Files
faunosteo/scripts/browser/r4-import-page/worker.js
T
bchanot 49591c53cd test(tests): R4.3a browser re-measure of large files, leak series, WebKit repetition (stream sink, BYOB)
Streaming into FileBytes removes the file's JavaScript copy in Chromium and Firefox
(50 MiB: 65 and 67 MiB summed growth against 56 and 79 for the TypeScript); WebKit keeps
about 180 MiB for any file read inside a worker (272 against 76), a main-thread BYOB read
feeding the worker's sink measured at 106. Corpus 666/678 on the stream path. Leak series
(3 engines × 3 documents × 3 repetitions): Chromium ≤ 0.5 MiB over the last 5, Firefox
within ±3 (idle only), WebKit copy +13 then level (12-import series −2.3). 20 dense
imports in WebKit: no worker error event.
2026-10-18 18:50:47 +02:00

274 lines
9.1 KiB
JavaScript

// Import worker of the R4 measurement page (R4.2e, R4.3a), the shape plan
// §B.1.5 gives the application at R4.3: a module worker created for one
// import, the glue imported statically, the module received from the main
// thread (structured clone) or fetched from its URL, the File read here, one
// SheetHandle, one result message. The main thread terminates it in every
// case. Two read paths, chosen per task (`readPath`):
// stream (default, R4.3a) `file.stream()` pushed chunk by chunk into
// a `FileBytes` sink in linear memory, then `import_pdf_bytes`
// or `restore_copy_bytes` (the core decodes the copy): the
// file lives once, in linear memory. A BYOB reader reads into
// one reused 1 MiB buffer (no garbage chunk per read); a
// stream that refuses BYOB is read with its default reader,
// each chunk dropped once pushed; without `stream` the file
// is read whole and pushed once. `readVia` records which;
// arrayBuffer (R4.2e, kept to measure both) one ArrayBuffer of the file,
// `import_pdf` (the glue copies it into linear memory) or the
// strict TextDecoder then `draft.restore`.
import init, { FileBytes, SheetHandle } from '/pkg/faunosteo_wasm.js'
const BYOB_BUFFER_BYTES = 1024 * 1024
const cspViolations = []
let memory = null
self.addEventListener('securitypolicyviolation', (event) => {
cspViolations.push(`${event.effectiveDirective} ${event.blockedURI}`)
})
const linearBytes = () => memory?.buffer.byteLength ?? null
function post(message) {
self.postMessage({ v: 1, ...message, csp: cspViolations })
}
function describe(error) {
const text = `${error?.name ?? 'Error'}: ${error?.message ?? error}`
return text.split('\n')[0].slice(0, 200)
}
/** Plan §B.1.5 kinds: a trap, an allocation failure, anything else. */
function kindOf(error) {
const name = error?.name ?? ''
if (name === 'RuntimeError' || name === 'InternalError') return 'trap'
if (name !== 'RangeError') return 'other'
return /memory|alloc|buffer/i.test(error.message) ? 'alloc' : 'trap'
}
function fail(phase, kind, error) {
post({ type: 'failed', phase, kind, error: describe(error) })
}
/** Instantiates once; `module` when the clone arrived, else the URL. */
async function start({ module, url }) {
const exports = await init({ module_or_path: module ?? url })
memory = exports.memory
post({
type: 'ready',
via: module ? 'module' : 'url',
linearBytes: linearBytes(),
})
}
/** A refusal decided here, in the envelope's shape. */
const refusal = (kind, detail) => JSON.stringify({ ok: false, kind, detail })
/** The size pre-check, before any read: the kinds the reader would give. */
function precheck(route, file, limits) {
const limit = route === 'copy' ? limits.copyBytes : limits.pdfBytes
if (file.size <= limit) return null
const kind = route === 'copy' ? 'invalid' : 'invalid_pdf'
return refusal(kind, `file size ${file.size} > ${limit}, not read`)
}
/** R4.2e: `import_pdf`, or the strict UTF-8 decode then `draft.restore`. */
function runImport(handle, route, bytes) {
if (route === 'pdf') return handle.import_pdf(new Uint8Array(bytes))
let text
try {
text = new TextDecoder('utf-8', { fatal: true }).decode(bytes)
} catch {
return refusal('invalid_utf8', 'strict TextDecoder refused the copy')
}
const input = JSON.stringify({ draft: text, archive: null })
return handle.call('draft.restore', input)
}
/** R4.3a: the sink consumed by the import (the wrapper is invalidated). */
function runSink(handle, route, sink) {
if (route === 'pdf') return handle.import_pdf_bytes(sink)
return handle.restore_copy_bytes(sink)
}
/** The R4.2e read: one ArrayBuffer, alive until the import returns. */
async function readWhole(file) {
let buffer
try {
buffer = await file.arrayBuffer()
} catch (error) {
return { kind: 'read', error }
}
const size = buffer.byteLength
return {
facts: { readVia: 'arrayBuffer', chunks: 1, maxChunkBytes: size },
run: (handle, route) => runImport(handle, route, buffer),
release: () => undefined,
}
}
/** A read's result, or `{ error }` when the browser cannot read the file. */
const settleRead = (pending) => pending.catch((error) => ({ error }))
const ignore = () => undefined
/** BYOB when the stream is a byte stream, else its default reader. */
function byobReader(stream) {
try {
return stream.getReader({ mode: 'byob' })
} catch {
return null
}
}
/**
* The stream's chunks: BYOB reads into one buffer, recycled after each push
* (its bytes were copied into the sink); else the stream's own chunks.
*/
function chunkSource(stream) {
const byob = byobReader(stream)
if (!byob) {
const reader = stream.getReader()
return {
via: 'stream-default',
next: () => settleRead(reader.read()),
recycle: ignore,
cancel: () => reader.cancel().catch(ignore),
}
}
let buffer = new ArrayBuffer(BYOB_BUFFER_BYTES)
return {
via: 'stream-byob',
next: () => settleRead(byob.read(new Uint8Array(buffer))),
recycle: (view) => (buffer = view.buffer),
cancel: () => byob.cancel().catch(ignore),
}
}
/** Pushes one chunk; `{ kind, error }` when refused or when it throws. */
function pushChunk(sink, chunk, facts) {
facts.chunks += 1
facts.maxChunkBytes = Math.max(facts.maxChunkBytes, chunk.byteLength)
try {
if (sink.push(chunk)) return null
} catch (error) {
return { kind: kindOf(error), error }
}
return { kind: 'read', error: new Error('more bytes than file.size') }
}
/**
* `file.stream()` into the sink, chunk by chunk (no chunk list, no
* concatenation), so the whole file exists only in the sink.
*/
async function streamInto(file, sink, facts) {
let source
try {
source = chunkSource(file.stream())
} catch (error) {
return { kind: 'read', error }
}
facts.readVia = source.via
for (;;) {
let next = await source.next()
if (next.error) return { kind: 'read', error: next.error }
if (next.done) return null
const refused = pushChunk(sink, next.value, facts)
source.recycle(next.value)
next = null // the chunk is unreachable before the next read suspends
if (refused) {
source.cancel()
return refused
}
}
}
/** No `file.stream()`: the file read whole, pushed once, then dropped. */
async function bufferInto(file, sink, facts) {
facts.readVia = 'arrayBuffer-fallback'
let buffer
try {
buffer = await file.arrayBuffer()
} catch (error) {
return { kind: 'read', error }
}
return pushChunk(sink, new Uint8Array(buffer), facts)
}
/** The file gathered in a `FileBytes` sink sized from `file.size`. */
async function readStreamed(file) {
let sink
try {
sink = new FileBytes(file.size)
} catch (error) {
return { kind: kindOf(error), error }
}
const facts = { readVia: 'stream', chunks: 0, maxChunkBytes: 0 }
const fill = typeof file.stream === 'function' ? streamInto : bufferInto
const failure = (await fill(file, sink, facts)) ?? shortRead(file, sink)
if (failure) {
sink.free()
return failure
}
return {
facts,
run: (handle, route) => runSink(handle, route, sink),
release: () => sink.free(),
}
}
/** A read that ended before `file.size`: a read failure, never a PDF. */
function shortRead(file, sink) {
if (sink.len() === file.size) return null
const text = `read ${sink.len()} of ${file.size} bytes`
return { kind: 'read', error: new Error(text) }
}
/** One step whose throwable becomes a `failed` message of its own kind. */
async function step(kind, work) {
try {
return { ok: true, value: await work() }
} catch (error) {
fail('task', kind ?? kindOf(error), error)
return { ok: false }
}
}
async function task({ route, file, sheet, limits, readPath = 'stream' }) {
const startedAt = performance.now()
const refused = precheck(route, file, limits)
if (refused) {
const diag = { precheck: true, readPath }
return post({ type: 'result', out: refused, diag })
}
const read = readPath === 'arrayBuffer' ? readWhole : readStreamed
const bytes = await read(file)
if (bytes.error) return fail('task', bytes.kind, bytes.error)
const readAt = performance.now()
const handle = await step('sheet', () => new SheetHandle(sheet))
if (!handle.ok) return bytes.release()
const sheetAt = performance.now()
const linearBeforeRun = linearBytes()
const out = await step(null, () => bytes.run(handle.value, route))
if (!out.ok) return
const doneAt = performance.now()
const diag = {
ms: doneAt - startedAt,
readMs: readAt - startedAt,
sheetMs: sheetAt - readAt,
runMs: doneAt - sheetAt,
linearBeforeRun,
linearBytes: linearBytes(),
readPath,
...bytes.facts,
}
post({ type: 'result', out: out.value, diag })
}
self.onmessage = (event) => {
const message = event.data
const phase = message?.type === 'init' ? 'start' : 'task'
const work = phase === 'start' ? start(message) : task(message)
work.catch((error) => fail(phase, kindOf(error), error))
}
self.onmessageerror = () => post({ type: 'clone_failed' })
self.onunhandledrejection = (event) =>
fail('task', kindOf(event.reason), event.reason)