Files
codegraph/__tests__/sync-rebuild-convergence.test.ts
T
Colby McHenryandClaude Opus 5 03893b0ab9 CG-33: converge incremental sync with a full rebuild
A live, auto-synced index did not converge to a clean rebuild of the same
tree — 4.3% of distinct edges wrong in both directions on this repo's own
index, overwhelmingly `calls`, which is what flow queries traverse and what
explore's file ranking weights. Silent: nothing warned, and the symptom read
as "codegraph isn't very good" rather than "this index needs rebuilding."

Two causes, and the fix needed both. Resolution binds a reference to one of
the same-named definitions PROJECT-WIDE, so a definition appearing or
vanishing changes the correct answer for references in files the sync never
touches — and those references resolved successfully once, which deletes
their unresolved_refs row, leaving nothing to revisit them with (#1240's
retry only revisits refs parked as failed). Separately, when nothing
disambiguated the candidates the winner came down to rowid, i.e. the order
files happened to be WRITTEN, which differs between a scan-order full index
and a sync that appends each file as it changes. That second one is why
re-resolution alone could not converge: re-resolving against the identical
graph still picked a different candidate.

So getNodesByName now orders by (file_path, start_line) — a property of the
code, not of the write order — and sync computes a definitionDelta and
re-opens the resolution edges whose answer it may have invalidated,
re-inserting each as the reference that created it for the orphan sweep to
bind against the post-sync graph.

The delta compares `file\0name` pairs per file rather than one name set over
the batch: a commit that adds `collect` to a new file while an unrelated
changed file already defines `collect` cancels out of a batch-wide set, and
that miss was the largest residual class in the first measurement.

Conservative where the failure modes are asymmetric — a wrong deletion is a
permanent edge loss, a missed rebind is only residual drift. Edges without a
refName stamp are never touched (nothing to restore them from), sources the
sync already re-extracted are skipped, and a per-name ceiling declines the
generic names. Edges are deleted before the sweep re-inserts, since
INSERT OR IGNORE against idx_edges_identity would otherwise keep both rows
when a reference rebinds elsewhere.

Replaying real commits of this repo through sync, then diffing against a
rebuild: 16 commits 48 -> 0; 80 commits 1,634 -> 361, with the actively
misleading direction (stale edges the index keeps asserting) 671 -> 2.
Index and sync wall-clock are unchanged; the ORDER BY costs 18% per uncached
name lookup, which never reaches wall-clock because the resolver memoizes it.

The 357-edge residual at 80 commits is one pre-existing class: refs to
generic names (`push`, `join`) parked above #1240's per-name retry ceiling,
which a rebuild resolves into cross-language garbage — a TS test file
"calling" an R method. Converging there would mean manufacturing wrong edges,
so it is left alone. And no drift metric in `codegraph status`: it cannot be
computed without the rebuild it would be recommending, and a proxy would fire
on that residual and train users to ignore it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 04:32:40 -05:00

272 lines
13 KiB
TypeScript

/**
* Incremental sync must converge to a full rebuild (CG-33).
*
* A long-lived, auto-synced index silently diverged from a clean rebuild of the
* identical tree: 4.3% of distinct edges wrong, in BOTH directions, on
* codegraph's own repo. Two mechanisms, both exercised here:
*
* 1. Resolution binds a reference to one of the same-named definitions
* PROJECT-WIDE, so adding or removing a definition changes the answer for
* references in files the sync never touches. Those references resolved once
* and their rows were deleted, so nothing revisited them — the index kept an
* answer that was only correct against an older graph.
* 2. When nothing disambiguated the candidates, the winner was whichever row
* the index scan reached first — i.e. the order files were WRITTEN. A full
* index writes in scan order; a sync appends each file as it changes, so the
* same tree resolved differently depending on how the index was built.
*
* The assertions here compare the whole edge SET, never counts: the divergence
* is bidirectional and nets out of a total (raw rows differed by 0.7% while
* 4.3% of edges were wrong), so a count check passes on a broken index.
*/
import { describe, it, expect, beforeEach, afterEach } from 'vitest';
import * as fs from 'fs';
import * as path from 'path';
import * as os from 'os';
import CodeGraph from '../src/index';
import { createDatabase } from '../src/db/sqlite-adapter';
describe('Incremental sync converges to a full rebuild (CG-33)', () => {
let testDir: string;
let cg: CodeGraph;
const write = (rel: string, content: string) => {
const full = path.join(testDir, rel);
fs.mkdirSync(path.dirname(full), { recursive: true });
fs.writeFileSync(full, content);
};
/**
* Every edge as a `source|target|kind` triple, read from the database with a
* second read-only connection. Node ids are `sha256(filePath:kind:name:line)`,
* so for an identical tree they are identical across a sync and a rebuild —
* which is what makes the two sets directly comparable.
*/
const edgeSet = (): Set<string> => {
const { db } = createDatabase(path.join(testDir, '.codegraph', 'codegraph.db'), { readOnly: true });
try {
const rows = db.prepare('SELECT source, target, kind FROM edges').all() as Array<{
source: string;
target: string;
kind: string;
}>;
return new Set(rows.map((r) => `${r.source}|${r.target}|${r.kind}`));
} finally {
db.close();
}
};
/** Human-readable diff, so a failure names the edges instead of just a count. */
const describeDiff = (synced: Set<string>, rebuilt: Set<string>): string => {
const missing = [...rebuilt].filter((e) => !synced.has(e));
const stale = [...synced].filter((e) => !rebuilt.has(e));
return `missing from synced: ${missing.length}, stale in synced: ${stale.length}`;
};
/**
* Rebuild the index from scratch over the CURRENT tree and return its edge
* set. `indexAll` recreates the database file, so this is the same ground
* truth a user gets from `codegraph index`.
*/
const rebuildEdgeSet = async (): Promise<Set<string>> => {
await cg.indexAll();
return edgeSet();
};
beforeEach(() => {
testDir = fs.mkdtempSync(path.join(os.tmpdir(), 'codegraph-cg33-'));
});
afterEach(() => {
cg?.destroy();
if (fs.existsSync(testDir)) fs.rmSync(testDir, { recursive: true, force: true });
});
/**
* The originating shape. `caller.ts` calls `pct` with no import, so it binds
* by name; at index time `zeta.ts` is the only definition. A later sync adds
* `alpha.ts`, which sorts FIRST and is therefore the rebuild's answer — but
* `caller.ts` never changes, so nothing re-resolves it.
*/
it('rebinds references in UNCHANGED files when a sync adds a competing definition', async () => {
write('src/caller.ts', `export function run(): number {\n return pct(1);\n}\n`);
write('src/zeta.ts', `export function pct(n: number): number {\n return n;\n}\n`);
cg = CodeGraph.initSync(testDir, { config: { include: ['**/*.ts'], exclude: [] } });
await cg.indexAll();
write('src/alpha.ts', `export function pct(n: number): number {\n return n * 2;\n}\n`);
const result = await cg.sync();
expect(result.filesAdded).toBe(1);
expect(result.definitionDelta).toContain('pct');
const synced = edgeSet();
const rebuilt = await rebuildEdgeSet();
expect(describeDiff(synced, rebuilt)).toBe('missing from synced: 0, stale in synced: 0');
});
/**
* The mirror direction: removing a definition narrows the candidate set too,
* so the delta must include names the sync DROPPED, not just names it added.
*
* This one already converged before the fix — a removal cascades the edge
* away and the #1240 removal path resurrects it, so the reference gets
* re-resolved for free. It is here as a standing guard on the invariant, and
* because the removal half of the delta has no other coverage: an
* implementation that only sampled post-sync names would still pass every
* other test in this file.
*/
it('rebinds references in UNCHANGED files when a sync removes a competing definition', async () => {
write('src/caller.ts', `export function run(): number {\n return pct(1);\n}\n`);
write('src/alpha.ts', `export function pct(n: number): number {\n return n;\n}\n`);
write('src/zeta.ts', `export function pct(n: number): number {\n return n * 2;\n}\n`);
cg = CodeGraph.initSync(testDir, { config: { include: ['**/*.ts'], exclude: [] } });
await cg.indexAll();
fs.rmSync(path.join(testDir, 'src', 'alpha.ts'));
const result = await cg.sync();
expect(result.filesRemoved).toBe(1);
const synced = edgeSet();
const rebuilt = await rebuildEdgeSet();
expect(describeDiff(synced, rebuilt)).toBe('missing from synced: 0, stale in synced: 0');
});
/**
* The delta must be computed per FILE. Comparing one name set across the whole
* changed batch cancels a name that is added in one changed file while another
* changed file already defined it — which is precisely the shape a commit that
* splits a module out has, and it was the largest residual class in the first
* measurement of this fix.
*/
it('flags a name added in one changed file even when another changed file already defines it', async () => {
write('src/caller.ts', `export function run(): number {\n return pct(1);\n}\n`);
write('src/zeta.ts', `export function pct(n: number): number {\n return n;\n}\nexport function keep(): number {\n return 0;\n}\n`);
cg = CodeGraph.initSync(testDir, { config: { include: ['**/*.ts'], exclude: [] } });
await cg.indexAll();
// One commit: a NEW file gains `pct`, and the file that already had `pct`
// is edited too (so a batch-wide name set would see `pct` on both sides).
write('src/alpha.ts', `export function pct(n: number): number {\n return n * 2;\n}\n`);
write('src/zeta.ts', `export function pct(n: number): number {\n return n + 1;\n}\nexport function keep(): number {\n return 0;\n}\n`);
const result = await cg.sync();
expect(result.definitionDelta).toContain('pct');
const synced = edgeSet();
const rebuilt = await rebuildEdgeSet();
expect(describeDiff(synced, rebuilt)).toBe('missing from synced: 0, stale in synced: 0');
});
/**
* The realistic case the issue was filed from: many edits driven through sync
* one after another, the way a watcher or a `git pull` applies them. Drift
* accumulated across syncs, so a single-edit test would not have caught it.
*/
it('stays converged across a sequence of adds, edits, renames and deletes', async () => {
write('src/caller.ts', `export function run(): number {\n return pct(1) + fmt(2) + collect(3);\n}\n`);
write('src/util/zeta.ts', `export function pct(n: number): number {\n return n;\n}\n`);
write('src/util/omega.ts', `export function fmt(n: number): number {\n return n;\n}\n`);
cg = CodeGraph.initSync(testDir, { config: { include: ['**/*.ts'], exclude: [] } });
await cg.indexAll();
// 1. add a competing `pct` that sorts before the existing one
write('src/util/alpha.ts', `export function pct(n: number): number {\n return n * 2;\n}\n`);
await cg.sync();
// 2. body-only edit — must produce NO definition delta, so the common sync
// pays nothing for this machinery
write('src/util/alpha.ts', `export function pct(n: number): number {\n return n * 3;\n}\n`);
const bodyOnly = await cg.sync();
expect(bodyOnly.filesModified).toBe(1);
expect(bodyOnly.definitionDelta).toBeUndefined();
// 3. a rename: `fmt` moves out of omega.ts into a file that sorts first
write('src/util/omega.ts', `export function other(n: number): number {\n return n;\n}\n`);
write('src/util/beta.ts', `export function fmt(n: number): number {\n return n;\n}\n`);
await cg.sync();
// 4. a symbol appears for a reference that never resolved at all
write('src/util/gamma.ts', `export function collect(n: number): number {\n return n;\n}\n`);
await cg.sync();
// 5. delete the current `pct` winner, so the reference must fall back...
fs.rmSync(path.join(testDir, 'src', 'util', 'alpha.ts'));
await cg.sync();
// 6. ...and then a later sync introduces a new winner ahead of it again.
// Ending here rather than on the delete matters: after the delete the
// binding happens to land back where it started, which a broken index
// also reaches. The final state must be one only re-resolution reaches.
write('src/util/aaa.ts', `export function pct(n: number): number {\n return n * 5;\n}\n`);
await cg.sync();
const synced = edgeSet();
expect(synced.size).toBeGreaterThan(0);
const rebuilt = await rebuildEdgeSet();
expect(describeDiff(synced, rebuilt)).toBe('missing from synced: 0, stale in synced: 0');
});
/**
* Guards the escape hatch itself: with the rebind pass off, the same sequence
* must still produce a structurally sound index (no lost or orphaned edges) —
* just a drifted one. If this ever fails, the pass is doing something the
* kill switch cannot undo.
*/
it('CODEGRAPH_NO_REBIND=1 disables the pass without corrupting the index', async () => {
write('src/caller.ts', `export function run(): number {\n return pct(1);\n}\n`);
write('src/zeta.ts', `export function pct(n: number): number {\n return n;\n}\n`);
cg = CodeGraph.initSync(testDir, { config: { include: ['**/*.ts'], exclude: [] } });
await cg.indexAll();
const before = edgeSet();
process.env.CODEGRAPH_NO_REBIND = '1';
try {
write('src/alpha.ts', `export function pct(n: number): number {\n return n * 2;\n}\n`);
await cg.sync();
} finally {
delete process.env.CODEGRAPH_NO_REBIND;
}
const after = edgeSet();
// Every edge that existed before is still there — the pass is the only
// thing that would have re-opened them, and it did not run.
for (const edge of before) expect(after.has(edge)).toBe(true);
});
});
/**
* Resolution's candidate order must be a property of the CODE, not of the order
* rows were written. This is the half of CG-33 that a re-resolution pass alone
* cannot fix: without it, re-resolving a reference against the very same graph
* can still pick a different winner than a rebuild does.
*/
describe('Same-name candidate order is content-derived, not insertion-derived (CG-33)', () => {
let testDir: string;
let cg: CodeGraph;
afterEach(() => {
cg?.destroy();
if (fs.existsSync(testDir)) fs.rmSync(testDir, { recursive: true, force: true });
});
it('getNodesByName orders by (file_path, start_line) even when rows were written in another order', async () => {
testDir = fs.mkdtempSync(path.join(os.tmpdir(), 'codegraph-cg33-order-'));
fs.mkdirSync(path.join(testDir, 'src'), { recursive: true });
fs.writeFileSync(path.join(testDir, 'src', 'mid.ts'), `export function pad(): void {}\nexport function dup(): number {\n return 2;\n}\n`);
fs.writeFileSync(path.join(testDir, 'src', 'zeta.ts'), `export function dup(): number {\n return 1;\n}\n`);
cg = CodeGraph.initSync(testDir, { config: { include: ['**/*.ts'], exclude: [] } });
await cg.indexAll();
// A sync APPENDS this file's nodes, so `alpha.ts` gets the highest rowids
// despite sorting first — exactly the divergence a full index never has,
// and the reason candidate order cannot come from the physical row order.
fs.writeFileSync(path.join(testDir, 'src', 'alpha.ts'), `export function dup(): number {\n return 3;\n}\n`);
await cg.sync();
const keys = cg.getNodesByName('dup').map((n) => `${n.filePath}:${String(n.startLine).padStart(6, '0')}`);
expect(keys.length).toBeGreaterThanOrEqual(3);
expect(keys).toEqual([...keys].sort());
expect(keys[0]).toContain('src/alpha.ts');
});
});