Skip to content

CHANGELOG.md

Latest commit

 

History

History
180 lines (158 loc) · 9.98 KB

CHANGELOG.md

File metadata and controls

180 lines (158 loc) · 9.98 KB
Raw

Changelog

All notable released ProcFlow changes are recorded here. Detailed verification evidence belongs in the canonical release note under docs/releases/.

v2.7.0 — Maintainability pass

  • Extracted src/token-utils.ts (shared token/AST helpers) and src/dialects-state.ts (PL/pgSQL and T-SQL transaction rules) from src/ir.ts; the analysis modules now form a one-way dependency chain (token-utils ← lineage ← columns ← columnflow), with no cycles.
  • Moved bundled demo SQL out of src/app.ts into src/samples.ts.
  • Unified duplicated logic: one CTE splitter (splitCTEs, now carrying column lists and body spans) and one shared dependency/report filter kernel (graphNarrowByFocus, graphFilteredClone).
  • toMermaid no longer mutates the input graph; added regression tests for the non-mutation guarantee and for Mermaid's securityLevel:'strict'.
  • Deleted dead helpers (workspaceExportText, defaultWorkspaceFilter, a no-op branch in column flow).
  • Added npm test (scripts/test-all.mjs) — runs golden, fuzz, ui, erd-ui, and security suites against a local static server with one command — and fixed a load-order race in the UI suites where the off-screen iframe could finish loading before the test script attached its listener.
  • Adopted Biome (npm run lint) and wired it into CI alongside typecheck.
  • Fixed npm run metrics on Windows: Chromium emits CRLF in --dump-dom, so the snapshot comparison now normalises line endings.
  • UI: tablists support arrow-key navigation (ARIA tabs pattern) and the app follows prefers-color-scheme: light with a legible light palette.
  • Docs: corrected the stale trust-model version and port, indexed ADR-003 and all canonical release notes, and moved 33 superseded RELEASE_NOTE_*, PR_NOTE_*, and historical planning documents to docs/archive/ with an archive README. ROADMAP.md, tasks/, and BENCHMARKS.md are marked as historical where they are no longer current.

v2.6.0 — Aggregates and grouping

  • Added per-column aggregate selectors: COUNT, COUNT DISTINCT, SUM, AVG, MIN, and MAX. The remaining picked columns become the GROUP BY list in pick order; GROUP BY is omitted when every pick is aggregated.
  • Aggregate expressions get deterministic aliases, sorting an aggregated column orders by the expression, and DISTINCT is disabled and explained while grouping because GROUP BY already collapses rows.
  • Query files move to version 2; version-1 files migrate forward with empty aggregates, and aggregates are saved, exported, restored, and pruned with their picks. Newer file versions are rejected with a diagnostic.
  • Added a plan tip for one-to-many multiplication with SUM/AVG.
  • Documentation: ADR-003 and an aggregates section in docs/QUERY_BUILDER.md.

v2.5.0 — Saved queries and file export

  • Added query persistence: Save to this browser, Restore saved, and Forget saved store one query locally with explicit actions only.
  • Added versioned query files: Export query file and Import query file use the procflow-erd-query format with a schema fingerprint. Foreign, future-version, and malformed files are rejected with diagnostics.
  • Restores prune references that no longer exist (tables, columns, taught joins, sorts, cross/excluded ids, path choices) and report the drops and a changed schema instead of failing.
  • Added Download .sql, which respects the Comments toggle; the optional name field drives export filenames and falls back to the schema fingerprint.
  • Added src/query-store.ts (pure serialization, validation, and pruning); schemaFingerprint is now shared by ERD layout and query files.
  • Documentation: ADR-002, a saving/exporting/restoring section in docs/QUERY_BUILDER.md, and ARCHITECTURE invariant 7.
  • Test coverage: query-store golden records (round-trip, rejection, tolerance, pruning, fingerprint, storage) and the ERD UI suite extended to 62 checks (export/import/download, save/restore/forget, activation hint).

v2.4.1 — Query builder hardening and release discipline

  • Hardened the query engine: shortest-path enumeration is iterative, so long declared chains cannot exhaust the call stack; a 1,000-table scale fixture asserts 999 joins, 998 bridges, and deterministic output.
  • Fixed plan-edge highlighting when the overlay redraw was starved by animation-frame scheduling; query-state changes now draw synchronously.
  • Excluding a problem table now prunes its sorts; the dropdown teach flow explains the same-table self-join case; the clipboard fallback is shared.
  • Moved pure DOM builders to src/ui/erd-query-view.ts (no behavior change).
  • Accessibility: one concise status announcement replaces noisy live regions, the floating query window is a labelled dialog, Q toggles query mode, and Escape cancels teaching without clearing the diagram selection.
  • Added tests/erd-ui.html (49 interaction checks) and registered it in CI; the Firefox smoke now loads erd.html.
  • Release discipline: package-lock.json tracks package.json, and CI guards the changelog, README, release index, and lockfile version.
  • Documentation: docs/QUERY_BUILDER.md, ADR-001, and the suite cookbook and header convention in docs/DEVELOPMENT.md.
  • Updated CI actions (actions/checkout, actions/setup-node).

v2.4.0 — Query builder on the ERD

  • Added the ERD query builder: a toolbar toggle turns column rows into checkboxes and opens a draggable, resizable floating window with the generated SQL, the join plan, and explicit resolutions for tables the declared foreign keys cannot reach.
  • Joins follow declared PRIMARY KEY / UNIQUE / FOREIGN KEY evidence only. Equal-cost paths are offered as a choice, bridges are tagged, and preserve or strict join policy controls INNER/LEFT with per-join overrides.
  • Added teach-by-clicking joins, typed predicates, opt-in CROSS JOIN, column exclusion (named in the SQL header), and self joins with a second aliased copy and per-column copy routing.
  • Added query ergonomics: DISTINCT, dialect-aware row caps (TOP / LIMIT / FETCH FIRST), per-column sort cycling, and a Distinct tip when a one-to-many join can repeat rows.
  • Generated SQL is dialect-quoted (T-SQL, PostgreSQL, DB2, SQLite) and syntax highlighted with byte-identical Copy output.
  • The diagram reacts to picks: plan edges stay lit while unrelated edges dim, unresolved tables carry a coral ring, join cards highlight their edge and endpoints on hover, Find matches columns, and Only used tables narrows the canvas.
  • Release quality: package-lock.json version now tracks package.json, and the ERD query UI ships with a committed browser suite and release-doc CI guards.

v2.3.0 — Readable estate layouts

  • Auto-arranged a schema on first parse; Reset still returns to declaration order.
  • Edges now anchor at the referenced foreign-key column row and route orthogonally through card-free layout gutters, with band corridors and an outer channel for long multi-band edges. Zero path/card intrusions verified on the DB2 sample, a 300-table chain, and a sampled 800-table chain.
  • Crossing reduction now uses median sweeps plus a monotonic adjacent-transpose pass, counts crossings by inversion counting, and supports LR/TB/Auto direction; Auto keeps whichever direction fits a 16:9 viewport better.
  • Added position pinning (per-card diamond, Unpin all) so Auto arrange keeps chosen positions, with deterministic de-overlap for the rest.
  • Added spacing presets (Compact/Normal/Roomy) and moved isolated tables and sourceless views to a downstream band.

v2.2.0 — Arrange and save the ERD

  • Added deterministic Auto arrange: layered parent→child flow with bounded barycenter crossing reduction, cycle-safe ordering, and view placement downstream of the tables referenced in the view body (FROM/JOIN, used as a declared layout hint only).
  • Added band wrapping so wide and tall estates (800-table chains, 799-leaf fans) become a readable grid instead of one extremely long line or column.
  • Added layout saving: opt-in Save/Restore/Forget in this browser plus versioned Export/Import layout files keyed by a schema fingerprint, with unmatched entities appended rather than overlapped.
  • Reworked the ERD canvas to absolute positions, so dragging is persistent across re-renders, re-parsing, and Compact mode; Reset returns to declaration order.

v2.1.0 — Entity relationship diagrams

  • Added the ERD page (erd.html): DDL parsing for declared tables, views, columns, primary/unique/foreign keys, and composite-key grouping across T-SQL, PostgreSQL, DB2, and SQLite, with raw type fidelity and source spans.
  • Added deterministic Mermaid erDiagram export with declared cardinality, optionality, and provenance comments. Declared constraints only; unresolved references stay explicit.
  • Added estate navigation: canvas panning, draggable tables, FK selection inspector, find/jump, compact boxes for large estates, and smooth wheel zoom with whole-estate Fit.
  • Added a draggable editor/diagram divider on both pages and a clear Flowchart/ERD page switcher.
  • Fixed view cards being hidden by a global class collision and Fit zooming out indefinitely on large estates.

v2.0.0 — Trustworthy SQL Analysis Contract

  • Added the cross-dialect adversarial semantic qualification matrix and its fixture-only dialectAdversarialSemanticAssertionRate metric.
  • Reject newer and corrupt saved workspace schemas without coercion, mutation, or automatic deletion.
  • Finalised the v2 accuracy and compatibility contract.

v1.14.1 — Targeted correctness patch

  • Preserved named PL/pgSQL LATERAL sources.
  • Kept DB2 PREPARE/EXECUTE dynamic SQL opaque instead of asserting an object read or procedure call.
  • Preserved unique sequential DB2 SESSION. temporary-table data flow.

See the v1.14.1 release note and the accuracy audit for the reproductions and verification record.