dsh-inner-browser-use 0.44.0 → 0.45.0

package/scripts/check-sites.mjs

@@ -0,0 +1,94 @@
+#!/usr/bin/env node
+/**
+ * What the second read of the same site costs.
+ *
+ *   node scripts/check-sites.mjs
+ *
+ * Serves two pages on two origins: an app shell with three hundred buttons and
+ * a document of twelve thousand characters. Reads each one three times and
+ * checks that the first read teaches the next — the app opens as a map, the
+ * document is asked for less — and that the lesson is on disk for the next run.
+ * Exits non-zero when a site stops being learned.
+ */
+import { createServer } from 'node:http'
+import { existsSync, mkdtempSync, readFileSync } from 'node:fs'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+
+process.env.DSH_INNER_BROWSER_STATE_DIR = mkdtempSync(join(tmpdir(), 'dsh-sites-'))
+
+const buttons = Array.from({ length: 300 }, (_, i) => `<button>action ${i + 1}</button>`).join('')
+const app = `<!doctype html><meta charset="utf-8"><title>app shell</title>
+<h1>app shell</h1><p>a shell with three hundred things to click.</p><div>${buttons}</div>`
+const words = Array.from({ length: 240 }, (_, i) => `Paragraph ${i + 1}: ` + 'the quick brown fox jumps over the lazy dog and keeps running through the field '.repeat(3)).join('</p><p>')
+const doc = `<!doctype html><meta charset="utf-8"><title>long document</title>
+<article><h1>long document</h1><p>${words}</p></article>`
+
+const server = createServer((request, response) => {
+  const body = String(request.url ?? '').startsWith('/doc') ? doc : app
+  response.writeHead(200, { 'content-type': 'text/html; charset=utf-8' })
+  response.end(body)
+})
+await new Promise((done) => server.listen(0, '127.0.0.1', done))
+const port = server.address().port
+const appUrl = `http://127.0.0.1:${port}/app`
+const docUrl = `http://localhost:${port}/doc`
+
+const { Sidecar } = await import('../sidecar.js')
+const { saveTactics, sidecarTools, sessionScope, workScope } = await import('../tools.js')
+
+const wait = (ms) => new Promise((settle) => setTimeout(settle, ms))
+const sidecar = new Sidecar({ profile: 'sites-check', headless: true })
+const tools = sidecarTools(sidecar, { byChat: new Map() }, {})
+const tool = (name) => tools.find((one) => one.name === name)
+const cwd = mkdtempSync(join(tmpdir(), 'dsh-sites-cwd-'))
+/* Same binding the plugin does: the chat and the project a call belongs to. */
+const run = (name, args) => sessionScope.run('sites-check', () => workScope.run(cwd, () => tool(name).execute(args, { agent: { id: 'sites-check' } })))
+let failed = 0
+const check = (ok, what) => {
+  console.log(`${ok ? 'ok  ' : 'FAIL'} ${what}`)
+  if (!ok) failed += 1
+}
+const text = (said) => String(said?.text ?? '')
+try {
+  await sidecar.launch('about:blank')
+  await sidecar.watch(() => {})
+  await sidecar.ensure({ session: 'sites-check' })
+
+  await run('browser_open', { url: appUrl })
+  const appFirst = text(await run('browser_view', {}))
+  const appSecond = text(await run('browser_view', {}))
+  const appThird = text(await run('browser_view', {}))
+  console.log('--- app, third read ---\n' + appThird.slice(0, 1200))
+  check(!/^map: /m.test(appFirst), 'the first read of a big site is the page itself')
+  check(!/^map: /m.test(appSecond), 'one read is not yet a lesson')
+  check(/^map: \d+ controls/m.test(appThird), 'the third read opens as a map of what is there')
+  check(/this site reads big \(2 read\(s\)/.test(appThird), 'the map says which site it learned and why')
+  check(appThird.length < appFirst.length, `the map costs less than the page (${String(appThird.length)} < ${String(appFirst.length)} chars)`)
+
+  await run('browser_open', { url: docUrl })
+  const docFirst = text(await run('browser_read', {}))
+  const docSecond = text(await run('browser_read', { force: true }))
+  const docThird = text(await run('browser_read', {}))
+  console.log('--- doc, third read (head) ---\n' + docThird.slice(0, 400))
+  console.log('--- doc, third read (tail) ---\n' + docThird.slice(-300))
+  check(docFirst.length > 4000, `a long document really is long (${String(docFirst.length)} chars)`)
+  check(docSecond.length > 4000, 'the forced second read is the whole thing again')
+  check(/this site reads long \(2 read\(s\), ~\d{4,} chars of text\): cut at 3500 characters/.test(docThird), 'the third read says it was cut short and how to go past it')
+  check(docThird.length < docFirst.length, `the learned read costs less (${String(docThird.length)} < ${String(docFirst.length)} chars)`)
+
+  await saveTactics()
+  const file = join(process.env.DSH_INNER_BROWSER_STATE_DIR, 'sites.json')
+  check(existsSync(file), 'the lesson is written where the next run will find it')
+  const kept = existsSync(file) ? JSON.parse(readFileSync(file, 'utf8')) : {}
+  check(kept[`http://127.0.0.1:${port}`]?.controls >= 60, 'the app origin remembers how many controls it has')
+  check(kept[`http://localhost:${port}`]?.text >= 5000, 'the document origin remembers how long its text is')
+} catch (error) {
+  failed += 1
+  console.log('FAIL ' + String(error?.stack ?? error))
+} finally {
+  await sidecar.close?.().catch(() => {})
+  server.close()
+}
+console.log(failed === 0 ? 'sites: ok' : `sites: ${String(failed)} failed`)
+process.exitCode = failed === 0 ? 0 : 1

package/scripts/soak.mjs

@@ -0,0 +1,110 @@
+#!/usr/bin/env node
+/**
+ * Does a long session grow without bound?
+ *
+ *   node scripts/soak.mjs [cycles]
+ *
+ * One headless browser, one tab, one chat, over and over: open a page, read it,
+ * take a snapshot, read the same page again (the unchanged short-circuit), open
+ * the other page. Memory and page count are sampled every few cycles. The run
+ * fails when RSS climbs past the ceiling or a second tab appears, so a leak in
+ * the screencast, the CDP client or the read cache shows up as a number here
+ * instead of as a slow machine next week.
+ *
+ * Bounded on purpose: 24 cycles by default, about twenty seconds, no internet.
+ */
+import { createServer } from 'node:http'
+import { mkdtempSync } from 'node:fs'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+
+process.env.DSH_INNER_BROWSER_STATE_DIR = mkdtempSync(join(tmpdir(), 'dsh-soak-'))
+const cycles = Math.max(4, Math.min(Number(process.argv[2] ?? 24) || 24, 200))
+
+const buttons = Array.from({ length: 120 }, (_, i) => `<button>action ${i + 1}</button>`).join('')
+const words = Array.from({ length: 80 }, (_, i) => `Paragraph ${i + 1}: ` + 'the quick brown fox jumps over the lazy dog. '.repeat(6)).join('</p><p>')
+const pages = {
+  '/app': `<!doctype html><meta charset="utf-8"><title>app shell</title><h1>app shell</h1><div>${buttons}</div>`,
+  '/doc': `<!doctype html><meta charset="utf-8"><title>long document</title><article><h1>long document</h1><p>${words}</p></article>`,
+}
+
+const server = createServer((request, response) => {
+  const body = pages[String(request.url ?? '').split('?')[0]] ?? pages['/app']
+  response.writeHead(200, { 'content-type': 'text/html; charset=utf-8' })
+  response.end(body)
+})
+await new Promise((done) => server.listen(0, '127.0.0.1', done))
+const port = server.address().port
+const appUrl = `http://127.0.0.1:${port}/app`
+const docUrl = `http://127.0.0.1:${port}/doc`
+
+const { Sidecar } = await import('../sidecar.js')
+const { sidecarTools, sessionScope, workScope } = await import('../tools.js')
+
+const sidecar = new Sidecar({ profile: 'soak', headless: true })
+const tools = sidecarTools(sidecar, { byChat: new Map() }, {})
+const tool = (name) => tools.find((one) => one.name === name)
+const cwd = mkdtempSync(join(tmpdir(), 'dsh-soak-cwd-'))
+/* Same binding the plugin does: the chat and the project a call belongs to. */
+const run = (name, args) => sessionScope.run('soak', () => workScope.run(cwd, () => tool(name).execute(args, { agent: { id: 'soak' } })))
+const text = (said) => String(said?.text ?? '')
+
+/* macOS reports every renderer's RSS in full, shared pages included, so a
+   healthy headless tree sits far above what one page needs; the bar is a
+   ceiling this machine never approaches plus growth over the run. */
+const ceilingMb = 2500
+const samples = []
+let failed = 0
+const check = (ok, what) => {
+  console.log(`${ok ? 'ok  ' : 'FAIL'} ${what}`)
+  if (!ok) failed += 1
+}
+
+const startedAt = Date.now()
+try {
+  await sidecar.launch('about:blank')
+  await sidecar.watch(() => {})
+  await sidecar.ensure({ session: 'soak' })
+
+  let reads = 0
+  let snapshots = 0
+  for (let i = 0; i < cycles; i += 1) {
+    const url = i % 2 === 0 ? appUrl : docUrl
+    await run('browser_open', { url })
+    reads += text(await run('browser_view', {})).length > 0 ? 1 : 0
+    await run('browser_view', {})
+    if (i % 3 === 0) {
+      snapshots += 1
+      await run('browser_shot', {})
+    }
+    if (i % 2 === 1) await run('browser_read', {})
+    if (i % 4 === 3 || i === cycles - 1) {
+      /* sidecar.state() already reports what the whole process tree holds, so
+         no pid has to be found and no second sampling path exists. */
+      const state = await sidecar.state()
+      const mb = Math.round(Number(state?.memory ?? 0) / 1e6)
+      samples.push({ cycle: i + 1, mb, pages: Number(state?.pages ?? 0) })
+      console.log(`     cycle ${String(i + 1).padStart(3)} · ${String(mb)} MB · pages ${String(state?.pages)}`)
+    }
+    /* A quiet moment, so a leak in the close/reopen path is not hidden by the
+       next navigation arriving before the old page is gone. */
+    if (i % 6 === 5) await new Promise((settle) => setTimeout(settle, 250))
+  }
+
+  const first = samples[0] ?? { mb: 0 }
+  const last = samples.at(-1) ?? first
+  const growth = last.mb - first.mb
+  const seconds = Math.round((Date.now() - startedAt) / 100) / 10
+  check(reads > 0, `the browser answered every cycle (${String(reads)} of ${String(cycles)})`)
+  check(samples.every((one) => one.pages <= 1), `one page per chat is kept (max ${String(Math.max(...samples.map((one) => one.pages)))})`)
+  check(last.mb < ceilingMb, `RSS stays inside this machine's ceiling (${String(last.mb)} < ${String(ceilingMb)} MB)`)
+  check(growth < 250, `RSS growth over ${String(cycles)} cycles is bounded (${growth >= 0 ? '+' : ''}${String(growth)} MB)`)
+  console.log(`soak: ${failed === 0 ? 'ok' : `${String(failed)} failed`} · ${String(cycles)} cycles, ${String(seconds)} s, rss ${String(first.mb)} → ${String(last.mb)} MB, pages ${String(last.pages)}, ${String(snapshots)} shots, ${String(reads)} reads`)
+} catch (error) {
+  failed += 1
+  console.log('FAIL ' + String(error?.stack ?? error))
+} finally {
+  await sidecar.close?.().catch(() => {})
+  server.close()
+}
+process.exitCode = failed === 0 ? 0 : 1

package/sidecar.js

@@ -175,6 +175,19 @@ export function flashRuntime() {
   return { binary: ready.electron, plugin: ready.plugin, version }
 }
 
+/**
+ * What the browser said about its own plugin list. A plugin runtime that opens
+ * its port but carries no PPAPI module looks fine from outside and plays
+ * nothing, so the list is read and turned into a sentence rather than trusted.
+ * @param {unknown} names - plugin names from the page.
+ * @returns {{ok: boolean, name: string}} whether a Flash plugin is there.
+ */
+export function flashVerdict(names) {
+  const list = (Array.isArray(names) ? names : []).map((one) => String(one ?? ''))
+  const found = list.find((one) => one.toLowerCase().includes('shockwave'))
+  return { ok: found !== undefined, name: found ?? '' }
+}
+
 /* A chat's tab address outlives a plugin reload: the browser is still running
    with that tab, and the chat that returns should get its page back, not a
    blank one. */
@@ -2118,6 +2131,9 @@ export class Sidecar {
     /* A movie page needs a runtime Chrome no longer has; the switch is the
        person's, and it swaps the browser rather than the protocol. */
     this.flash = options.flash === undefined ? process.env.DSH_INNER_BROWSER_FLASH === '1' || readFlashChoice() : options.flash === true
+    /* Whether the launched runtime really carries the plugin, asked once at
+       launch: null means it has not been asked. */
+    this.flashCheck = null
     /* The Flash app serves its own frames; the screencast cannot see a plugin. */
     this.flashBridge = null
     this.headless = true
@@ -2634,6 +2650,29 @@ export class Sidecar {
     return this.starting
   }
 
+  /**
+   * Ask the Flash-era browser which plugins it actually loaded.
+   * @returns {Promise<{ok: boolean, name: string} | null>} the answer, or null when it could not be asked.
+   */
+  async checkFlash() {
+    this.flashCheck = null
+    const page = this.page
+    if (page === null || page === undefined) return null
+    try {
+      const out = await page.cdp.send('Runtime.evaluate', {
+        expression: 'JSON.stringify(Array.from(navigator.plugins).map((one) => one.name))',
+        returnByValue: true,
+      })
+      const said = out?.result?.value
+      const names = JSON.parse(typeof said === 'string' ? said : '[]')
+      this.flashCheck = flashVerdict(names)
+    } catch {
+      /* A page that cannot answer is not evidence either way. */
+      this.flashCheck = null
+    }
+    return this.flashCheck
+  }
+
   /** @param {string | undefined} url - first address. @returns {Promise<SidecarPage>} the page. */
   async launch(url) {
     /* Chromium 87 opening the profile a current Chrome owns would downgrade it
@@ -2656,6 +2695,10 @@ export class Sidecar {
     await this.restoreSessionCookies()
     await this.page.cast()
     if (url !== undefined && url !== '') await this.page.navigate(url)
+    /* A runtime that opens a port is not yet a runtime that plays: ask the page
+       whether the plugin is really in it, so a broken install is said out loud
+       instead of showing up as a black movie. */
+    if (this.flash) await this.checkFlash()
     this.touch()
     /* Sign-ins outlive this browser: what works now is worth a copy, taken
        while the cookies are known good rather than after they are lost. */
@@ -3326,6 +3369,8 @@ export class Sidecar {
       alive: this.alive, // ui-copy-ignore
       profile: this.profile, // ui-copy-ignore
       headless: this.headless, // ui-copy-ignore
+      flash: this.flash, // ui-copy-ignore
+      flashCheck: this.flashCheck, // ui-copy-ignore
       url: this.page?.url ?? '', // ui-copy-ignore
       pages: this.pages.size,
       title: this.page?.title ?? '',