path: root/modules/test/test-memory-graph.js

// node modules/test/test-memory-graph.js
// test-memory-graph: five confirmed defects in module-memory-graph.js.
//
//   T1  addEdge() averaged weights, so every repeat of a strong edge pulled it down toward the
//       constant its caller passed. A relation that was used once at 0.9 and then observed at
//       0.5 decayed to 0.5 for being used. Evidence must be monotonic: stronger wins.
//   T2  recordTask() keyed task nodes on Date.now(), so re-submitting the same task created a
//       brand new node every millisecond instead of reinforcing the existing one.
//   T3  recordFileAccess() recorded the graph's own store (/.memory_graph/*) as user files,
//       so the graph observed itself and grew every time it saved.
//   T4  failed tool calls were persisted at weight 0.8 and successes at 0.5, which made
//       showStats() rank failures as the strongest relationships in the graph.
//   T5  concept ids were built from a whitespace-only split, so ids carried punctuation
//       ("concept:parser," ) and never matched an id built from the same word elsewhere.
//
// Set MEMORY_GRAPH_PATH to test a copy of the module (used to prove the test is RED on the
// pre-fix file). prune() is deliberately untouched and T6.5 pins that.
const fs = require('fs'), path = require('path');
const ROOT = path.join(__dirname, '..', 'core');
const MG_FILE = process.env.MEMORY_GRAPH_PATH || path.join(ROOT, 'module-memory-graph.js');

const store = {};
global.localStorage = {
    getItem: k => (k in store ? store[k] : null),
    setItem: (k, v) => { store[k] = String(v); },
    removeItem: k => { delete store[k]; }
};
global.indexedDB = { open() { throw new Error('no idb'); }, deleteDatabase() { throw new Error('no idb'); } };
global.window = global;

const load = (file, cls) => new Function(fs.readFileSync(file, 'utf8') + '\nreturn ' + cls + ';')();

let VFS, MG;
try {
    VFS = load(path.join(ROOT, 'virtual-fs.js'), 'VirtualFileSystem');
    MG = load(MG_FILE, 'MemoryGraphModule');
} catch (e) {
    console.log('FAIL could not load module — ' + e.message);
    process.exit(1);
}
if (typeof MG !== 'function') {
    console.log('FAIL module-memory-graph.js did not expose MemoryGraphModule (got ' + typeof MG + ')');
    process.exit(1);
}

let fails = 0, passes = 0;
function ok(name, cond, detail) {
    if (cond) { passes++; return; }
    fails++;
    console.log('FAIL ' + name + (detail ? ' — ' + detail : ''));
}

function mk() {
    const out = [];
    const vfs = new VFS();
    vfs.saveToStorage = () => true;
    const term = { print: (m, t) => out.push({ text: String(m), type: t || '' }), log: () => {}, fs: vfs };
    const mg = new MG();
    mg.init(term);
    return { mg, vfs, out, term };
}
const text = arr => arr.map(o => o.text).join('\n');
// Legacy task ids are Date.now()-based: make sure two calls land in different milliseconds so
// the RED run cannot pass by the two calls sharing one tick.
const tick = ms => { const t = Date.now(); while (Date.now() - t < ms) { /* spin */ } };
const iso = new Date('2024-01-05T10:00:00.000Z').toISOString();
const sessionOf = d => 'session-session-' + Math.floor(new Date(d).getTime() / 86400000);

// ─── T1: addEdge must not erode a strong edge ───
{
    const { mg } = mk();
    mg.addEdge('a', 'b', 'used', 0.9);
    mg.addEdge('a', 'b', 'used', 0.5);
    mg.addEdge('a', 'b', 'used', 0.5);
    const e = mg._edges[0];
    ok('T1.1 a strong edge is not eroded by repeated weak observations',
        !!e && e.weight === 0.9, 'weight ' + (e && e.weight));
    ok('T1.2 repeats still count as use', !!e && e.use_count === 3, 'use_count ' + (e && e.use_count));
    ok('T1.3 repeats do not duplicate the edge', mg._edges.length === 1, 'edges ' + mg._edges.length);

    mg.addEdge('a', 'b', 'used', 1);
    ok('T1.4 stronger evidence still raises the weight', e.weight === 1, 'weight ' + e.weight);
}

// ─── T2: task ids are content-derived, so identical tasks merge ───
{
    const { mg } = mk();
    const TASK = 'Read the parser and fix the lexer bug';
    mg.recordTask(TASK, 'ok', 'agent-a');
    tick(3);
    mg.recordTask(TASK, 'ok', 'agent-a');
    const tasks = () => Object.values(mg._nodes).filter(n => n.type === 'task');
    ok('T2.1 identical re-submits collapse to one task node',
        tasks().length === 1, 'task nodes: ' + tasks().map(t => t.id).join(','));
    ok('T2.2 the merged task records both uses',
        tasks().length === 1 && tasks()[0].access_count === 2, 'access_count ' + (tasks()[0] || {}).access_count);
    const ex = mg._edges.filter(e => e.type === 'executed');
    ok('T2.3 one executed edge, reinforced',
        ex.length === 1 && ex[0].use_count === 2, 'edges ' + ex.length + ' use_count ' + (ex[0] || {}).use_count);

    tick(3);
    mg.recordTask(TASK, 'ok', 'agent-b');
    ok('T2.4 a different agent keeps its own task node', tasks().length === 2,
        'task nodes: ' + tasks().map(t => t.id).join(','));

    tick(3);
    mg.recordTask('Deploy the release notes', 'ok', 'agent-a');
    ok('T2.5 different text gives a different task id', tasks().length === 3,
        'task nodes: ' + tasks().length);

    tick(3);
    mg.recordTask('  read the PARSER and fix the lexer bug  ', 'ok', 'agent-a');
    ok('T2.6 whitespace and case variants still dedupe', tasks().length === 3,
        'task nodes: ' + tasks().map(t => t.label).join(' | '));
}

// ─── T3: the graph must not record its own store ───
{
    const { mg } = mk();
    const nodesBefore = Object.keys(mg._nodes).length;
    const edgesBefore = mg._edges.length;
    mg.recordFileAccess('/.memory_graph/nodes.json');
    mg.recordFileAccess('/.memory_graph/edges.json', 'agent-a');
    mg.recordFileAccess('/.memory_graph/');
    mg.recordFileAccess('.memory_graph/export.json');

    const selfFiles = Object.keys(mg._nodes).filter(k => k.indexOf('.memory_graph') !== -1);
    ok('T3.1 store paths are not recorded as file nodes', selfFiles.length === 0, selfFiles.join(','));
    ok('T3.2 no extra nodes from the store', Object.keys(mg._nodes).length === nodesBefore,
        'before ' + nodesBefore + ' after ' + Object.keys(mg._nodes).length);
    ok('T3.3 no extra edges from the store', mg._edges.length === edgesBefore,
        'before ' + edgesBefore + ' after ' + mg._edges.length);

    mg.recordFileAccess('/src/app.js');
    ok('T3.4 ordinary paths are still recorded', !!mg.getNode('file', '/src/app.js'),
        Object.keys(mg._nodes).join(','));
    ok('T3.5 an ordinary path still gets its accessed edge',
        mg._edges.some(e => e.type === 'accessed' && e.target === 'file:/src/app.js'));
}

// ─── T4: a failure must not outrank a success ───
{
    const { mg, out } = mk();
    mg.recordToolCall('read_file', true);
    mg.recordToolCall('write_file', false);
    const used = mg._edges.find(e => e.type === 'used');
    const failed = mg._edges.find(e => e.type === 'failed');
    ok('T4.1 both outcomes are recorded', !!used && !!failed,
        mg._edges.map(e => e.type + ':' + e.weight).join(','));
    ok('T4.2 a failed tool call is weaker than a successful one',
        !!used && !!failed && failed.weight < used.weight,
        'used ' + (used || {}).weight + ' failed ' + (failed || {}).weight);
    ok('T4.3 success keeps its weight', !!used && used.weight === 0.5, String((used || {}).weight));

    const mark = out.length;
    mg.showStats();
    const stats = out.slice(mark).map(o => o.text).join('\n');
    const lines = stats.split('\n');
    const at = lines.findIndex(l => l.indexOf('Strongest relationships') !== -1);
    const strong = at === -1 ? [] : lines.slice(at + 1, at + 4).filter(Boolean);
    ok('T4.4 showStats still lists strongest relationships', strong.length >= 2, stats.slice(0, 300));
    ok('T4.5 the failure is not ranked above the success',
        strong.length >= 2 && strong[0].indexOf('used') !== -1 && strong[1].indexOf('failed') !== -1,
        strong.join(' | '));
}

// ─── T5: concept ids are slugged ───
{
    const { mg } = mk();
    mg.recordTask('Refactor: the Parser (v2!), then ship it.', 'done', 'agent-a');
    const concepts = Object.values(mg._nodes).filter(n => n.type === 'concept');
    ok('T5.1 task keywords become concept nodes', concepts.length > 0, 'none');
    ok('T5.2 concept ids carry no punctuation',
        concepts.every(c => /^concept:[a-z0-9]+(-[a-z0-9]+)*$/.test(c.id)),
        concepts.map(c => c.id).join(','));
    const involves = mg._edges.filter(e => e.type === 'involves');
    ok('T5.3 every involves edge points at a real concept node',
        involves.length > 0 && involves.every(e => !!mg._nodes[e.target]),
        involves.map(e => e.target).join(','));

    mg.recordConcept('Neural VFS, v2!', '/src/nvfs.js');
    ok('T5.4 recordConcept slugifies its id', !!mg.getNode('concept', 'neural-vfs-v2'),
        Object.keys(mg._nodes).filter(k => k.indexOf('concept:') === 0).join(','));
    ok('T5.5 the concept→file edge uses the slugged id',
        mg._edges.some(e => e.type === 'maps_to' && e.source === 'concept:neural-vfs-v2'));
}

// ─── T6: graphs already stored keep loading, and prune() is untouched ───
{
    const { vfs } = mk();
    const recent = new Date().toISOString();
    const legacyNodes = {
        'session:session-legacy': { id: 'session:session-legacy', type: 'session', label: 'Legacy session',
            metadata: {}, created: iso, updated: iso, access_count: 1, namespace: 'global', source_session: 'session-legacy' },
        'task:task-1700000000000': { id: 'task:task-1700000000000', type: 'task', label: 'legacy task',
            metadata: { full: 'legacy task' }, created: iso, updated: iso, access_count: 1,
            namespace: 'global', source_session: 'session-legacy' },
        'concept:orphan': { id: 'concept:orphan', type: 'concept', label: 'orphan',
            metadata: {}, created: iso, updated: iso, access_count: 1, namespace: 'global', source_session: 'session-legacy' }
    };
    const legacyEdges = [
        { source: 'session:session-legacy', target: 'task:task-1700000000000', type: 'executed',
          weight: 0.6, created: iso, last_used: recent, use_count: 1, source_agents: [] },
        { source: 'session:session-legacy', target: 'tool:write_file', type: 'failed',
          weight: 0.8, created: iso, last_used: recent, use_count: 1, source_agents: ['agent-a'] }
    ];
    vfs.mkdirRecursive('/.memory_graph');
    vfs.fsWriteJSON('/.memory_graph/nodes.json', legacyNodes);
    vfs.fsWriteJSON('/.memory_graph/edges.json', legacyEdges);

    const out2 = [];
    const mg2 = new MG();
    mg2.init({ print: (m, t) => out2.push({ text: String(m), type: t || '' }), log: () => {}, fs: vfs });

    ok('T6.1 a legacy graph still loads', Object.keys(mg2._nodes).length >= 4,
        'nodes ' + Object.keys(mg2._nodes).length);
    ok('T6.2 legacy Date.now() task ids are still addressable',
        !!mg2.getNode('task', 'task-1700000000000'));
    ok('T6.3 legacy edges keep their shape',
        mg2._edges.every(e => Array.isArray(e.source_agents) && typeof e.weight === 'number'),
        JSON.stringify(mg2._edges));
    const legFailed = mg2._edges.find(e => e.type === 'failed');
    ok('T6.4 a legacy 0.8 failure is capped below success on load',
        !!legFailed && legFailed.weight < 0.5, 'weight ' + (legFailed || {}).weight);
    ok('T6.5 unrelated legacy weights are untouched',
        (mg2._edges.find(e => e.type === 'executed') || {}).weight === 0.6,
        String((mg2._edges.find(e => e.type === 'executed') || {}).weight));

    // prune() is correct and must stay correct: it drops zero-degree nodes and keeps the rest.
    mg2.addNode('concept', 'second-orphan', 'second orphan');
    const orphanPresent = !!mg2._nodes['concept:second-orphan'];
    mg2.prune();
    ok('T6.6 prune still drops zero-degree nodes (prune untouched)',
        orphanPresent && !mg2._nodes['concept:second-orphan'], Object.keys(mg2._nodes).join(','));
    ok('T6.7 prune keeps edges that are still in use', mg2._edges.length === 2,
        'edges ' + mg2._edges.length);
}

// T7. slugging must not erase non-Latin scripts: an a-z0-9 slug turned a Cyrillic task
// into '' (no concept nodes at all) and German 'Ärger' into 'rger'.
{
    const { mg } = mk();
    ok('T7.1 cyrillic slug is not empty', mg._slug('Исправить парсер ошибок') === 'исправить-парсер-ошибок', mg._slug('Исправить парсер ошибок'));
    ok('T7.2 cjk slug is not empty', mg._slug('修复解析器的错误') === '修复解析器的错误', mg._slug('修复解析器的错误'));
    ok('T7.3 accents survive', mg._slug('Ärger mit Größe') === 'ärger-mit-größe', mg._slug('Ärger mit Größe'));
    ok('T7.4 ascii is unchanged', mg._slug('Neural VFS, v2!') === 'neural-vfs-v2', mg._slug('Neural VFS, v2!'));
    mg.recordTask('Исправить парсер ошибок', 'done', 'agent-ru');
    const cyr = Object.keys(mg._nodes).filter(k => k.indexOf('concept:') === 0);
    ok('T7.5 a cyrillic task still yields concept nodes', cyr.length > 0, JSON.stringify(cyr.slice(0, 3)));
}

console.log('test-memory-graph: ' + passes + ' passed, ' + fails + ' failed');
process.exit(fails ? 1 : 0);