Dosu Logo
Ask
DocumentsMag Stacker
0007-magpul-constraint-domain-layer-not-db
0007-magpul-constraint-domain-layer-not-db
Type
External
Status
Published
Created
Jul 4, 2026
Updated
Jul 4, 2026
Source
View

ADR-0007: Enforce the Magpul label constraint in the domain/UX layer, not the database#

Date: 2026-07-01
Status: accepted
Deciders: unclesp1d3r (with Claude Code)

Context#

Issue #21 asked to constrain magazine.label to the PMAG Gen M3 dot-matrix set (A-Z, 0-9, -, max 4) and originally proposed a global rule with a database check backstop. During brainstorm/planning this was reworked into an opt-in, per-account "Magpul mode": the constraint applies only when the magazine owner's mode is on, must grandfather existing nonconforming labels, is owner-governed under grant-based sharing, and must hold on every write path (single add/edit, bulk add, server actions, paste).

Decision#

Enforce the constraint in the domain/service layer (authoritative) plus a client form input mask (affordance only). No database CHECK constraint or trigger. The owner's mode is resolved server-side inside the write transaction and passed to a pure validator; labels are normalized (uppercase + outer-trim) and stored only when the owner's mode is on; invalid input is rejected, never truncated.

Alternatives Considered#

Alternative 1: Table-level CHECK constraint (the original #21 proposal)#

  • Pros: DB-guaranteed invariant; cannot be bypassed by any write path.
  • Cons: cannot read the owner's per-account flag or express "only when mode on"; cannot grandfather existing rows.
  • Why not: the rule is conditional on another table's column — structurally inexpressible as a CHECK.

Alternative 2: BEFORE INSERT/UPDATE trigger#

  • Pros: DB-enforced; could join the user row to read the flag.
  • Cons: cross-table coupling and operational complexity in the storage layer for what is a UX/validation rule; duplicates domain logic in SQL.
  • Why not: the coupling and complexity aren't justified for a validation feature that isn't a storage invariant.

Alternative 3: Global constraint on every label#

  • Pros: one code path; simplest rule.
  • Cons: punishes owners who mark magazines with arbitrary stencilled/painted identifiers and never opted into the dot-matrix workflow.
  • Why not: the constraint is only meaningful for owners who opt in.

Alternative 4: Per-magazine "marking type" flag (vs per-account)#

  • Pros: mixed marking schemes within one account.
  • Cons: more state, no demonstrated need.
  • Why not: YAGNI; a per-account opt-in signals intent for hard enforcement. Revisit if mixed per-magazine marking becomes a real need.

Consequences#

Positive#

  • Supports conditional / owner-governed / grandfathering behavior that a DB constraint cannot express.
  • The pure validator is trivially unit-tested; owner resolution + normalization live in the service and are integration-tested.
  • Shared constants (MAX_LABEL_LENGTH, allowed-set regex, normalizeMagpulLabel) back single add, bulk add, and the form, and are reused by #20 (rendering) and #22 (auto-numbering).

Negative#

  • No DB backstop: a future write path that bypasses the domain service could persist a nonconforming label. Mitigation: all writes route through the service; bulk add was explicitly wired in.
  • The client mask keys on the owner's flag only for self-owned magazines; shared-magazine edits by a grantee defer to authoritative domain validation (no client mask).

Risks#

  • The per-write owner-mode lookup adds a read — bounded (indexed primary-key lookup) and inside the existing transaction. The update path locks the row (SELECT … FOR UPDATE) so a stale read can't skip normalization/validation under concurrent shared edits.