perf(understand): parallelise file I/O in compute-batches + extract-import-map (#76)

The /understand pipeline reads every code file twice during analysis:
once in compute-batches (`extractExports` for the cross-batch neighbour
map) and once again in extract-import-map (per-language config loaders).
Both sites used sequential `readFileSync` loops, so on the iOS repo in
issue #226 (~15k files) the disk-read time was effectively serialised
behind a single libuv thread while the rest of the pool sat idle.

## Changes

- `extractExports` now batches files into `IO_PARALLELISM = 64` slices
  and issues all `readFile` calls in each slice through `Promise.all`,
  letting libuv's worker-thread pool overlap disk reads. The
  tree-sitter parse stays on the main thread because `web-tree-sitter`
  is single-threaded WASM — pipelining the I/O while parses run is
  where the wall-time savings come from.
- `loadTsConfigs`, `loadGoModules`, `loadPhpAutoloads` and
  `buildResolutionContext` switch to async / `Promise.all` for the
  same reason. `buildResolutionContext` also runs the three loader
  passes concurrently (`Promise.all([...])`) since they're independent.
- A small `readFilesParallel(paths)` helper is added at the top of
  `extract-import-map.mjs` so the three loaders share the same
  error-preserving shape.

## Why behavior stays identical

- Each loader collects its candidate paths in `files[]` order *before*
  issuing reads, then iterates `reads` in the same order to emit
  warnings + populate output maps. So stderr order and the final map
  contents are byte-identical to the previous sequential loops.
- `extractExports` collects per-file errors in-place in the
  `Promise.all` callbacks and emits warnings during the post-read
  serial loop, again in chunk order — so warning text and order match
  the previous implementation.
- Tree-sitter parsing is unchanged: parses still run serially on the
  main thread, just with reads pipelined alongside.

## What's NOT in this PR

- `buildFingerprintStore` and `analyzeChanges` in `core/fingerprint.ts`
  have the same sequential pattern. They're left alone here because
  they're part of the public `@understand-anything/core` API; making
  them async would be a breaking change worth its own discussion.
  Internal-only `.mjs` scripts are safe to refactor without API churn.
- No change to scan-project: most of its sync I/O is `statSync`
  (metadata, not content) plus a handful of small `.gitignore` /
  `.understandignore` reads. The parallelism win is marginal there.

## Verification

- `pnpm lint` clean
- `pnpm --filter @understand-anything/core build` clean
- `pnpm --filter @understand-anything/skill build` clean
- `pnpm test`: 196/196 — including
  `test_compute_batches.test.mjs` (19 tests) and
  `test_extract_import_map.test.mjs` (40 tests), which exercise both
  changed pipelines end-to-end with fixture projects. No output
  diff vs main.

Refs #76

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Tirth Kanani
2026-05-31 20:25:39 +01:00
Unverified
parent 470cc01dc5
commit eea73b656d
2 changed files with 123 additions and 53 deletions
@@ -13,10 +13,19 @@
*/
import { readFileSync, writeFileSync, existsSync, realpathSync } from 'node:fs';
import { readFile } from 'node:fs/promises';
import { dirname, join, resolve } from 'node:path';
import { fileURLToPath, pathToFileURL } from 'node:url';
import { createRequire } from 'node:module';
/**
* Chunk size for parallel file I/O. Bounded so a 15k-file repo doesn't try
* to open every descriptor at once (would hit `EMFILE`) while still keeping
* libuv's worker-thread pool saturated. Empirically chosen to keep memory
* around tens of MB even when the average file is ~10 KB.
*/
const IO_PARALLELISM = 64;
const __filename = fileURLToPath(import.meta.url);
const PLUGIN_ROOT = resolve(dirname(__filename), '../..');
const require = createRequire(resolve(PLUGIN_ROOT, 'package.json'));
@@ -57,31 +66,55 @@ async function extractExports(projectRoot, codeFiles) {
}
const exportsByPath = new Map();
for (const file of codeFiles) {
const abs = join(projectRoot, file.path);
let content;
try {
content = readFileSync(abs, 'utf-8');
} catch (err) {
process.stderr.write(
`Warning: compute-batches: exports extraction failed for ${file.path} ` +
`(read error: ${err.message}) — symbols=[] in neighborMap — ` +
`cross-batch edges to this file limited to file-level\n`,
);
exportsByPath.set(file.path, []);
continue;
}
try {
const analysis = registry.analyzeFile(file.path, content);
const names = (analysis?.exports || []).map(e => e.name).filter(Boolean);
exportsByPath.set(file.path, names);
} catch (err) {
process.stderr.write(
`Warning: compute-batches: exports extraction failed for ${file.path} ` +
`(analyze error: ${err.message}) — symbols=[] in neighborMap — ` +
`cross-batch edges to this file limited to file-level\n`,
);
exportsByPath.set(file.path, []);
// I/O is parallelised in bounded chunks (libuv worker threads handle the
// disk reads concurrently) while the actual tree-sitter parse stays on
// the main thread, since web-tree-sitter is single-threaded WASM. For a
// 15k-file iOS repo (#226), the sequential `readFileSync` loop dominated;
// letting reads pipeline drops wall time roughly proportional to the
// share of the loop spent waiting on disk.
for (let start = 0; start < codeFiles.length; start += IO_PARALLELISM) {
const slice = codeFiles.slice(start, start + IO_PARALLELISM);
// Read every file in the slice concurrently. Errors per file are
// captured in-place so a single bad file does not abort the chunk.
const reads = await Promise.all(
slice.map(async (file) => {
const abs = join(projectRoot, file.path);
try {
const content = await readFile(abs, 'utf-8');
return { file, content, readError: null };
} catch (err) {
return { file, content: null, readError: err };
}
}),
);
// Serialise the CPU-bound tree-sitter work and the stderr warning emits
// so log order remains identical to the previous sequential loop. This
// also keeps existing fixture-comparison tests stable.
for (const { file, content, readError } of reads) {
if (readError) {
process.stderr.write(
`Warning: compute-batches: exports extraction failed for ${file.path} ` +
`(read error: ${readError.message}) — symbols=[] in neighborMap — ` +
`cross-batch edges to this file limited to file-level\n`,
);
exportsByPath.set(file.path, []);
continue;
}
try {
const analysis = registry.analyzeFile(file.path, content);
const names = (analysis?.exports || []).map(e => e.name).filter(Boolean);
exportsByPath.set(file.path, names);
} catch (err) {
process.stderr.write(
`Warning: compute-batches: exports extraction failed for ${file.path} ` +
`(analyze error: ${err.message}) — symbols=[] in neighborMap — ` +
`cross-batch edges to this file limited to file-level\n`,
);
exportsByPath.set(file.path, []);
}
}
}
return exportsByPath;
@@ -37,6 +37,29 @@ import { createRequire } from 'node:module';
import { dirname, resolve, join, posix } from 'node:path';
import { fileURLToPath, pathToFileURL } from 'node:url';
import { existsSync, readFileSync, realpathSync, writeFileSync } from 'node:fs';
import { readFile } from 'node:fs/promises';
/**
* Read a list of files concurrently while preserving result order. Failures
* are returned in-place as `{ raw: null, err }` so callers can emit the same
* per-file warnings they did under the previous sequential `readFileSync`
* loops.
*
* `paths` is a list of `{ key, absPath }` pairs; `key` is whatever the caller
* wants to attach the result to (typically a project-relative POSIX path).
*/
async function readFilesParallel(paths) {
return Promise.all(
paths.map(async ({ key, absPath }) => {
try {
const raw = await readFile(absPath, 'utf-8');
return { key, raw, err: null };
} catch (err) {
return { key, raw: null, err };
}
}),
);
}
const __dirname = dirname(fileURLToPath(import.meta.url));
// skills/understand/ -> plugin root is two dirs up
@@ -180,20 +203,25 @@ function parseTsConfigText(raw) {
* where the stripper damaged a string literal containing `//`.
* 3. If both fail, warn and skip — that tsconfig contributes no aliases.
*/
function loadTsConfigs(projectRoot, files) {
async function loadTsConfigs(projectRoot, files) {
const out = new Map();
// Collect the candidate paths in the original file order before reading,
// so warning emit order matches the previous sequential implementation.
const candidates = [];
for (const f of files) {
const p = toPosix(f.path);
const base = p.includes('/') ? p.slice(p.lastIndexOf('/') + 1) : p;
if (base !== 'tsconfig.json') continue;
const absPath = join(projectRoot, p);
if (!existsSync(absPath)) continue;
let raw;
try {
raw = readFileSync(absPath, 'utf-8');
} catch (err) {
candidates.push({ key: p, absPath });
}
const reads = await readFilesParallel(candidates);
for (const { key: p, raw, err } of reads) {
if (err) {
// absPath isn't carried through the helper return shape; reconstruct it.
process.stderr.write(
`Warning: extract-import-map: tsconfig.json at ${absPath} failed ` +
`Warning: extract-import-map: tsconfig.json at ${join(projectRoot, p)} failed ` +
`to read (${err.message}) — path aliases from this config will ` +
`not be applied — relative imports unaffected\n`,
);
@@ -202,7 +230,7 @@ function loadTsConfigs(projectRoot, files) {
const parsed = parseTsConfigText(raw);
if (!parsed) {
process.stderr.write(
`Warning: extract-import-map: tsconfig.json at ${absPath} failed ` +
`Warning: extract-import-map: tsconfig.json at ${join(projectRoot, p)} failed ` +
`to parse — path aliases from this config will not be applied ` +
`— relative imports unaffected\n`,
);
@@ -237,20 +265,20 @@ function loadTsConfigs(projectRoot, files) {
* The resolver uses each module's prefix to translate
* `import "github.com/foo/bar/x"` into the project-internal `x/<file>.go`.
*/
function loadGoModules(projectRoot, files) {
async function loadGoModules(projectRoot, files) {
const out = new Map();
const candidates = [];
for (const f of files) {
const p = toPosix(f.path);
const base = p.includes('/') ? p.slice(p.lastIndexOf('/') + 1) : p;
if (base !== 'go.mod') continue;
const absPath = join(projectRoot, p);
if (!existsSync(absPath)) continue;
let raw;
try {
raw = readFileSync(absPath, 'utf-8');
} catch {
continue;
}
candidates.push({ key: p, absPath });
}
const reads = await readFilesParallel(candidates);
for (const { key: p, raw, err } of reads) {
if (err) continue;
let moduleName = '';
for (const line of raw.split(/\r?\n/)) {
const trimmed = line.replace(/\/\/.*$/, '').trim();
@@ -306,10 +334,17 @@ function findNearestConfigDir(startDir, configMap) {
*
* Build once; pass everywhere.
*/
function buildResolutionContext(projectRoot, files) {
async function buildResolutionContext(projectRoot, files) {
const fileSet = new Set(files.map(f => toPosix(f.path)));
const tsConfigs = loadTsConfigs(projectRoot, files);
const goModules = loadGoModules(projectRoot, files);
// The three config-loader passes are independent and each does its own
// batched parallel I/O; run them concurrently so the wait for a slow
// tsconfig.json read doesn't block go.mod / composer.json scanning.
const [tsConfigs, goModules, phpAutoloads] = await Promise.all([
loadTsConfigs(projectRoot, files),
loadGoModules(projectRoot, files),
loadPhpAutoloads(projectRoot, files),
]);
// Index .go files by their parent directory so the Go resolver can
// expand a package-level import to all member .go files in O(1).
@@ -331,8 +366,6 @@ function buildResolutionContext(projectRoot, files) {
const kotlinIndex = buildSuffixIndex(files, p => p.endsWith('.kt'));
const csIndex = buildSuffixIndex(files, p => p.endsWith('.cs'));
const phpAutoloads = loadPhpAutoloads(projectRoot, files);
return {
projectRoot,
fileSet,
@@ -1019,20 +1052,22 @@ function parseComposerAutoloadText(raw) {
* at the bad file and skips it. The rest of the project's PHP imports keep
* resolving via whichever composer.json files parsed cleanly.
*/
function loadPhpAutoloads(projectRoot, files) {
async function loadPhpAutoloads(projectRoot, files) {
const out = new Map();
const candidates = [];
for (const f of files) {
const p = toPosix(f.path);
const base = p.includes('/') ? p.slice(p.lastIndexOf('/') + 1) : p;
if (base !== 'composer.json') continue;
const absPath = join(projectRoot, p);
if (!existsSync(absPath)) continue;
let raw;
try {
raw = readFileSync(absPath, 'utf-8');
} catch (err) {
candidates.push({ key: p, absPath });
}
const reads = await readFilesParallel(candidates);
for (const { key: p, raw, err } of reads) {
if (err) {
process.stderr.write(
`Warning: extract-import-map: composer.json at ${absPath} failed ` +
`Warning: extract-import-map: composer.json at ${join(projectRoot, p)} failed ` +
`to read (${err.message}) — PSR-4 namespace mapping from this ` +
`composer.json unavailable — PHP imports under this package ` +
`will not resolve\n`,
@@ -1042,7 +1077,7 @@ function loadPhpAutoloads(projectRoot, files) {
const parsed = parseComposerAutoloadText(raw);
if (parsed === null) {
process.stderr.write(
`Warning: extract-import-map: composer.json at ${absPath} failed ` +
`Warning: extract-import-map: composer.json at ${join(projectRoot, p)} failed ` +
`to parse — PSR-4 namespace mapping unavailable — PHP imports ` +
`under this package will not resolve\n`,
);
@@ -1412,8 +1447,10 @@ async function main() {
);
}
// Build resolution context (cached configs)
const ctx = buildResolutionContext(projectRoot, files);
// Build resolution context (cached configs). The loader pass for the
// tsconfig/go.mod/composer.json files inside is parallelised — see
// `buildResolutionContext`.
const ctx = await buildResolutionContext(projectRoot, files);
const importMap = {};
let filesWithImports = 0;