@preventive/triage 1.0.0-alpha.5 → 1.0.0-alpha.50

package/risk/actions/scripts/shell.js

@@ -0,0 +1,169 @@
+// A `run:` script read into the commands it runs, as a shell would split
+// them, for commands.js. A quick read, not a shell's parse: a command built
+// as the script runs goes unseen.
+import { SHELLS } from '../../files/runs-code.js'
+import { resolved } from '../../npm/files.js'
+
+// A shell, by its program, wherever it is, Windows' `.exe` too (risk's
+// runs-code.js SHELLS).
+const SHELL = new RegExp(String.raw`^(?:.*[\/\\])?${SHELLS}(?:\.exe)?$`, 'iu')
+
+// A line of a script, its continuations joined (an escaped line's end, or a
+// pipe, `&&` or `||` the next line goes on from), and the line it starts
+// on, counting from 0.
+export function scriptLines(script) {
+  // An expression (`${{ … }}`), which GitHub writes in before the shell reads
+  // the script, read as a variable is: a value unknown here, each line it
+  // spans past its first continuing the one it is on; a `}}` in one of its
+  // strings (`format('{{0}}')`) not its end.
+  const lines = script.replaceAll(/\$\{\{(?:'[^']*'|[^'])*?\}\}/gu, expression => `$EXPRESSION${' \\\n'.repeat(expression.split('\n').length - 1)}`).split('\n')
+  const joined = []
+  for (let at = 0; at < lines.length; at++) {
+    const start = at
+    let text = lines[at].replace(/\r$/u, '')
+    for (;;) {
+      const escaped = /(?:^|[^\\])(?:\\\\)*[\\`^]$/u.test(text), piped = /(?:\||&&)\s*$/u.test(text) && !/(?:^|\s)#/u.test(text)
+      if (!(escaped || piped) || at + 1 === lines.length) break
+      text = `${escaped ? text.slice(0, -1) : text} ${lines[++at].replace(/\r$/u, '')}`
+    }
+    joined.push({ text, line: start })
+  }
+  return joined
+}
+
+const WRAPPERS = new Set(['sudo', 'env', 'time', 'nohup', 'exec', 'command', 'nice', 'stdbuf', 'xvfb-run', 'timeout', 'corepack'])
+// A wrapper's options taking a value, as `sudo -u user`, `timeout 10m` do.
+const WRAPPER_VALUES = new Map([['sudo', /^-[ugCDhpRrTU]$/u], ['env', /^-[uC]$/u], ['nice', /^-n$/u], ['timeout', /^-[sk]$/u]])
+
+// What a command runs (`argv`): its words past settings, kept (`env`), and
+// wrappers, a wrapper's options and their values, and `timeout`'s duration.
+function argvOf(words) {
+  const env = {}
+  let at = 0
+  for (;;) {
+    for (; at < words.length && /^[A-Za-z_]\w*=/u.test(words[at]); at++) env[words[at].slice(0, words[at].indexOf('='))] = words[at].slice(words[at].indexOf('=') + 1)
+    const wrapper = words[at]?.replace(/^.*\//u, '')
+    if (!WRAPPERS.has(wrapper)) return { argv: words.slice(at), env }
+    at++
+    while (at < words.length && words[at].startsWith('-')) at += WRAPPER_VALUES.get(wrapper)?.test(words[at]) ? 2 : 1
+    if (wrapper === 'timeout' && /^\d/u.test(words[at] ?? '')) at++
+  }
+}
+
+// A command's words, quotes taken off, redirections dropped, and the file
+// its output is written to (`> file`, `>> file`), or null.
+function wordsOf(text) {
+  const words = []
+  let quote = null, word = null
+  for (let at = 0; at <= text.length; at++) {
+    const char = text[at]
+    if (quote) {
+      if (char === quote) quote = null
+      else if (char === undefined) break
+      else word += quote === '"' && char === '\\' && /["\\$`]/u.test(text[at + 1] ?? '') ? text[++at] : char
+    } else if (char === undefined || /\s/u.test(char)) {
+      if (word !== null) words.push(word)
+      word = null
+    } else if (char === '"' || char === "'") { quote = char; word ??= '' }
+    // A redirection joined to a word before it (`url>file`) starts its own.
+    else if ((char === '>' || char === '<') && word !== null && !/^\d*&?$|[<>]$/u.test(word)) { words.push(word); word = char }
+    else word = (word ?? '') + (char === '\\' && at + 1 < text.length ? text[++at] : char)
+  }
+  const kept = []
+  let out = null
+  for (let at = 0; at < words.length; at++) {
+    const redirect = /^(\d*)(>>?|<<?|&>>?)(.*)$/u.exec(words[at])
+    if (!redirect) { kept.push(words[at]); continue }
+    const target = redirect[3] === '' ? words[++at] : redirect[3]
+    if (redirect[2].includes('>') && (redirect[1] === '' || redirect[1] === '1') && target && !target.startsWith('&')) out = target
+  }
+  return { words: kept, out }
+}
+
+// Where a `$(…)` starting at `from` (its `(`) closes, or the line's end.
+function closing(text, from) {
+  let depth = 0, quote = null
+  for (let at = from; at < text.length; at++) {
+    const char = text[at]
+    if (quote) { if (char === quote) quote = null; else if (char === '\\') at++ }
+    else if (char === "'" || char === '"') quote = char
+    else if (char === '(') depth++
+    else if (char === ')' && --depth === 0) return at
+  }
+  return text.length
+}
+
+// A line's commands, each `{ argv, env, piped, out, nested }`, `env` the
+// settings it runs with, `piped` where a pipe feeds it, `out` the file it writes its output to and `nested` its
+// command substitutions' text, to read as lines of their own; and those of
+// none of them (`nested`).
+function commandsOf(text) {
+  const commands = []
+  let nested = [], piped = false, quote = null, start = 0
+  const end = (at, next) => {
+    const { words, out } = wordsOf(text.slice(start, at))
+    const { argv, env } = argvOf(words)
+    if (argv.length > 0) { commands.push({ argv, env, piped, out, nested }); nested = [] }
+    piped = next
+  }
+  for (let at = 0; at < text.length; at++) {
+    const char = text[at]
+    if (quote === "'") { if (char === "'") quote = null; continue }
+    if (char === '\\') { at++; continue }
+    if (char === '$' && text[at + 1] === '(') {
+      const close = closing(text, at + 1)
+      nested.push(text.slice(at + 2, close))
+      at = close
+    } else if (quote === '"') { if (char === '"') quote = null }
+    else if (char === '"' || char === "'") quote = char
+    else if (char === '#' && (at === 0 || /\s/u.test(text[at - 1]))) { end(at, false); start = text.length; break }
+    else if ((char === '&' || char === '|') && text[at + 1] === char) { end(at, false); start = ++at + 1 }
+    else if (char === ';' || (char === '&' && !/[<>]/u.test(text[at - 1] ?? '') && text[at + 1] !== '>')) { end(at, false); start = at + 1 }
+    else if (char === '|') { end(at, true); start = at + (text[at + 1] === '&' ? 2 : 1); if (text[at + 1] === '&') at++ }
+  }
+  if (start < text.length) end(text.length, false)
+  return { commands, nested }
+}
+
+// The command line a shell is given to run (`bash -euc '…'`, `pwsh
+// -Command '…'`), else null.
+function shellLine([program, ...words]) {
+  if (!SHELL.test(program)) return null
+  const option = /(?:pwsh|powershell)(?:\.exe)?$/iu.test(program) ? /^-(?:c|command)$/iu : /^-[a-z]*c[a-z]*$/u
+  const at = words.findIndex(word => option.test(word))
+  return at === -1 ? null : words[at + 1] ?? null
+}
+
+// A path `to` from `dir`, both from the step's directory ('' for it); as
+// written where a variable, `~`, `-` or an absolute path names it.
+export const pathFrom = (dir, to) => /^[$~/-]/u.test(to) ? to : resolved(dir ? dir.split('/') : [], to)?.join('/') ?? [dir, to].filter(Boolean).join('/')
+
+// Each command of a script, in the order they run (`order`): its `line` from
+// 0, the directory the `cd`s before it moved to (`dir`, by pathFrom), and
+// `pipedTo`, the commands its output is piped through; a command
+// substitution's before its command's, a shell's command line's (`sh -c`)
+// after it.
+export function scriptCommands(script) {
+  const found = []
+  let dir = '', read = 0
+  // A substitution's or a shell's command line (`apart`) moves no `cd` of its caller's.
+  const readLine = (text, line, apart = false) => {
+    if (read++ >= 64) return
+    const outer = dir, { commands, nested } = commandsOf(text)
+    commands.forEach(({ nested: substituted, ...command }, at) => {
+      for (const each of substituted) readLine(each, line, true)
+      const end = commands.findIndex((each, after) => after > at && !each.piped)
+      found.push({ ...command, line, dir, order: found.length, pipedTo: commands.slice(at + 1, end === -1 ? undefined : end) })
+      if (command.argv[0] === 'cd') dir = pathFrom(dir, command.argv.slice(1).find(word => !/^-[LPe@]+$/u.test(word)) ?? '~')
+      const shell = shellLine(command.argv)
+      if (shell) readLine(shell, line, true)
+    })
+    for (const each of nested) readLine(each, line, true)
+    if (apart) dir = outer
+  }
+  for (const { text, line } of scriptLines(script)) {
+    read = 0
+    readLine(text, line)
+  }
+  return found
+}

package/risk/actions/uses.js

@@ -0,0 +1,76 @@
+// What a `uses:` names, as GitHub reads it (`owner/repo[/path]@ref`, `./path`,
+// `$/path`, `docker://image`), and an image a job runs in.
+
+const GITHUB_OWNERS = new Set(['actions', 'github'])
+
+const OWNER = /^[\da-z](?:[\da-z]|-(?=[\da-z])){0,38}$/iu
+const NAME = /^(?!\.{1,2}$)[\w.-]{1,100}$/u
+const SHA = /^[\da-f]{40}$/u
+const SHORT_SHA = /^[\da-f]{7,39}$/u
+// A version a comment beside a pin names, as Dependabot, Renovate, pinact
+// and ratchet write one (`# v4.1.1`, `# tag=v4.1.1`, `# ratchet:x@v4`).
+const VERSION_COMMENT = /^#\s*(?:tag\s*[=:]\s*|pin@|ratchet:[^\s@]+@)?(v?\d+(?:\.\d+){0,3}(?:[-+][\w.-]+)?)(?![\w.])/iu
+
+// `sha`, `short-sha`, `ref` (a tag or a branch), or `expression`, which GitHub refuses.
+function pinOf(ref) {
+  if (/\$\{\{/u.test(ref)) return 'expression'
+  if (SHA.test(ref)) return 'sha'
+  return SHORT_SHA.test(ref) ? 'short-sha' : 'ref'
+}
+
+// An image's name, tag and digest (`[host[:port]/]path[:tag][@digest]`), null
+// for none; null where it isn't one.
+export function readImage(text) {
+  if (typeof text !== 'string' || text === '' || /\$\{\{|\s/u.test(text)) return null
+  const at = text.indexOf('@')
+  const digest = at === -1 ? null : text.slice(at + 1)
+  const named = at === -1 ? text : text.slice(0, at)
+  const colon = named.lastIndexOf(':')
+  const tagged = colon > named.lastIndexOf('/')
+  if (digest !== null && !/^sha256:[\da-f]{64}$/u.test(digest)) return null
+  return { image: tagged ? named.slice(0, colon) : named, tag: tagged ? named.slice(colon + 1) : null, digest }
+}
+
+const commentVersion = comment => typeof comment === 'string' ? VERSION_COMMENT.exec(comment.trim())?.[1] ?? null : null
+
+export const isWorkflowPath = path => /^\.github\/workflows\/[^/]+\.ya?ml$/u.test(path)
+
+// `{ kind, uses, … }`: `action`, or `workflow` at a job's level (`job`), with
+// its `repo`, `path`, `ref`, `pin` and `comment`'s version; `local`, `self`
+// at the commit running, `job` a reusable workflow's, at the calling
+// workflow's commit; `docker`; or `invalid`, which GitHub won't run.
+export function readUses(text, { job = false, comment = null } = {}) {
+  const uses = typeof text === 'string' ? text.trim() : ''
+  if (uses.startsWith('./') || uses.startsWith('$/')) {
+    return { kind: 'local', uses, path: uses.slice(2).replace(/\/+$/u, ''), ...uses.startsWith('$/') ? { self: true } : {}, ...job ? { job: true } : {} }
+  }
+  if (uses.startsWith('docker://')) {
+    const image = job ? null : readImage(uses.slice('docker://'.length))
+    return image ? { kind: 'docker', uses, ...image } : { kind: 'invalid', uses }
+  }
+  const at = uses.indexOf('@')
+  const [owner = '', name = '', ...path] = (at === -1 ? uses : uses.slice(0, at)).split('/')
+  const ref = at === -1 ? '' : uses.slice(at + 1)
+  if (!OWNER.test(owner) || !NAME.test(name) || ref === '' || path.some(part => !NAME.test(part)) || (job && !isWorkflowPath(path.join('/')))) {
+    return { kind: 'invalid', uses }
+  }
+  return { kind: job ? 'workflow' : 'action', uses, repo: `${owner}/${name}`, path: path.join('/'), ref, pin: pinOf(ref), comment: commentVersion(comment) }
+}
+
+// The key a reference is resolved once by, a commit's with its comment's version.
+export const usesKey = ({ repo, path, ref, pin, comment }) => `${repo.toLowerCase()}${path ? `/${path}` : ''}@${ref}${pin === 'sha' && comment ? ` # ${comment}` : ''}`
+
+// A reference from its key, a reusable workflow's by its path's.
+export function usesOfKey(key) {
+  const m = typeof key === 'string' ? /^([^@\s]+@\S+)(?: # (\S+))?$/u.exec(key) : null
+  const job = Boolean(m && /^[^/]+\/[^/]+\/\.github\/workflows\/[^/@]+\.ya?ml@/u.test(m[1]))
+  const read = m ? readUses(m[1], { job, comment: m[2] ? `# ${m[2]}` : null }) : null
+  return read && (read.kind === 'action' || read.kind === 'workflow') && usesKey(read) === key ? read : null
+}
+
+// GitHub's own, the workflow's owner's, or another's.
+export function ownerKind(repo, owner) {
+  const of = repo.split('/')[0].toLowerCase()
+  if (GITHUB_OWNERS.has(of)) return 'github'
+  return typeof owner === 'string' && of === owner.toLowerCase() ? 'own' : 'third-party'
+}

package/risk/actions/workflow.js

@@ -0,0 +1,110 @@
+// A workflow, or an action's metadata, read for what it runs: its events, its
+// jobs' steps (`uses`, `run`) and images, each on its line from 1.
+import { YamlError, parseYaml } from '@preventive/lockfile/yaml.js'
+import { readImage, readUses } from './uses.js'
+
+const isMap = value => value !== null && typeof value === 'object' && !Array.isArray(value)
+const stringOf = value => typeof value === 'string' ? value : null
+
+// The line, from 1, `key` of a mapping or an entry of a sequence is on.
+function lineOf(file, container, key, fallback) {
+  const at = container !== null && typeof container === 'object' ? file.positions.get(container) : undefined
+  const line = at instanceof Map ? at.get(key) : at?.[key]
+  return line === undefined ? fallback : line + 1
+}
+
+// A step's or a job's `uses`, on its line, with the comment that line ends with.
+function usesOf(file, owner, fallback, job = false) {
+  if (typeof owner.uses !== 'string') return null
+  const line = lineOf(file, owner, 'uses', fallback)
+  return { ...readUses(owner.uses, { job, comment: /(?:^|[ \t])(#.*)$/u.exec(file.lines[line - 1] ?? '')?.[1] ?? null }), line }
+}
+
+function triggersOf(on) {
+  if (typeof on === 'string') return [on]
+  if (Array.isArray(on)) return on.filter(event => typeof event === 'string')
+  return isMap(on) ? Object.keys(on) : []
+}
+
+// The line a `|` block scalar's text starts on, after its key's; else null.
+const LITERAL_HEADER = /:[ \t]+(?:&\S+[ \t]+)?\|[1-9+-]{0,2}[ \t]*(?:#.*)?$/u
+const literalStart = (file, line) => line !== null && LITERAL_HEADER.test(file.lines[line - 1] ?? '') ? line + 1 : null
+
+// A `run:` step's script, and its directory, else `directory` (defaults.run);
+// `production` where its `env`, else the job's or the workflow's (`env`),
+// sets NODE_ENV so, as an install then leaves devDependencies out.
+const envOf = (owner, outer = {}) => ({ ...outer, ...isMap(owner?.env) ? owner.env : {} })
+function runOf(file, step, directory, env) {
+  const script = stringOf(step.run)
+  if (script === null) return null
+  const line = lineOf(file, step, 'run', null)
+  return { script, line, start: literalStart(file, line), workingDirectory: stringOf(step['working-directory']) ?? directory,
+    production: envOf(step, env).NODE_ENV === 'production' }
+}
+
+// A step's or a job's `with`, and the line each input is on.
+function inputsOf(file, owner, line) {
+  const inputs = isMap(owner.with) ? owner.with : null
+  return { with: inputs, withLines: inputs ? Object.fromEntries(Object.keys(inputs).map(key => [key, lineOf(file, inputs, key, line)])) : {} }
+}
+
+function stepsOf(file, steps, directory, env = {}) {
+  if (!Array.isArray(steps)) return []
+  return steps.flatMap((step, index) => {
+    if (!isMap(step)) return []
+    const line = lineOf(file, steps, index, null)
+    return [{ line, uses: usesOf(file, step, line), run: runOf(file, step, directory, env), ...inputsOf(file, step, line) }]
+  })
+}
+
+// The directory a workflow's or a job's `defaults.run` names, else `outer`.
+const directoryOf = (owner, outer) => stringOf(isMap(owner?.defaults) && isMap(owner.defaults.run) ? owner.defaults.run['working-directory'] : null) ?? outer
+
+// The images a job runs in: its container's and its services'.
+function imagesOf(file, job) {
+  const images = []
+  const container = isMap(job.container) ? job.container.image : job.container
+  if (typeof container === 'string') images.push({ text: container, line: lineOf(file, isMap(job.container) ? job.container : job, isMap(job.container) ? 'image' : 'container', null) })
+  if (isMap(job.services)) {
+    for (const service of Object.values(job.services)) {
+      if (isMap(service) && typeof service.image === 'string') images.push({ text: service.image, line: lineOf(file, service, 'image', null) })
+    }
+  }
+  return images.map(image => ({ ...image, read: readImage(image.text) }))
+}
+
+function jobsOf(file, doc) {
+  if (!isMap(doc.jobs)) throw new YamlError('a workflow has no jobs')
+  const directory = directoryOf(doc, null)
+  return Object.values(doc.jobs).filter(isMap).map(job => ({ uses: usesOf(file, job, null, true), secrets: job.secrets ?? null, ...inputsOf(file, job, null),
+    steps: stepsOf(file, job.steps, directoryOf(job, directory), envOf(job, envOf(doc))), images: imagesOf(file, job) }))
+}
+
+// An action's `runs`: its runtime, a composite one's steps as a job, and a
+// Docker one's image or Dockerfile.
+function runsOf(file, doc) {
+  const runs = isMap(doc.runs) ? doc.runs : null
+  if (!runs || typeof runs.using !== 'string') throw new YamlError('an action has no runs.using')
+  const using = runs.using.toLowerCase()
+  const usingLine = lineOf(file, runs, 'using', null)
+  const image = using === 'docker' ? stringOf(runs.image) : null
+  const imageLine = lineOf(file, runs, 'image', usingLine)
+  const docker = image?.startsWith('docker://') ? { text: image.slice('docker://'.length), line: imageLine, read: readImage(image.slice('docker://'.length)) } : null
+  const steps = using === 'composite' ? stepsOf(file, runs.steps, null) : []
+  return { using, usingLine, dockerfile: image && !docker ? { path: image.replace(/^\.\//u, ''), line: imageLine } : null, jobs: [{ uses: null, steps, images: docker ? [docker] : [] }] }
+}
+
+// A workflow (`kind` 'workflow') or an action's metadata ('action') read, or
+// the YamlError of one GitHub can't run.
+export function readWorkflow(kind, text) {
+  try {
+    const file = { lines: text.split('\n'), positions: new WeakMap() }
+    const doc = parseYaml(text, { workflow: true, lines: file.positions })
+    if (!isMap(doc)) throw new YamlError(`${kind === 'action' ? 'an action' : 'a workflow'} is not a mapping`)
+    if (kind === 'action') return { triggers: [], ...runsOf(file, doc) }
+    return { triggers: triggersOf(doc.on), using: null, usingLine: null, dockerfile: null, jobs: jobsOf(file, doc) }
+  } catch (error) {
+    if (error instanceof YamlError) return error
+    throw error
+  }
+}

package/risk/files/characters.js

@@ -0,0 +1,196 @@
+// The characters a text holds that can make it read other than it runs, or
+// hide text in it: control and bidirectional characters, and the invisible
+// characters a payload or a name no one sees can be spelled in (variation
+// selectors, Hangul fillers, tags). Shared with a code viewer that shows each
+// as a labelled box (ui/view/control-chars.js), which leaves what spells an
+// emoji, a recommended flag or a line's end as it is (passed).
+
+// What a text holds beyond printing characters and its line breaks: the C0
+// controls other than tab, line feed and carriage return, DEL, the C1
+// controls, and the bidirectional controls that can make code read other than
+// it runs ("Trojan Source"); and the invisible characters that can hide text
+// in it (invisibleCharacters), as a payload hidden in
+// variation selectors or a variable named by a Hangul filler.
+export const CONTROLS = /(?![\t\n\r])\p{Cc}|[\u202A-\u202E\u2066-\u2069]/gu
+const NON_ASCII = /\P{ASCII}/u
+
+// A file's text: `kind` 'ascii' or 'utf8', by whether it holds anything past
+// ASCII, else 'binary' (which the server tells by its bytes, sending no
+// text); `controls` the control and invisible characters a text holds, how
+// often by code point, or null for none.
+export function npmTextEncoding(text) {
+  if (typeof text !== 'string') return { kind: 'binary', controls: null }
+  const controls = new Map()
+  for (const [char] of text.matchAll(CONTROLS)) controls.set(char.codePointAt(0), (controls.get(char.codePointAt(0)) ?? 0) + 1)
+  // The invisible characters are all past ASCII: looked for in a text that
+  // has any, from its first.
+  const past = text.search(NON_ASCII)
+  if (past !== -1) for (const [code, times] of invisibleCharacters(text, past) ?? []) controls.set(code, times)
+  return { kind: past === -1 ? 'ascii' : 'utf8', controls: controls.size > 0 ? controls : null }
+}
+
+// A pictograph is an assigned one (Emoji): Extended_Pictographic also takes
+// in code points reserved for emoji to come, unassigned, each boxed itself,
+// and a joiner or selector beside one would go unseen.
+const EMOJI_BEFORE = /(?=\p{Emoji})\p{Extended_Pictographic}(?:\u{FE0F}|[\u{1F3FB}-\u{1F3FF}])?$/u
+const EMOJI_AFTER = /^(?=\p{Emoji})\p{Extended_Pictographic}/u
+const PICTOGRAPH_BEFORE = /(?=\p{Emoji})\p{Extended_Pictographic}$/u
+const KEYCAP_BASE = /^[\d#*]$/u
+const isTag = code => code >= 0xe0020 && code <= 0xe007f
+// The subdivision flags Unicode recommends (RGI), their tags spelling the
+// subdivision after a black flag: England's, Scotland's and Wales's. Others
+// show as a black flag and their tags as nothing, as a payload in them would.
+const FLAGS = new Set(['gbeng', 'gbsct', 'gbwls'])
+const tagged = (text, from, to) => to - from === 10 ? String.fromCodePoint(...Array.from({ length: 5 }, (_, i) => text.codePointAt(from + 2 * i) - 0xe0000)) : ''
+
+// Where to go on from where `code` at `at` in `text` stays text, else -1 for
+// one to show. A CR before a newline is part of the line end, as is one that
+// ends `text` where `lineEnd` says a newline follows it. A joiner between two
+// emoji joins them into one, a recommended subdivision flag's tags, up to the
+// cancel tag, name its region (FLAGS), and the text or emoji presentation
+// selector after a pictograph, or the emoji one between a keycap's digit, #
+// or * and its mark, draws it; after a letter or a digit alone it is another
+// name.
+export function passed(text, at, code, lineEnd) {
+  if (code === 0x0d && (text[at + 1] === '\n' || lineEnd && at + 1 === text.length)) return at + 1
+  if (code === 0x200d && EMOJI_BEFORE.test(text.slice(Math.max(0, at - 4), at)) && EMOJI_AFTER.test(text.slice(at + 1, at + 3))) return at + 1
+  if ((code === 0xfe0e || code === 0xfe0f) && PICTOGRAPH_BEFORE.test(text.slice(Math.max(0, at - 2), at))) return at + 1
+  if (code === 0xfe0f && KEYCAP_BASE.test(text[at - 1] ?? '') && text[at + 1] === '\u20E3') return at + 1
+  if (isTag(code) && !isTag(text.codePointAt(at - 2))) {
+    let end = at
+    while (isTag(text.codePointAt(end))) end += 2
+    if (text.codePointAt(at - 2) === 0x1f3f4 && text.codePointAt(end - 2) === 0xe007f && FLAGS.has(tagged(text, at, end - 2))) return end
+  }
+  return -1
+}
+
+// Those text is hidden in, as a file's Invisible chars label counts them
+// (readability.js), where they show: variation selectors (a payload, a
+// byte a selector), Hangul fillers (a name no one sees) and tags (ASCII no one
+// sees). Not the format characters, a BOM or a zero-width space, which text
+// has its uses for, nor the other marks that show as nothing (the combining
+// grapheme joiner, the Khmer inherent vowels and the Mongolian free variation
+// selectors), which tables of the characters a name may hold, as JavaScript
+// parsers have, hold too; nor what of these hides nothing (benign).
+const INVISIBLE = /[\u115F\u1160\u3164\uFFA0\u{E0001}\u{E0020}-\u{E007F}]|[\uFE00-\uFE0F]|[\u{E0100}-\u{E01EF}]/u
+// The code units they are in, to look for them by, faster than by code point:
+// the high surrogate starting those past the Basic Multilingual Plane (tags,
+// the selectors' supplement) too, told apart from its others once found. No
+// `u`, which would read a pair as the one code point it spells.
+// eslint-disable-next-line require-unicode-regexp
+const INVISIBLE_UNITS = /[\uFE00-\uFE0F\u115F\u1160\u3164\uFFA0\uDB40]/g
+// A table's code points, from its ranges in hex (`30 2018-2019`).
+const codePoints = ranges => new Set(ranges.split(' ').flatMap(range => {
+  const [from, to = from] = range.split('-').map(digits => Number.parseInt(digits, 16))
+  return Array.from({ length: to - from + 1 }, (_, i) => from + i)
+}))
+// The variation sequences Unicode registers, which hide nothing, though Code
+// shows their selectors after a letter or a digit, another name there: VS1 to
+// VS14's on other than an ideograph (StandardizedVariants.txt, Unicode
+// 18.0.0), each selector's bases, as an HTML entity table's math symbols, and
+// the emoji with a text and an emoji presentation, VS15 and VS16 after each
+// (emoji-variation-sequences.txt), as a heading's anchor keeps one (`#`).
+const STANDARDIZED = new Map([
+  [0xfe00, '30 1000 1002 1004 1010-1011 1015 1019-101a 101c-101d 1022 1031 1075'
+    + ' 1078 107a 1080 2018-2019 201c-201d 210b 2110 2112 211b 212c 2130-2131'
+    + ' 2133 2205 2229-222a 223d 2242-2243 2248 2268-2269 2272-2273 228a-228b'
+    + ' 2293-2295 2297 229c 22cd 22da-22dd 29b7 2a3c-2a3d 2a6c 2a9d-2a9e'
+    + ' 2aac-2aad 2acb-2acc 3001-3002 31c0 31c6 31c9 31cf-31d2 31d4-31d5 31d7'
+    + ' 31db-31dd 31e5 a856 a85c a85e-a860 a868 aa60-aa66 aa6b-aa6c aa6f aa7a'
+    + ' ff01 ff0c ff0e ff10 ff1a-ff1b ff1f 10ac5-10ac6 10ad6-10ad7 10ae1 13091'
+    + ' 1310f 1311c 13121 13127 13139 131a0 131b1 131b8-131b9 131cb 131e0'
+    + ' 131f9-131fa 1327b 1327f 13285 1328c 132aa 132cb 132dc 132e7 13307'
+    + ' 1331b 13322 1333c 13377-13378 13399-1339a 133d3 133e4-133e5 133e7'
+    + ' 133ee 133f2 133f5-133f6 13416 13419-1341a 13423 13443-13446'
+    + ' 13be8-13bea 13f72 1d49c 1d49e-1d49f 1d4a2 1d4a5-1d4a6 1d4a9-1d4ac'
+    + ' 1d4ae-1d4b5 1db10'],
+  [0xfe01, '2018-2019 201c-201d 210b 2110 2112 211b 212c 2130-2131 2133 3001-3002'
+    + ' 31c6 31cf 31dd ff01 ff0c ff0e ff1a-ff1b ff1f 13093 13132 13139 13187'
+    + ' 131b1 131ee 131f8-131fa 13257 1327f 1328b 132a4 13308 13312-13314'
+    + ' 1331b 13321-13322 13331 133e8 13419 13f1f 14274 1d49c 1d49e-1d49f'
+    + ' 1d4a2 1d4a5-1d4a6 1d4a9-1d4ac 1d4ae-1d4b5'],
+  [0xfe02, '2018-2019 201c-201d 13117 13139 13183 131a0 131ba 131ee 13216 1327b'
+    + ' 132a4 132e7 132e9 132f8 132fd 13302-13303 13310-13314 1331c 13321'
+    + ' 13331 1334a 13361 13373 1337d 13385 133af-133b0 133bf 133db 133dd'
+    + ' 13419 1342c 1342e 14274 1d4cf'],
+  [0xfe03, '13012 130b8 130ba 1310f 1312f 13139 13184 1319c-1319d 1319f 131b1'
+    + ' 131db 13205 13296 13338 13419'],
+  [0xfe04, '13139 1340d'],
+  [0xfe05, '13139 14274'],
+  [0xfe06, '1312f 13139 13184 131db 131ee 13205 132a4 132e9 1332b 13338'
+    + ' 13370-13371 13419 14274'],
+].map(([selector, bases]) => [selector, codePoints(bases)]))
+const EMOJI = codePoints('23 2a 30-39 a9 ae 203c 2049 2122 2139 2194-2199 21a9-21aa 231a-231b 2328'
+  + ' 23cf 23e9-23f3 23f8-23fa 24c2 25aa-25ab 25b6 25c0 25fb-25fe 2600-2604 260e'
+  + ' 2611 2614-2615 2618 261d 2620 2622-2623 2626 262a 262e-262f 2638-263a 2640'
+  + ' 2642 2648-2653 265f-2660 2663 2665-2666 2668 267b 267e-267f 2692-2697 2699'
+  + ' 269b-269c 26a0-26a1 26a7 26aa-26ab 26b0-26b1 26bd-26be 26c4-26c5 26c8'
+  + ' 26ce-26cf 26d1 26d3-26d4 26e9-26ea 26f0-26f5 26f7-26fa 26fd 2702 2705'
+  + ' 2708-270d 270f 2712 2714 2716 271d 2721 2728 2733-2734 2744 2747 274c 274e'
+  + ' 2753-2755 2757 2763-2764 2795-2797 27a1 27b0 27bf 2934-2935 2b05-2b07'
+  + ' 2b1b-2b1c 2b50 2b55 3030 303d 3297 3299 1f004 1f170-1f171 1f17e-1f17f'
+  + ' 1f202 1f21a 1f22f 1f237 1f30d-1f30f 1f315 1f31c 1f321 1f324-1f32c 1f336'
+  + ' 1f378 1f37d 1f393 1f396-1f397 1f399-1f39b 1f39e-1f39f 1f3a7 1f3ac-1f3ae'
+  + ' 1f3c2 1f3c4 1f3c6 1f3ca-1f3ce 1f3d4-1f3e0 1f3ed 1f3f3 1f3f5 1f3f7 1f408'
+  + ' 1f415 1f41f 1f426 1f43f 1f441-1f442 1f446-1f449 1f44d-1f44e 1f453 1f46a'
+  + ' 1f47d 1f4a3 1f4b0 1f4b3 1f4bb 1f4bf 1f4cb 1f4da 1f4df 1f4e4-1f4e6'
+  + ' 1f4ea-1f4ed 1f4f7 1f4f9-1f4fb 1f4fd 1f508 1f50d 1f512-1f513 1f549-1f54a'
+  + ' 1f550-1f567 1f56f-1f570 1f573-1f579 1f587 1f58a-1f58d 1f590 1f5a5 1f5a8'
+  + ' 1f5b1-1f5b2 1f5bc 1f5c2-1f5c4 1f5d1-1f5d3 1f5dc-1f5de 1f5e1 1f5e3 1f5e8'
+  + ' 1f5ef 1f5f3 1f5fa 1f610 1f687 1f68d 1f691 1f694 1f698 1f6ad 1f6b2'
+  + ' 1f6b9-1f6ba 1f6bc 1f6cb 1f6cd-1f6cf 1f6e0-1f6e5 1f6e9 1f6f0 1f6f3')
+// The blocks of variation selectors, as those tables write them out as ranges.
+const BLOCKS = [[0xfe00, 0xfe0f], [0xe0100, 0xe01ef]]
+// The compatibility jamo but their filler.
+const JAMO = [[0x3131, 0x3163], [0x3165, 0x318e]]
+
+// The code point before `at` in `text`, a pair of surrogates read as one.
+function codePointBefore(text, at) {
+  if (at === 0) return undefined
+  const pair = at >= 2 ? text.codePointAt(at - 2) : 0
+  return pair > 0xffff ? pair : text.codePointAt(at - 1)
+}
+const within = (code, ...ranges) => code !== undefined && ranges.some(([from, to]) => code >= from && code <= to)
+
+// Whether the one shown as `code` at `at` hides nothing: a Hangul filler
+// composing Hangul (the initial consonant's before a vowel, a syllable with
+// none, or the vowel's after an initial consonant, one with none), or a
+// compatibility or halfwidth jamo's among those jamo, as a codepage's table
+// lists them, though not one after a syllable, a name that reads as it; a
+// variation sequence Unicode registers; a block of selectors written out as a
+// range in a table (`VS1-VS16`); or an emoji's presentation selector doubled,
+// as copying text leaves it.
+function benign(text, at, code) {
+  const end = at + (code > 0xffff ? 2 : 1)
+  const after = text.codePointAt(end), before = codePointBefore(text, at)
+  switch (code) {
+    case 0x115f: return within(after, [0x1161, 0x11a7], [0xd7b0, 0xd7c6])
+    case 0x1160: return within(before, [0x1100, 0x115e], [0xa960, 0xa97c])
+    case 0x3164: return within(before, ...JAMO) || within(after, ...JAMO)
+    case 0xffa0: return within(before, [0xffa1, 0xffdc]) || within(after, [0xffa1, 0xffdc])
+    default: return (code === 0xfe0e || code === 0xfe0f ? EMOJI.has(before) : STANDARDIZED.get(code)?.has(before))
+      || BLOCKS.some(([from, to]) => code === from && text[end] === '-' && text.codePointAt(end + 1) === to
+        || code === to && text[at - 1] === '-' && codePointBefore(text, at - 1) === from)
+      || (code === 0xfe0e || code === 0xfe0f) && before === code && PICTOGRAPH_BEFORE.test(text.slice(Math.max(0, at - 3), at - 1))
+  }
+}
+
+// How many of each a text holds, by code point, or null for none: each one
+// a code viewer shows (passed), visited alone, not each character it shows. `from`:
+// where to look from, as the first character past ASCII, since all of them are.
+export function invisibleCharacters(text, from = 0) {
+  let found = null
+  INVISIBLE_UNITS.lastIndex = from
+  for (let match = INVISIBLE_UNITS.exec(text); match; match = INVISIBLE_UNITS.exec(text)) {
+    const at = match.index, code = text.codePointAt(at)
+    if (!INVISIBLE.test(String.fromCodePoint(code))) continue
+    const next = passed(text, at, code, false)
+    if (next !== -1) INVISIBLE_UNITS.lastIndex = next
+    else if (!benign(text, at, code)) {
+      found ??= new Map()
+      found.set(code, (found.get(code) ?? 0) + 1)
+    }
+  }
+  return found
+}
+