#!/usr/bin/env node
// One-shot ADR importer for the ruflo-adr plugin.
//
// Walks the working directory (or ADR_ROOT override), parses every ADR file
// under */docs/adr/ or */docs/adrs/, persists records to the `adr-patterns`
// namespace, persists causal edges to the `adr-edges` namespace, prints a
// summary with status counts + relationship breakdown + dangling-ref check.
//
// Handles two ADR formats:
//   1. v3-style:  `# ADR-097: Title` heading + `**Status**: Proposed` line
//   2. plugin-style: YAML frontmatter (`status: Proposed`) at file head
//
// Usage:
//   node scripts/import.mjs                         # markdown summary to stdout
//   IMPORT_FORMAT=json node scripts/import.mjs       # JSON summary
//   IMPORT_DRY_RUN=1 node scripts/import.mjs         # parse + summarize, skip memory_store
//   ADR_ROOT=/path/to/repo node scripts/import.mjs   # override scan root (default: cwd)
//
// Why a script, not raw MCP calls: 70+ ADRs × multiple memory_store calls each
// is hundreds of MCP round-trips. spawnSync over the CLI is materially faster
// and avoids shell-quoting pitfalls in the ADR titles.

import { readFileSync, readdirSync, statSync } from 'node:fs';
import { join, basename } from 'node:path';
import { spawnSync } from 'node:child_process';

// ADR-100 / #1748 Issue 3 — CLI_CORE=1 routes to lite cli-core (~2s cold-cache).
// Note: cli-core's JsonMemoryBackend overwrites by default, so the
// "exists" / UNIQUE-constraint detection below collapses to "ok" under CLI_CORE.
// Re-running import in CLI_CORE mode is therefore idempotent (records refreshed)
// rather than incremental (records skipped). For incremental imports across
// many runs, leave CLI_CORE unset.
const CLI_PKG = process.env.CLI_CORE === '1'
  ? '@claude-flow/cli-core@alpha'
  : '@claude-flow/cli@latest';

const ROOT = process.env.ADR_ROOT || process.cwd();
// #2474 bonus: \`.claude/worktrees/*\` mirrors the repo so every ADR was
// indexed 2-3x. Skip the whole \`.claude\` tree — it's all ruflo runtime
// state (worktrees, scheduled tasks, etc.), never authored content.
const SKIP_DIRS = new Set(['node_modules', '.git', 'dist', 'v2', '.next', '.turbo', 'build', '.claude']);

// #2474 Bug 4: parseId padded ADR numbers, but extractAdrRefs's
// padStart-then-strip pipeline produced different results for the same
// numeric value. A repo mixing `0001-foo.md` (parseId → ADR-0001) with
// a body line `Supersedes: ADR-0001` (extractAdrRefs → ADR-001) drops
// every edge as dangling. Single normalizer keeps both paths in lockstep.
function normalizeAdrId(raw) {
  const digits = String(raw).replace(/^ADR-?/i, '').trim();
  if (!/^\d+$/.test(digits)) return `ADR-${raw}`;
  // Canonical form: ≤3 digits → zero-pad to 3 (legacy default);
  // ≥4 digits → keep as-is. Either way, the same numeric input from
  // a filename and from a body reference now produce the same key.
  return digits.length >= 4 ? `ADR-${digits}` : `ADR-${digits.padStart(3, '0')}`;
}

function findAdrs(dir, out = []) {
  let entries;
  try { entries = readdirSync(dir); } catch { return out; }
  for (const e of entries) {
    if (SKIP_DIRS.has(e)) continue;
    const p = join(dir, e);
    let st;
    try { st = statSync(p); } catch { continue; }
    if (st.isDirectory()) {
      findAdrs(p, out);
    } else if (e.endsWith('.md') && (p.includes('/docs/adr/') || p.includes('/docs/adrs/'))) {
      out.push(p);
    }
  }
  return out;
}

function parseAdr(path) {
  const text = readFileSync(path, 'utf-8');
  const id = parseId(path, text);
  const title = parseTitle(text);
  const status = parseStatus(text);
  const date = parseDate(text);
  const tags = parseTags(text);
  const context = parseContextFirstParagraph(text);
  const links = parseLinks(text, id);
  return { id, title, status, date, tags, context, links, file: path.replace(ROOT + '/', '') };
}

function parseId(path, text) {
  const fm = /^---\s*$([\s\S]*?)^---\s*$/m.exec(text);
  if (fm) {
    const m = /^id:\s*(\S+)/m.exec(fm[1]);
    if (m) return normalizeAdrId(m[1]);
  }
  const fname = basename(path, '.md');
  const m = /^(ADR-?\d+|\d{3,4})/i.exec(fname);
  if (m) return normalizeAdrId(m[1]);
  return fname;
}

function parseTitle(text) {
  const fm = /^---\s*$([\s\S]*?)^---\s*$/m.exec(text);
  if (fm) {
    const m = /^title:\s*(.+)$/m.exec(fm[1]);
    if (m) return m[1].trim();
  }
  const m = /^#\s*(?:ADR-?\d+:?\s*)?(.+?)$/m.exec(text);
  return m ? m[1].trim() : '(untitled)';
}

function parseStatus(text) {
  const fm = /^---\s*$([\s\S]*?)^---\s*$/m.exec(text);
  if (fm) {
    const m = /^status:\s*(.+)$/im.exec(fm[1]);
    if (m) return m[1].trim();
  }
  // #2474 Bug 2: also accept `**Status:**` (colon inside the bold span),
  // the widely-used Nygard / MADR style. Previous regex only matched
  // `**Status**:` (colon outside) and dropped every Nygard-style ADR
  // to status=Unknown. Now the colon can sit on either side of the `**`.
  // Strip parenthetical qualifiers like "Proposed (v3.6.x)" -> "Proposed".
  let m = /^\*\*Status:?\*\*:?\s*([A-Za-z][A-Za-z\- ]*?)(?:\s*\(.*?\))?\s*$/m.exec(text);
  if (m) return m[1].trim();
  // Also handle full-bold MADR style: **Status: Value** (entire phrase bolded)
  m = /^\*\*Status:\s*([A-Za-z][A-Za-z\- ]*?)(?:\s*\([^)]*\))?\*\*\s*$/m.exec(text);
  return m ? m[1].trim() : 'Unknown';
}

function parseDate(text) {
  const fm = /^---\s*$([\s\S]*?)^---\s*$/m.exec(text);
  if (fm) {
    const m = /^date:\s*(\S+)/m.exec(fm[1]);
    if (m) return m[1];
  }
  const m = /^\*\*Date\*\*:\s*(\S+)/m.exec(text);
  return m ? m[1] : '';
}

function parseTags(text) {
  const fm = /^---\s*$([\s\S]*?)^---\s*$/m.exec(text);
  if (fm) {
    const m = /^tags:\s*\[([^\]]+)\]/m.exec(fm[1]);
    if (m) return m[1].split(',').map((s) => s.trim()).filter(Boolean);
  }
  const m = /^\*\*Tags\*\*:\s*(.+)$/m.exec(text);
  return m ? m[1].split(',').map((s) => s.trim()).filter(Boolean) : [];
}

function parseContextFirstParagraph(text) {
  const m = /^##\s*Context\s*$\s*([\s\S]+?)(?=^##\s|\Z)/m.exec(text);
  if (!m) return '';
  return m[1].trim().split(/\n\s*\n/)[0].replace(/\s+/g, ' ').slice(0, 400);
}

// Extract ADR-NNN references from a link line. CRITICAL: must distinguish ADR
// references from GitHub issue numbers (#1697 etc.) which the prior version
// false-positively captured as "ADR-1697". We only recognize bare numbers as
// ADR refs when they appear in a known ADR-link section AND they don't look
// like GitHub issues (no leading #, no leading "issue").
function parseLinks(text, selfId) {
  const out = [];
  // Frontmatter relationships
  const fm = /^---\s*$([\s\S]*?)^---\s*$/m.exec(text);
  if (fm) {
    for (const [field, relation] of [
      ['supersedes', 'supersedes'],
      ['amended-by', 'amends'],
      ['amends', 'amends'],
      ['related', 'related'],
      ['depends-on', 'depends-on'],
    ]) {
      const re = new RegExp(`^${field}:\\s*\\[?([^\\]\\n]+)\\]?$`, 'mi');
      const m = re.exec(fm[1]);
      if (m) for (const ref of extractAdrRefs(m[1])) {
        if (relation === 'supersedes') out.push({ from: ref, to: selfId, relation });
        else out.push({ from: selfId, to: ref, relation });
      }
    }
  }
  // Body relationship lines. #2474 Bug 3: loosened to accept both colon
  // placements (`**Supersedes:**` and `**Supersedes**:`) and an optional
  // parenthetical qualifier like `**Supersedes (partial):**` — same
  // tolerance as parseStatus.
  const REL = (label) => new RegExp(`^\\*\\*${label}(?:\\s*\\([^)]*\\))?:?\\*\\*:?\\s*(.+)$`, 'mi');
  const supersedes = REL('Supersedes').exec(text);
  if (supersedes) for (const ref of extractAdrRefs(supersedes[1])) out.push({ from: ref, to: selfId, relation: 'supersedes' });
  const amended = REL('(?:Amended[ -]by|Amends)').exec(text);
  if (amended) for (const ref of extractAdrRefs(amended[1])) out.push({ from: selfId, to: ref, relation: 'amends' });
  const related = REL('Related').exec(text);
  if (related) for (const ref of extractAdrRefs(related[1])) out.push({ from: selfId, to: ref, relation: 'related' });
  const dependsOn = REL('Depends[ -]on').exec(text);
  if (dependsOn) for (const ref of extractAdrRefs(dependsOn[1])) out.push({ from: selfId, to: ref, relation: 'depends-on' });
  return out;
}

function extractAdrRefs(s) {
  const refs = new Set();
  // Strip GitHub issue / commit references first to prevent false positives.
  const cleaned = s
    .replace(/#\d+/g, '')               // #1697
    .replace(/issue[s]?\s*\d+/gi, '')    // issue 1697
    .replace(/PR\s*\d+/gi, '')           // PR 1234
    .replace(/commit\s*[`a-f0-9]+/gi, '') // commit `abc123`
    .replace(/`[^`]*`/g, '');             // any backtick-quoted span
  const re = /\bADR-?(\d+)\b/gi;
  let m;
  // #2474 Bug 4: route every ADR ref through the same normalizer parseId
  // uses, so a Supersedes: ADR-0001 line lands on the same node key as
  // the file ADR-0001.md (was: padStart-then-strip produced ADR-001).
  while ((m = re.exec(cleaned))) refs.add(normalizeAdrId(m[1]));
  return [...refs];
}

function memoryStore(namespace, key, value) {
  const valueStr = typeof value === 'string' ? value : JSON.stringify(value);
  // #2474 Bug 1 (fatal): ADR titles like "ADR-005 — Repository …" contain
  // a U+2014 em-dash. \`npm exec\` runs argv validation BEFORE handing args
  // to the underlying bin, and \`commander\`-style argv with a non-ASCII
  // dash that starts an arg makes it reject with:
  //   npm error arg Argument starts with non-ascii dash, this is probably invalid: — …
  // Result: every store failed → \`Records stored: 0/N\`.
  //
  // Use the \`--flag=value\` form so npm sees a single \`--value=…\` token
  // and doesn't try to interpret the leading character of the value.
  // This works on both legacy and current npm; the underlying CLI accepts
  // \`--flag=value\` and \`--flag value\` equivalently.
  const r = spawnSync('npx', [
    CLI_PKG, 'memory', 'store',
    `--namespace=${namespace}`,
    `--key=${key}`,
    `--value=${valueStr}`,
  ], { stdio: ['ignore', 'pipe', 'pipe'], encoding: 'utf-8' });
  if (r.status !== 0) {
    if (/UNIQUE constraint/i.test(r.stderr || r.stdout || '')) return 'exists';
    return 'error: ' + (r.stderr || '').slice(0, 100);
  }
  return 'ok';
}

const dryRun = process.env.IMPORT_DRY_RUN === '1';
const fmt = process.env.IMPORT_FORMAT || 'markdown';

const files = findAdrs(ROOT);
const adrs = files.map(parseAdr);
const byId = new Map();
const allEdges = [];
for (const a of adrs) {
  byId.set(a.id, a);
  allEdges.push(...a.links);
}

let storedRecords = 0, storedEdges = 0;
const errors = [];
if (!dryRun) {
  for (const a of adrs) {
    const r = memoryStore('adr-patterns', `${a.id}::${basename(a.file, '.md')}`,
      `${a.title} — ${a.context || '(no context)'}\n\nfile: ${a.file}\nstatus: ${a.status}\ndate: ${a.date}\ntags: ${a.tags.join(',')}`);
    if (r === 'ok' || r === 'exists') storedRecords++;
    else errors.push(`${a.id} ${a.file}: ${r}`);
  }
  for (const e of allEdges) {
    const key = `${e.relation}:${e.from}->${e.to}:${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
    const r = memoryStore('adr-edges', key, JSON.stringify({ ...e, capturedAt: new Date().toISOString() }));
    if (r === 'ok' || r === 'exists') storedEdges++;
  }
}

const danglingRefs = allEdges.filter((e) => !byId.has(e.to));
const supersededIds = new Set(allEdges.filter((x) => x.relation === 'supersedes').map((x) => x.from));
const statusMismatches = [];
for (const id of supersededIds) {
  const a = byId.get(id);
  if (a && !/superseded/i.test(a.status)) statusMismatches.push({ id, status: a.status, file: a.file });
}

const byStatus = {};
for (const a of adrs) {
  const k = (a.status || 'unknown').toLowerCase();
  byStatus[k] = (byStatus[k] || 0) + 1;
}
const byRelation = {};
for (const e of allEdges) byRelation[e.relation] = (byRelation[e.relation] || 0) + 1;
const bySource = {};
for (const a of adrs) {
  const src = a.file.split('/docs/')[0];
  bySource[src] = (bySource[src] || 0) + 1;
}

const result = {
  scannedRoot: ROOT,
  total: adrs.length,
  sourceDirs: Object.keys(bySource).length,
  storedRecords, storedEdges, dryRun,
  byStatus, byRelation, bySource,
  edges: allEdges.length,
  danglingRefs, statusMismatches, errors,
};

if (fmt === 'json') {
  console.log(JSON.stringify(result, null, 2));
  process.exit(0);
}

console.log('## ADR Index Summary');
console.log('');
console.log(`Total ADRs: **${result.total}** across ${result.sourceDirs} source dirs (root: ${ROOT})`);
console.log(`Records stored to \`adr-patterns\`: ${result.storedRecords}/${result.total}${dryRun ? ' (dry-run, skipped)' : ''}`);
console.log(`Edges stored to \`adr-edges\`: ${result.storedEdges}/${result.edges}${dryRun ? ' (dry-run, skipped)' : ''}`);
console.log('');
console.log('### By status');
for (const [k, n] of Object.entries(byStatus).sort((a, b) => b[1] - a[1])) console.log(`- ${k}: ${n}`);
console.log('');
console.log(`### Relationships: **${result.edges}** edges`);
for (const [k, n] of Object.entries(byRelation).sort((a, b) => b[1] - a[1])) console.log(`- ${k}: ${n}`);
console.log('');
console.log('### Issues found');
console.log(`- Dangling refs (edge → non-existent ADR): ${danglingRefs.length}`);
for (const d of danglingRefs.slice(0, 10)) console.log(`  - ${d.relation} ${d.from} → ${d.to} (missing)`);
console.log(`- Status mismatches (superseded but not marked): ${statusMismatches.length}`);
for (const m of statusMismatches.slice(0, 10)) console.log(`  - ${m.id} status='${m.status}' (${m.file})`);
console.log(`- Storage errors: ${errors.length}`);
for (const e of errors.slice(0, 5)) console.log(`  - ${e}`);
console.log('');
console.log('### Source breakdown');
for (const [s, n] of Object.entries(bySource).sort((a, b) => b[1] - a[1]).slice(0, 12)) console.log(`- ${s}: ${n}`);