Release v1.11.0 — Column lineage pipelines

[ikelaiah]

PR: v1.11.0 — Column lineage pipelines

Summary

Implements the v1.11.0 — Column lineage pipelines milestone from
ROADMAP.md in full and nothing from v1.12.0 onwards: column-flow edges
through CTEs, views, temporary tables, transformations, and
catalogue-resolved object boundaries (README roadmap item 6 delivered);
F column metadata and column-flow export styles; E
column-resolution signals; and the layout engine's documented column
graph class. The golden suite grows from 209 to 210; fuzz (400) and UI (22)
stay green; 15 column-flow fixtures assert end-to-end traces and opaque
ambiguity, plus 2 column layout-budget fixtures. RDL / report import (v1.12.0)
remains deferred.

What's included

Column-flow module (src/columnflow.ts)

A new runtime module (loaded after columns.js, before exporters.js) that
walks an object's statements in source order and tracks the column-level
reaching definition of each column-carrying object:

• Reaching definitions across statements. SELECT … INTO (v1.10.0
  fixtures already bound the projection), INSERT … SELECT (including explicit
  column lists and positional mapping onto a CREATE TABLE schema), CREATE TABLE DDL column lists, and UPDATE … SET transformations each define or
  refine an object's columns. Sequential writes replace the reaching
  definition (most-recent-writer wins, mirroring the v1.5.0 flow graph);
  conditional writes inside a branch/loop/handler mark the object multi at
  merge time.
• End-to-end traces. Every produced column carries its ultimate origin
  object and column (flattened through temp chains, e.g. SELECT … INTO #b FROM #a reports the original source, not #a), as {source, sourceColumn, sourceSpan, resolution}.
• Consumers stay conservative. A consumer binds back through an object's
  known columns; an ambiguous reaching definition (multi) or a column the
  object's definition does not produce stays explicitly opaque with a
  region-scoped, span-attached column_flow_opaque diagnostic — no binding is
  ever invented. A T-SQL SELECT col = expr alias target is no longer mistaken
  for a source-column reference.
• Object boundaries. Views defined earlier in the same script (seeded
  through analyseEstate → opts.define) and catalogue-proven external
  objects resolve column identity by definition; CTE scopes act as provable
  origins. analyseColumns trims trailing semicolons and a depth-0 GO batch
  separator at its entry point so raw view bodies analyse cleanly.
• Export graph (buildColumnGraph): a plain Graph on its own documented
  column layout class — a step spine in statement order plus column-carrying
  object nodes routed as data lanes — that toMermaid/toDrawio render with
  the new colstep/colobj canonical styles and full provenance metadata.

Analysis integration (src/ir.ts)

analyse() computes result.columnFlow (the ColumnFlow model) and
result.columnFlowGraph, merges the region-scoped column_flow_opaque
diagnostics into the findings panel, and feeds a column_flow construct into
the existing construct-coverage surface (E signals with no new UI).
analyseEstate() threads same-script object definitions into later objects so
view boundaries resolve by definition. Confidence, coverage, exports, and all
pre-existing fixtures are untouched: clean pipelines produce no new
diagnostics.

Types, styles, layout

• src/types.d.ts — ColumnFlowOutput, ColumnFlowObject, ColumnFlowStep,
  ColumnFlowConsume, ColumnFlowEdge, ColumnFlow, AnalysisResult.columnFlow
  / columnFlowGraph, AnalyseOptions.define, PROCFLOW_COLUMNFLOW_* globals.
• src/exporters.ts — canonical colstep and colobj node styles shared by
  both exporters (the distinct data-flow edge style already derives from the
  data edge kind).
• tests/parity.ts — the column layout class is documented in
  PROCFLOW_LAYOUT_CLASSES (node limit 28, crossing budget 0); enforcement
  lives in the new suite.

Fixtures and tests

• tests/column-flow.ts (+ tests/index.html, tests/metrics.html,
  column-flow.js, and src/columnflow.js wiring) — 15 fixture-corpus
  records plus 2 integration records, and 2 column layout-budget records:
  end-to-end temp pipelines (SELECT … INTO, UPDATE, INSERT … SELECT
  with a CREATE TABLE schema and an explicit column list, CTE-fed temps),
  two-hop temp chains that flatten to the source, ambiguous IF/ELSE, branch
  and loop writes that read opaque, sequential-producer non-multi replacement,
  same-script view boundaries, catalogue-proven boundaries, clean plain/dynamic
  invariants, and per-fixture Mermaid/draw.io export parity with provenance
  round-trip and data-edge routing. Gates the golden page via
  PROCFLOW_COLUMNFLOW_PASS + PROCFLOW_COLUMNFLOW_RESULT.
• tests/tests.ts — gates the golden page on the v1.11.0 column-flow suite.
• tests/metrics.ts → docs/metrics-v1.11.0.json — adds
  columnFlowPassRate, columnLayoutPassRate, and the columnFlow (15) /
  columnLayout (2) corpus counts; the v1.10.0 snapshot is renamed to
  v1.11.0 (old snapshot removed). All pre-existing corpus aggregates are
  byte-identical to v1.10.0.
• examples/dbo.v1110_demo.sql — a view + three procedures demonstrating
  end-to-end temp pipelines, ambiguous branch reaching definitions, and
  same-script view boundary resolution (with a catalogue to paste).

Verification

• npm run typecheck — passes
• npm run build — passes
• npm run test:file — passes (with CHROME_PATH set)
• Golden suite — 210/210 (was 209)
• Fuzz suite — 400 deterministic mutation cases
• UI suite — 22/22 (construct coverage now also shows the column_flow row)
• npm run metrics — snapshot is current; docs/metrics-v1.11.0.json
  committed (v1.10.0 snapshot renamed), 100 % column-flow pass rate
• Local-only check clean; dist/ committed in sync (adds dist/src/columnflow.js
  and dist/tests/column-flow.js)

Not changed / deferred (per roadmap)

• Everything from v1.12.0 onwards — RDL/report import, report → dataset →
  object → column views, and report export — is out of scope. Interactive
  column views remain scheduled for v1.13.0.
• Existing v1.1.0–v1.10.0 semantics are untouched; the corpus aggregates in
  docs/metrics-v1.11.0.json are byte-identical to v1.10.0 for every
  pre-existing metric, confirming no parser regression, and the full golden
  suite stays green.