@preventive/upstream 1.0.0-alpha.9 → 1.0.1

package/advisories.d.ts

@@ -1,9 +1,8 @@
-// Hand-written against advisories.js; a change to either belongs with the
-// other.
-
 import type { Client, RepoName } from './github.js'
+import type { CacheStore } from './npm.js'
 
 export { HttpError } from './npm.js'
+export type { CacheStore } from './npm.js'
 
 // As purl and stasis name them.
 export type Ecosystem = 'npm' | 'cargo' | 'composer' | 'soldeer' | 'github'
@@ -36,9 +35,16 @@ export interface AdvisoryOptions {
   // Also asks each npm, cargo and composer package's GitHub repository
   // for its published advisories, which it has before GitHub reviews them
   // into the databases above: the package's `github` where given, else
-  // the one npm's metadata, crates.io or Packagist names, looked up
-  // through the cache. Needs `github`.
+  // the one looked up. An npm package's is the one the document of its
+  // newest version asked names, kept for good as getMeta keeps it (none
+  // where the registry does not have that version); a crate's and a
+  // composer package's, the one crates.io or Packagist names, kept a
+  // month. Needs `github`.
   repoAdvisories?: boolean
+  // Where what this call caches is kept, as CacheOptions (npm.js) has it:
+  // a repository's listing, and each package's repository looked up (an
+  // npm package's as its version's document). See `advisories`.
+  cache?: CacheStore | false
   // Also returns each advisory's full text, as `details`. OSV's records
   // and a repository's listing already carry it; npm's registry does not,
   // and its rows take OSV's record of their GHSA, one more request each
@@ -53,7 +59,7 @@ export interface Advisory {
   name: string
   // `registry`: npm's, as `npm audit` has it, one row per vulnerable
   // range. `osv`: OSV's records for cargo and composer, RustSec's, GitHub's
-  // and malicious packages' (MAL-…) among them; of the records naming each
+  // and malicious packages' (MAL-...) among them; of the records naming each
   // other as aliases, directly or through others, a version is reported
   // under one only: RustSec's, else the GHSA, else the first by id. Each
   // has the others' ids among its aliases.
@@ -62,17 +68,17 @@ export interface Advisory {
   // under that GHSA. The only source for `soldeer` and `github`, where
   // every range counts, whichever package it names.
   source: 'registry' | 'osv' | 'repository'
-  // A GHSA, RUSTSEC-…, MAL-…, DRUPAL-CORE-… or another OSV id, or
+  // A GHSA, RUSTSEC-..., MAL-..., DRUPAL-CORE-... or another OSV id, or
   // npm:<id> for a registry row without a GHSA.
   id: string
   // The id itself, or an OSV record's one GHSA alias.
   ghsa?: string
   // GitHub's page for the GHSA, with it: a `repository` row's on its
-  // repository (https://github.com/owner/name/security/advisories/GHSA-…),
+  // repository (https://github.com/owner/name/security/advisories/GHSA-...),
   // which has it before GitHub reviews it into its advisory database; any
-  // other's in that database (https://github.com/advisories/GHSA-…). An
+  // other's in that database (https://github.com/advisories/GHSA-...). An
   // `osv` row without one has osv.dev's page for its record
-  // (https://osv.dev/vulnerability/RUSTSEC-…). Built, never fetched: only
+  // (https://osv.dev/vulnerability/RUSTSEC-...). Built, never fetched: only
   // a registry row without a GHSA (npm:<id>) has none.
   url?: string
   aliases: string[]
@@ -108,6 +114,11 @@ export interface Advisory {
 // nothing, and a GHSA OSV does not have adds no `details`. Any other
 // failure, a malformed answer, or one about something not asked, throws:
 // nothing is left out quietly.
+// A repository's listing is kept for 90 minutes, as one entry for every
+// spelling of its name, holding only what rows are made from, `details`
+// included: in setCacheDir's cache gzipped, in a `cache` store as it is,
+// or with `cache` false, nowhere. 90 minutes is how late an advisory its
+// maintainer publishes can be seen. One gone or malformed is not kept.
 // Sorted by ecosystem and name. Versions are matched by npm's semver, from
 // the npm beside node, or by the semver peer where there is no npm.
 export function advisories(packages: Iterable<Package>, options?: AdvisoryOptions): Promise<Advisory[]>

package/cargo.d.ts

@@ -1,5 +1,3 @@
-// Hand-written against cargo.js; a change to either belongs with the other.
-//
 // A name is asserted to be a crate name, a version an exact semver
 // version, and a checksum 64 lowercase hex digits, before any request or
 // cache path. The cache is setCacheDir's (npm.js).

package/git.d.ts

@@ -1,13 +1,11 @@
-// Hand-written against git.js; a change to either belongs with the other.
-
 // The shape getRepo (package.js) answers in, and HEAD's commit. Each
 // field best-effort: left out where it cannot be read.
 export interface GitCheckout {
   // `owner/name`, when the origin remote is a GitHub one. Never the
   // remote's URL, which can name a private host or carry a token.
   github?: string
-  // Where the directory sits in its checkout, `/`-separated; absent at
-  // the root.
+  // Where the directory sits in its checkout, `/`-separated; `''` at the
+  // root.
   directory?: string
   // `https://github.com/<github>`, whenever there is a `github`.
   url?: string

package/github/write.d.ts

@@ -1,5 +1,4 @@
-// Hand-written against github/write.js; a change to either belongs with
-// the other. Arguments are checked as github.d.ts describes.
+// Arguments are checked as github.d.ts describes.
 
 import type { Client, RepoName } from '../github.js'
 

package/github.d.ts

@@ -1,5 +1,3 @@
-// Hand-written against github.js; a change to either belongs with the other.
-//
 // Every method checks its arguments hard before any request is built: an
 // options object holding only the keys listed here, a repo that is
 // `owner/name` by GitHub's rules, branches and tags that git would take,
@@ -7,11 +5,12 @@
 // component, titles and headlines on one line. A bad one is a rejection
 // naming the method; nothing is sent. Every URL is built from checked,
 // encoded segments and has to come back out of URL parsing unchanged.
-// Redirects are refused, except the tarballs', so an answer is about the
-// repo asked for. A response is read up to a size limit (64 MiB of JSON,
-// 128 MiB of a file, 512 MiB of a tarball) and within a timeout (30
-// seconds, 5 minutes for a tarball), and what it says goes into an error
-// message escaped.
+// Redirects are refused, so an answer is about the repo asked for, except
+// the tarballs' and, where asked, getRepo's for a repo renamed or
+// transferred. A response is read up to a size limit (64 MiB of JSON, 128
+// MiB of a file, 512 MiB of a tarball) and within a timeout (30 seconds, 5
+// minutes for a tarball), and what it says goes into an error message
+// escaped.
 
 // `owner/name`.
 export type RepoName = string
@@ -35,14 +34,62 @@ export interface RepoDirEntry {
   sha: string
 }
 
-// A failed request: `status` is the HTTP status GitHub answered with.
+// A lockfile's format, by the tool that writes it, with the names it
+// writes one under and those of the project files it locks as
+// src/lockfiles.js lists them. Those @preventive/lockfile reads are as it
+// names the modules that do, yarn as yarn1.js, which reads Yarn 1's alone:
+// only the text tells a later Yarn's apart.
+export type LockfileFormat =
+  | 'pnpm' | 'npm' | 'yarn' | 'bun' | 'deno'
+  | 'uv' | 'poetry' | 'pylock' | 'pipenv' | 'pdm' | 'pixi' | 'conda-lock'
+  | 'bundler' | 'composer' | 'cargo' | 'soldeer' | 'foundry'
+  | 'cocoapods' | 'swiftpm' | 'carthage'
+  | 'go' | 'dep' | 'nuget' | 'paket' | 'gradle' | 'pub' | 'mix' | 'rebar' | 'julia'
+  | 'stack' | 'cabal' | 'shards' | 'opam' | 'nix' | 'conan' | 'bazel' | 'helm' | 'renv' | 'terraform'
+
+// A file by a lockfile's name: `path` from the repo's root, as getRepoFile
+// takes it, and `sha` its blob's id.
+export interface RepoLockfile {
+  path: string
+  format: LockfileFormat
+  sha: string
+}
+
+// A file by a project file's name, as RepoLockfile is, and the formats of
+// the lockfiles that may lock its project, in src/lockfiles.js' order:
+// whether one does, and where, is the tool's to say, a workspace's at its
+// root.
+export interface RepoProjectFile {
+  path: string
+  formats: LockfileFormat[]
+  sha: string
+}
+
+// A failed request: `status` is the HTTP status GitHub answered with, and
+// `headers` its response's (a redirect's `location` among them), on every
+// call but a GraphQL one (github/write.js).
 export class HttpError extends Error {
   name: 'HttpError'
   status: number
+  headers?: Headers
 }
 
 // Reads only.
 export interface Client {
+  // Every file by a lockfile's name, and every project file of a tool that
+  // writes one, in the full commit `sha`, or under its `directory`, found as
+  // getRepoTreeId finds it, each list sorted by path; and a project file no
+  // name tells, by a lockfile beside it named after it: a Python script
+  // `uv lock --script` locked, or a Stack or Cabal project file of another
+  // name than stack.yaml or cabal.project. From one listing of the tree
+  // with every tree under it, each hashed back to its id, so none is left
+  // out or changed; where GitHub cuts that short, as it does past 100,000
+  // entries, from the tree's own listing and then each subtree's in turn.
+  // Not looked in: a submodule, which is another repo's commit, or a
+  // `node_modules`, of packages installed. A symlink is no file. Refused
+  // where a path found is one getRepoFile does not take, such as one with a
+  // control character in it.
+  findRepoLockfiles(options: { repo: RepoName; sha: string; directory?: string }): Promise<{ lockfiles: RepoLockfile[]; projects: RepoProjectFile[] }>
   // An advisory by its GHSA id. With `repo`, that repository's published
   // copy: the maintainer's latest text, there before GitHub reviews it,
   // and shaped as a repository advisory (`state`, `patched_versions`, no
@@ -58,7 +105,14 @@ export interface Client {
   // Refused unless GitHub answers with that pull request in that repo.
   getPullRequest(options: { repo: RepoName; number: number }): Promise<{ title: string; status: PullRequestStatus }>
   // GitHub's repository object, refused unless it is the repo asked for.
-  getRepo(options: { repo: RepoName }): Promise<any>
+  // With `followRedirect`, a repository renamed or transferred, which GitHub
+  // answers with a redirect to it by its id
+  // (`https://api.github.com/repositories/<id>`), is asked for there, with
+  // the same token, once: the repository as it is now, refused unless it
+  // has that id and a `full_name` that is `owner/name`, the name to ask its
+  // other calls with. A redirect anywhere else, or a second, is refused as
+  // without it: an HttpError with the redirect's status.
+  getRepo(options: { repo: RepoName; followRedirect?: boolean }): Promise<any>
   // A file's contents as text, at `ref` or the default branch. Refused for
   // a directory, a symlink or a submodule, unless the bytes hash to the
   // blob GitHub names for the path, and unless valid UTF-8.

package/npm.d.ts

@@ -1,5 +1,3 @@
-// Hand-written against npm.js; a change to either belongs with the other.
-//
 // Every call that reaches the registry first asserts that the name is a
 // string of the shape npm takes, and a version a string that npm's
 // semver.valid answers with unchanged; neither goes near a request, or a
@@ -7,8 +5,9 @@
 // a timeout, as github.d.ts describes, and lookups for many names go to
 // the registry eight at a time.
 //
-// Where NPM_TOKEN is set, it is sent with the requests for a scoped
-// package, so a private one can be read.
+// Where PREVENTIVE_MEASURES_NPM_TOKEN, STASIS_NPM_TOKEN or NPM_TOKEN is
+// set, the first of them not empty, in that order, is sent with the
+// requests for a scoped package, so a private one can be read.
 
 // Where a program called `name` keeps its cache for this user, as each
 // platform has it, read from the environment when called. macOS:
@@ -32,26 +31,59 @@ export const defaultCacheDir: string | undefined
 // cache: every read of it misses and every write is skipped. getTarball
 // reads ours all the same, set or not.
 //
-// The directory is trusted: a package's repo is answered from it as it
-// was written, with no request. Point it only at storage that nothing
-// less trusted than the caller can write, never a cache shared with or
-// restored from lower-trust jobs. Tarballs, crates (cargo.js), zips
+// The directory is trusted: a package's repo, as recorded or as a version
+// document kept here names it, and for 90 minutes a repository's published
+// advisories (advisories.js), are answered from it as they were written,
+// with no request. Point it only at storage that
+// nothing less trusted than the caller can write, never a cache shared
+// with or restored from lower-trust jobs. Tarballs, crates (cargo.js), zips
 // (soldeer.js) and tree tarballs (github.js) don't rely on this, since
-// they are checked against their integrity on every call.
+// they are checked against their integrity on every call; and the version
+// documents kept here (getMeta) never supply an integrity, only confirm
+// one a caller gives (verifyDist, and getMeta with a `dist`). A call's `cache` option, where it has
+// one, keeps what it caches elsewhere, or writes none of it here.
 export function setCacheDir(dir?: string | false): void
 
+// A store of the caller's, a database's say, for what would otherwise be
+// kept in setCacheDir's cache: `read` answers the value last written
+// under that `type` and `key`, or null or undefined for none. A `type` is
+// what is kept and a `key` which one: `npm/versions` and `npm/tarballs`
+// with `<name>@<version>` for a version's document and tarball; and for
+// advisories.js, `github/advisories` with the repository's `owner/name` in
+// lowercase for its listing, and `cargo/repos`, `composer/repos` and
+// `soldeer/repos` with a package's name for the repository it names.
+// Values are plain JSON data, which `read` may answer as written or as a
+// copy, but a tarball's are its bytes, a Uint8Array. Whatever it answers
+// is checked as a cached file is: one malformed, stale or kept differently
+// is a miss, and a tarball's bytes that do not match throw. A rejection
+// from either is the call's. It is trusted as setCacheDir's directory is.
+export interface CacheStore {
+  read(type: string, key: string): Promise<unknown>
+  write(type: string, key: string, value: unknown): Promise<void>
+}
+
+// Where a call keeps what it caches: in setCacheDir's cache where `cache`
+// is left out; in a store of the caller's in place of it; or with `false`,
+// nowhere, though setCacheDir's cache is read all the same. Other tools'
+// caches and ours are read for tarballs whatever it is (getTarball).
+export interface CacheOptions {
+  cache?: CacheStore | false
+}
+
 // A failed request: `status` is the HTTP status the registry answered
-// with (404 for a package or version it does not have).
+// with (404 for a package or version it does not have), and `headers` its
+// response's.
 export class HttpError extends Error {
   name: 'HttpError'
   status: number
+  headers?: Headers
 }
 
-// A package's GitHub repo as its registry metadata names it — getRepo
+// A package's GitHub repo as its registry metadata names it -- getRepo
 // (package.js) over its `latest` document: `github` is `owner/name`, and
-// `directory` is where in the repo the package sits, absent for a
-// package at the repo root. Throws when the package cannot be fetched or
-// names no GitHub repo.
+// `directory` is where in the repo the package sits, `''` at the repo
+// root, absent where it declares none. Throws when the package cannot be
+// fetched or names no GitHub repo.
 export interface GitHubLink {
   github: string
   directory?: string
@@ -71,7 +103,10 @@ export function resolvePackageRepos(packageNames: Iterable<string>, options?: {
 // The disk cache resolvePackageRepos reads and writes, a month per entry.
 // A read answers null for anything but a usable entry, one written by a
 // release that resolved repos by other rules included; a write answers
-// false where it could not write, and never throws.
+// false where it could not write, and never throws. A write takes the
+// `github` a lookup gives, `owner/name` with no trailing `.git`, so what
+// it writes reads back; a `directory` left out is unknown, and `''` the
+// repo root.
 export function readPackageRepoCache(name: string): Promise<PackageRepo | null>
 export function writePackageRepoCache(name: string, github: string, directory?: string): Promise<boolean>
 
@@ -84,25 +119,83 @@ export interface Dist {
   integrity: string
 }
 
+// `gitHead` is the commit the document names, as npm reads it off the git
+// checkout it publishes from, where it is a full commit id: 40 lowercase
+// hex digits, or 64. `repository`, `homepage` and `bugs` are the
+// document's, as the package.json published had them, or npm filled them
+// in from one another: each a string as it is, or for `repository` and
+// `bugs` an object, of which only the strings named below are read. Each is
+// absent where the document has none, or has another shape, or an object
+// with none of those strings. All of it is the publisher's word, held to no
+// repository.
 export interface PackageMeta {
   name: string
   version: string
   dist: Dist
+  gitHead?: string
+  repository?: string | PackageRepository
+  homepage?: string
+  bugs?: string | PackageBugs
+}
+
+export interface PackageRepository {
+  type: string | undefined
+  url: string | undefined
+  directory: string | undefined
+}
+
+export interface PackageBugs {
+  url: string | undefined
+  email: string | undefined
+}
+
+// getMeta's options: `cache`, and the `dist` a caller already has of that
+// version, as a lockfile gives it, held to the rules above.
+export interface MetaOptions extends CacheOptions {
+  dist?: Dist
 }
 
 // The registry's version document, refused unless it is for that name and
 // version and its dist is held to the rules above.
-export function getMeta(name: string, version: string): Promise<PackageMeta>
+//
+// With `dist`, npm's own cache comes first, where getTarball looks for
+// npm's tarballs: where it keeps the full packument of that name, as npm 10
+// and later fetch it, and that version's entry there has the given dist's
+// tarball and integrity, that entry answers, with no request, and neither
+// `cache` nor setCacheDir's cache is read or written. An abbreviated
+// packument, as earlier npm fetches, is passed over, as is one whose
+// content does not match the sha512 npm filed it by. What it answers
+// beside the dist is taken from npm's cache as it is from setCacheDir's.
+// Else the document `cache` keeps answers, with no request, where its
+// tarball and integrity are the given dist's; else the registry's is asked
+// for, and throws where they are not.
+//
+// Without `dist`, the registry is asked on every call: a cache never
+// supplies a dist. Kept whole, as `cache` says, and for good once it
+// passes, as the registry never takes a version twice: in setCacheDir's
+// cache, compressed, where `cache` is left out. One refused is not kept.
+export function getMeta(name: string, version: string, options?: MetaOptions): Promise<PackageMeta>
+
+// When the registry says each version of a package it lists was published,
+// by version, as Date#toISOString writes it: from the whole package's
+// document, the only one with `time`, megabytes for some. A version whose
+// time is missing, not an ISO 8601 UTC time, or not one there is, as
+// 2018-02-29, is left out. Refused unless the document is for that name.
+export function getPublishTimes(name: string): Promise<Map<string, string>>
 
-// Fetches the registry's dist for that version and throws unless its
-// integrity is the one given, as for a dist read off a lockfile.
-export function verifyDist(name: string, version: string, dist: Dist): Promise<void>
+// Throws unless the integrity of that version is the one given, as for a
+// dist read off a lockfile: by the version document `cache` keeps, with no
+// request, else by the registry's, kept then as getMeta keeps it. One kept
+// with another integrity, refused, or for another name or version, is
+// passed over for the registry's, which throws where it has another.
+export function verifyDist(name: string, version: string, dist: Dist, options?: CacheOptions): Promise<void>
 
 // A published version's gzipped tarball, whole, in memory. Without `dist`,
-// the version document is fetched on every call, cached tarball or not;
-// with one, it is trusted as given and nothing but the tarball is asked
-// for. Either way the bytes are checked against `dist.integrity`, whether
-// they were downloaded (before they are cached) or read from a cache.
+// the version document is fetched on every call, cached tarball or not,
+// and kept as getMeta keeps it; with one, it is trusted as given and
+// nothing but the tarball is asked for. Either way the bytes are checked
+// against `dist.integrity`, whether they were downloaded (before they are
+// cached) or read from a cache.
 //
 // Other caches are read first, and never written: npm's (cacache under
 // npm_config_cache, else npm's default, `~/.npm` outside Windows; and npm
@@ -110,5 +203,7 @@ export function verifyDist(name: string, version: string, dist: Dist): Promise<v
 // `~/.audit/cache/tgz/<org>:<name>-<version>.tgz`, then defaultCacheDir
 // and cacheDirFor('stasis'), as setCacheDir files a tarball in either,
 // whether set or not. A file there that does not match is passed over.
-// Then setCacheDir's cache, where one that does not match throws.
-export function getTarball(name: string, version: string, dist?: Dist): Promise<Uint8Array>
+// Then setCacheDir's cache, or a `cache` store in its place, where bytes
+// that do not match throw. A tarball downloaded is kept there, unless
+// `cache` is false.
+export function getTarball(name: string, version: string, dist?: Dist, options?: CacheOptions): Promise<Uint8Array>

package/npm.js

@@ -1,4 +1,4 @@
 export { cacheDirFor, defaultCacheDir, setCacheDir } from './src/cache.js'
 export { HttpError } from './src/http.js'
 export { getGitHub, readPackageRepoCache, resolvePackageRepos, writePackageRepoCache } from './src/npm/repos.js'
-export { getMeta, getTarball, verifyDist } from './src/npm/tarball.js'
+export { getMeta, getPublishTimes, getTarball, verifyDist } from './src/npm/tarball.js'

package/package.d.ts

@@ -1,11 +1,12 @@
-// Hand-written against package.js; a change to either belongs with the other.
-
 export interface PackageRepoLink {
   // `owner/name`, by GitHub's own rules.
   github?: string
   // Where in that repo the package sits, `/`-separated, as the tree path
-  // itself (a homepage URL's decoded); absent at its root, and whenever
-  // `github` is.
+  // itself (a homepage URL's decoded; a `repository.directory`'s `\` read
+  // as `/`, its empty and `.` parts dropped); `''` at its root, where
+  // `repository.directory` declares it (`./`, `/`, `/.`, `.`, `''`). Absent
+  // where unknown: neither field names a directory in it (one with a `..`
+  // part names none), and whenever `github` is absent.
   directory?: string
   // `https://github.com/<github>`, whenever there is a `github`.
   url?: string

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@preventive/upstream",
-  "version": "1.0.0-alpha.9",
+  "version": "1.0.1",
   "description": "Minimal npm, crates.io and Soldeer registry clients, a GitHub client, and a git checkout reader: the calls we use, nothing more",
   "type": "module",
   "exports": {
@@ -85,9 +85,12 @@
     "src/github/read.js",
     "src/github/write.js",
     "src/http.js",
+    "src/lockfiles.js",
+    "src/npm/cacache.js",
     "src/npm/registry.js",
     "src/npm/repos.js",
     "src/npm/tarball.js",
+    "src/npm/versions.js",
     "src/package.js",
     "src/pool.js",
     "src/remote.js",

package/semver.d.ts

@@ -1,5 +1,3 @@
-// Hand-written against semver.js; a change to either belongs with the other.
-
 // npm's own semver, borrowed from the npm beside node and passed through
 // as it is; where there is no npm, the `semver` package, an optional peer
 // dependency. Without either, every call throws, except valid() on a plain

package/soldeer.d.ts

@@ -1,6 +1,3 @@
-// Hand-written against soldeer.js; a change to either belongs with the
-// other.
-//
 // The name is asserted to be one Soldeer takes, the version a string of
 // letters, digits, `.`, `_`, `+` and `-`, and the checksum 64 lowercase hex
 // digits, before any request or cache path. The cache is setCacheDir's