instructions live in the source repository linked above.