@owlmeans/viable-common 0.0.4

package/src/convert/helpers.ts

@@ -0,0 +1,340 @@
+import {
+  ArchitectureCase, ConversionDecision, ConversionStage, CONVERSION_STAGE_ORDER,
+  CONVERTED_ORIGIN_DIR, EntropyClass, FileClass, ORIGIN_CARD_MAX_CHARS, SizeClass,
+  SIZE_CLASS_BOUNDS
+} from './consts.js'
+import type { OriginCard, StoryProof } from './types.js'
+
+/**
+ * Pure, deterministic derivations over the conversion vocabulary.
+ *
+ * Everything here answers the same way for the same input, in every runtime, forever — which is
+ * what makes the documents a conversion writes byte-stable across two runs, and therefore what
+ * makes the prompt cache prefix they are injected into actually hit. Nothing in this file reads a
+ * file, calls a model or looks at a clock.
+ */
+
+/** Which {@link SizeClass} a byte count falls in. Boundaries are exclusive upper bounds. */
+export const sizeClassOf = (bytes: number): SizeClass => {
+  if (bytes < SIZE_CLASS_BOUNDS[SizeClass.Tiny]) return SizeClass.Tiny
+  if (bytes < SIZE_CLASS_BOUNDS[SizeClass.Small]) return SizeClass.Small
+  if (bytes < SIZE_CLASS_BOUNDS[SizeClass.Medium]) return SizeClass.Medium
+  if (bytes < SIZE_CLASS_BOUNDS[SizeClass.Large]) return SizeClass.Large
+
+  return SizeClass.Huge
+}
+
+/**
+ * Extension → class, for the extensions that decide a file on their own.
+ *
+ * Deliberately not exhaustive: a path rule below overrides it wherever the location says more than
+ * the tail does (a `.ts` under `tests/` is a test, a `.json` under `seed/` is data).
+ */
+const EXT_CLASS: Record<string, FileClass> = {
+  ts: FileClass.Source, tsx: FileClass.Source, js: FileClass.Source, jsx: FileClass.Source,
+  mjs: FileClass.Source, cjs: FileClass.Source, mts: FileClass.Source, cts: FileClass.Source,
+  vue: FileClass.Source, svelte: FileClass.Source, java: FileClass.Source, kt: FileClass.Source,
+  kts: FileClass.Source, go: FileClass.Source, py: FileClass.Source, rb: FileClass.Source,
+  php: FileClass.Source, cs: FileClass.Source, ex: FileClass.Source, exs: FileClass.Source,
+  rs: FileClass.Source, sql: FileClass.Source, sh: FileClass.Source,
+
+  css: FileClass.Style, scss: FileClass.Style, sass: FileClass.Style, less: FileClass.Style,
+  styl: FileClass.Style,
+
+  md: FileClass.Doc, mdx: FileClass.Doc, rst: FileClass.Doc, adoc: FileClass.Doc,
+  txt: FileClass.Doc,
+
+  json: FileClass.Config, yaml: FileClass.Config, yml: FileClass.Config, toml: FileClass.Config,
+  ini: FileClass.Config, env: FileClass.Config, properties: FileClass.Config,
+  xml: FileClass.Config, conf: FileClass.Config,
+
+  html: FileClass.Template, htm: FileClass.Template, hbs: FileClass.Template,
+  ejs: FileClass.Template, pug: FileClass.Template, twig: FileClass.Template,
+  erb: FileClass.Template, jinja: FileClass.Template, j2: FileClass.Template,
+
+  csv: FileClass.Data, tsv: FileClass.Data, ndjson: FileClass.Data, jsonl: FileClass.Data,
+  parquet: FileClass.Data, dump: FileClass.Data,
+
+  png: FileClass.Binary, jpg: FileClass.Binary, jpeg: FileClass.Binary, gif: FileClass.Binary,
+  webp: FileClass.Binary, ico: FileClass.Binary, pdf: FileClass.Binary, zip: FileClass.Binary,
+  gz: FileClass.Binary, tar: FileClass.Binary, woff: FileClass.Binary, woff2: FileClass.Binary,
+  ttf: FileClass.Binary, otf: FileClass.Binary, mp4: FileClass.Binary, mp3: FileClass.Binary,
+  wasm: FileClass.Binary, jar: FileClass.Binary, so: FileClass.Binary, dll: FileClass.Binary,
+}
+
+/** Exact basenames that are a package manifest whatever their extension says. */
+const MANIFEST_NAMES = [
+  'package.json', 'deno.json', 'pom.xml', 'build.gradle', 'build.gradle.kts',
+  'settings.gradle', 'settings.gradle.kts', 'go.mod', 'cargo.toml', 'pyproject.toml',
+  'requirements.txt', 'pipfile', 'composer.json', 'gemfile', 'mix.exs',
+]
+
+/** Exact basenames that are a lockfile — never read, never counted as source. */
+const LOCK_NAMES = [
+  'bun.lock', 'bun.lockb', 'package-lock.json', 'yarn.lock', 'pnpm-lock.yaml', 'go.sum',
+  'cargo.lock', 'composer.lock', 'gemfile.lock', 'poetry.lock', 'pipfile.lock', 'mix.lock',
+]
+
+/**
+ * Path segments whose contents are produced by a build and say nothing about the application.
+ *
+ * `bin` is deliberately absent though `obj` is here: it is .NET/Java output in two of the nine
+ * ecosystems the converter reads and a real entrypoint directory in the rest (`bin/cli.ts`,
+ * `bin/console`, `bin/rails`), and this function is handed one path with no tree around it to
+ * tell the two apart. Dropping those entrypoints out of every source listing is the worse error.
+ */
+const GENERATED_SEGMENTS = [
+  'node_modules', 'dist', 'build', 'out', 'target', 'vendor', '.next', '.nuxt', '.svelte-kit',
+  '.turbo', '.gradle', '__pycache__', 'coverage', '.venv', 'obj',
+]
+
+/** Path segments that make a file a test wherever its extension would have said otherwise. */
+const TEST_SEGMENTS = ['test', 'tests', '__tests__', 'spec', 'specs', 'e2e']
+
+const segmentsOf = (path: string): string[] => path.split('/').filter(part => part !== '')
+
+const hasSegment = (path: string, of: string[]): boolean =>
+  segmentsOf(path).some(part => of.includes(part.toLowerCase()))
+
+/**
+ * The classes a seed or dump LOCATION is allowed to override.
+ *
+ * A seed directory says what a file holds, not what language it is written in — so it reclassifies
+ * a payload and never a program. `seed/users.json` is data; `src/seedUsers.ts` is the code that
+ * loads it, and `src/exportReport.ts` is a screen, and pushing either into the data budget takes
+ * real source out of the census's reach.
+ */
+const DATA_OVERRIDABLE: FileClass[] = [
+  FileClass.Data, FileClass.Config, FileClass.Binary, FileClass.Unknown,
+]
+
+/**
+ * The extensions that are as often a dump as they are a program.
+ *
+ * A `.sql` under `migrations/` is schema somebody wrote and a coder has to reproduce; the same
+ * tail under `dumps/` is an export nothing should read into a prompt. The location is the only
+ * thing that separates them, so it outranks the tail for these and for no other language.
+ */
+const DATA_AMBIGUOUS_EXT = ['sql']
+
+/**
+ * What one file of the origin is.
+ *
+ * Location outranks the tail wherever it says more — a `.ts` under `__tests__/` is a test, a
+ * `.json` under `seed/` is data — but only where it says more. The seed/dump rule therefore runs
+ * AFTER the extension table and applies only to {@link DATA_OVERRIDABLE}: asked first, it answered
+ * `Data` for every path merely containing the word (`tests/fixtures/user.ts`, `src/seedUsers.ts`,
+ * `src/exportReport.ts`), which is the census's source budget spent on the wrong files twice over.
+ */
+export const fileClassOf = (path: string, ext: string): FileClass => {
+  const lower = path.toLowerCase()
+  const base = segmentsOf(lower).pop() ?? lower
+  const tail = (ext.startsWith('.') ? ext.slice(1) : ext).toLowerCase()
+
+  if (LOCK_NAMES.includes(base)) return FileClass.Lock
+  if (hasSegment(lower, GENERATED_SEGMENTS)) return FileClass.Generated
+  if (MANIFEST_NAMES.includes(base)) return FileClass.Manifest
+  if (hasSegment(lower, TEST_SEGMENTS) || base.includes('.test.') || base.includes('.spec.')) {
+    return FileClass.Test
+  }
+  if (base.startsWith('.env')) return FileClass.Config
+
+  const byExt = EXT_CLASS[tail] ?? FileClass.Unknown
+  if (
+    (DATA_OVERRIDABLE.includes(byExt) || DATA_AMBIGUOUS_EXT.includes(tail))
+    && (isSeedPath(lower) || isBulkPath(lower))
+  ) {
+    return FileClass.Data
+  }
+
+  return byExt
+}
+
+/**
+ * Extensions whose tail already answers the binary question, either way.
+ *
+ * Deliberately two lists and not one with a default: an extension the census has never seen must
+ * be READ rather than guessed at, because a repository nobody here wrote carries tails nobody
+ * here chose. Nothing in either list is a guess about content — a `.ts` holding a NUL byte is a
+ * corrupt file, and a `.png` that is not binary is not a `.png`.
+ */
+const TEXT_EXTENSIONS: readonly string[] = [
+  '.ts', '.tsx', '.js', '.jsx', '.mjs', '.cjs', '.mts', '.cts', '.json', '.jsonc',
+  '.md', '.mdx', '.txt', '.css', '.scss', '.sass', '.less', '.html', '.htm', '.xml', '.svg',
+  '.yml', '.yaml', '.toml', '.ini', '.env', '.sh', '.bash', '.zsh', '.sql', '.graphql', '.gql',
+  '.py', '.rb', '.go', '.rs', '.java', '.kt', '.php', '.cs', '.c', '.h', '.cpp', '.hpp', '.swift',
+  '.vue', '.svelte', '.astro', '.lock', '.csv',
+]
+
+const BINARY_EXTENSIONS: readonly string[] = [
+  '.png', '.jpg', '.jpeg', '.gif', '.webp', '.avif', '.bmp', '.ico', '.icns', '.tiff',
+  '.mp3', '.mp4', '.wav', '.ogg', '.webm', '.mov', '.avi',
+  '.woff', '.woff2', '.ttf', '.otf', '.eot',
+  '.pdf', '.zip', '.gz', '.tgz', '.bz2', '.xz', '.7z', '.rar', '.tar',
+  '.wasm', '.node', '.so', '.dylib', '.dll', '.exe', '.bin', '.class', '.jar',
+  '.db', '.sqlite', '.sqlite3',
+]
+
+/**
+ * The binary verdict a path already carries, or `null` for a file that has to be probed.
+ *
+ * The three executors that walk a tree — the library-local helper, the publisher and the connector
+ * SDK — must answer the same verdict for the same file, and a probe alone does not: a small `.ico`
+ * with no NUL in its first bytes reads as text, so a census run on a laptop and the same census
+ * run in the slot disagreed about the same repository. The tail is asked first and the probe only
+ * where the tail says nothing, which is also why the probe is affordable at census scale — the
+ * extensions above are the bulk of any repository.
+ *
+ * The extension is taken exactly the way `path.extname` takes it, because two of the three callers
+ * used to: the last dot of the last SEGMENT, and nothing for a name with no dot or for a dotfile.
+ */
+export const binaryByExtension = (path: string): boolean | null => {
+  const base = segmentsOf(path).pop() ?? path
+  const dot = base.lastIndexOf('.')
+  if (dot < 1) return null
+  const ext = base.slice(dot).toLowerCase()
+
+  return TEXT_EXTENSIONS.includes(ext) ? false : BINARY_EXTENSIONS.includes(ext) ? true : null
+}
+
+/**
+ * How usable a file's head is as evidence.
+ *
+ * The printable ratio over the sampled head, and nothing cleverer: the question is only whether a
+ * model can read it. A minified bundle is printable and useless, so a second rule catches the
+ * shape that gives it away — very long lines with almost no whitespace.
+ */
+export const entropyClassOf = (head: string): EntropyClass => {
+  if (head.length < 1) return EntropyClass.Text
+
+  let printable = 0
+  let whitespace = 0
+  for (const char of head) {
+    const code = char.codePointAt(0) ?? 0
+    if (code === 9 || code === 10 || code === 13 || code === 32) {
+      ++whitespace
+      ++printable
+    } else if (code >= 32 && code !== 127) {
+      ++printable
+    }
+  }
+
+  const ratio = printable / head.length
+  if (ratio < 0.85) return EntropyClass.Opaque
+  // Prose, code and markup all break lines and indent. A body that does neither is minified,
+  // encoded or packed — printable, and not something a model can reason over.
+  if (whitespace / head.length < 0.05) return EntropyClass.Dense
+
+  return EntropyClass.Text
+}
+
+/** Path segments and name fragments that mark bulk data — an export, not the app's own data. */
+const BULK_HINTS = ['dump', 'dumps', 'export', 'exports', 'backup', 'backups', 'snapshot', 'archive']
+
+/** Path segments that mark seed data — what the application needs in order to make sense. */
+const SEED_HINTS = ['seed', 'seeds', 'fixture', 'fixtures', 'sample', 'samples', 'demo-data']
+
+const matchesHint = (path: string, hints: string[]): boolean => {
+  const lower = path.toLowerCase()
+  if (hasSegment(lower, hints)) return true
+  const base = segmentsOf(lower).pop() ?? lower
+
+  return hints.some(hint => base.includes(hint))
+}
+
+export const isBulkPath = (path: string): boolean => matchesHint(path, BULK_HINTS)
+
+export const isSeedPath = (path: string): boolean => matchesHint(path, SEED_HINTS)
+
+/**
+ * How much work re-implementing a story is, as a multiplier on the per-story price band.
+ *
+ * Derived from what the extraction found, never asked of a model: a story backed by a handful of
+ * proofs and no algorithm is ordinary CRUD, one that carries described algorithms is business
+ * logic somebody has to reproduce. Two runs over the same extraction therefore price identically,
+ * which is the point — a band that moved between two viewings of the same screen would be noise.
+ */
+export const storyComplexity = (proof: Pick<StoryProof, 'proofs' | 'algorithms'>): number => {
+  const proofs = proof.proofs?.length ?? 0
+  const algorithms = proof.algorithms?.length ?? 0
+
+  if (algorithms > 1 || proofs > 8) return 2
+  if (algorithms > 0 || proofs > 3) return 1.5
+
+  return 1
+}
+
+/**
+ * The stage after this one.
+ *
+ * The last stage answers with itself: there is nothing after implementation, and returning
+ * `undefined` would only move the same check to every call site.
+ */
+export const stageAfter = (stage: ConversionStage): ConversionStage => {
+  const at = CONVERSION_STAGE_ORDER.indexOf(stage)
+
+  return CONVERSION_STAGE_ORDER[at + 1] ?? stage
+}
+
+/**
+ * The decision that ADVANCES out of a stage.
+ *
+ * Implementation's is {@link ConversionDecision.Leave}, because a finished conversion has nothing
+ * further to run and leaving it as it stands is the ordinary outcome rather than an abandonment.
+ */
+export const decisionFor = (stage: ConversionStage): ConversionDecision => {
+  switch (stage) {
+    case ConversionStage.Intake: return ConversionDecision.Analyze
+    case ConversionStage.Analysis: return ConversionDecision.Extract
+    case ConversionStage.Extraction: return ConversionDecision.Implement
+    default: return ConversionDecision.Leave
+  }
+}
+
+/**
+ * Whether a conversion at `from` may enter `to`.
+ *
+ * The same stage is legal — that is a retry. The next one is legal. Nothing else is: every stage
+ * reads the documents the stage before it wrote, so skipping one means running against documents
+ * that were never produced, and going back means overwriting the record a later stage is built on.
+ */
+export const canEnter = (from: ConversionStage, to: ConversionStage): boolean => {
+  const at = CONVERSION_STAGE_ORDER.indexOf(from)
+  const next = CONVERSION_STAGE_ORDER.indexOf(to)
+  if (at < 0 || next < 0) return false
+
+  return next === at || next === at + 1
+}
+
+/** The size of the rendered origin card, so a composer can trim before it exceeds its ceiling. */
+export const originCardChars = (card: OriginCard): number => [
+  card.stack as string,
+  card.case as ArchitectureCase as string,
+  card.purpose,
+  ...card.packages,
+  ...card.entities,
+  ...card.flows,
+  ...card.conventions,
+].reduce((total, part) => total + (part?.length ?? 0), 0)
+
+/** Whether a rendered card still fits the budget every coder prompt carries it in. */
+export const originCardFits = (card: OriginCard): boolean =>
+  originCardChars(card) <= ORIGIN_CARD_MAX_CHARS
+
+/** Whether a target-relative path names the origin tree, or something inside it. */
+export const isOriginPath = (path: string): boolean =>
+  path === CONVERTED_ORIGIN_DIR || path.startsWith(`${CONVERTED_ORIGIN_DIR}/`)
+
+/**
+ * A path inside the origin tree, from a path relative to the origin's own root.
+ *
+ * Idempotent: a path already under the origin comes back unchanged, so a caller that has lost
+ * track of which side of the move it holds cannot produce `__viable_converted/__viable_converted/`.
+ */
+export const originPath = (relative: string): string => {
+  const path = relative.replace(/^\/+/, '')
+  if (path === '') return CONVERTED_ORIGIN_DIR
+  if (isOriginPath(path)) return path
+
+  return `${CONVERTED_ORIGIN_DIR}/${path}`
+}

package/src/convert/index.ts

@@ -0,0 +1,10 @@
+export * from './consts.js'
+export * from './schemas.js'
+export * from './helpers.js'
+export type * from './types.js'
+/**
+ * Declared beside the slot command that produces it and re-exported here, because the census is
+ * the only thing that reads a whole tree of them: one declaration, reachable from both barrels
+ * under the name every appendix uses. Two copies would drift the moment one gained a field.
+ */
+export type { FileStat } from '../slot/types.js'