'use strict'
const { resolve } = require('path')
const { parser, arrayDelimiter } = require('@npmcli/query')
const localeCompare = require('@isaacs/string-locale-compare')('en')
const log = require('proc-log')
const minimatch = require('minimatch')
const npa = require('npm-package-arg')
const pacote = require('pacote')
const semver = require('semver')
// handle results for parsed query asts, results are stored in a map that has a
// key that points to each ast selector node and stores the resulting array of
// arborist nodes as its value, that is essential to how we handle multiple
// query selectors, e.g: `#a, #b, #c` <- 3 diff ast selector nodes
class Results {
#currentAstSelector
#initialItems
#inventory
#outdatedCache = new Map()
#pendingCombinator
#results = new Map()
#targetNode
constructor (opts) {
this.#currentAstSelector = opts.rootAstNode.nodes[0]
this.#inventory = opts.inventory
this.#initialItems = opts.initialItems
this.#targetNode = opts.targetNode
this.currentResults = this.#initialItems
// We get this when first called and need to pass it to pacote
this.flatOptions = opts.flatOptions || {}
// reset by rootAstNode walker
this.currentAstNode = opts.rootAstNode
}
get currentResults () {
return this.#results.get(this.#currentAstSelector)
}
set currentResults (value) {
this.#results.set(this.#currentAstSelector, value)
}
// retrieves the initial items to which start the filtering / matching
// for most of the different types of recognized ast nodes, e.g: class (aka
// depType), id, *, etc in different contexts we need to start with the
// current list of filtered results, for example a query for `.workspace`
// actually means the same as `*.workspace` so we want to start with the full
// inventory if that's the first ast node we're reading but if it appears in
// the middle of a query it should respect the previous filtered results,
// combinators are a special case in which we always want to have the
// complete inventory list in order to use the left-hand side ast node as a
// filter combined with the element on its right-hand side
get initialItems () {
const firstParsed =
(this.currentAstNode.parent.nodes[0] === this.currentAstNode) &&
(this.currentAstNode.parent.parent.type === 'root')
if (firstParsed) {
return this.#initialItems
}
if (this.currentAstNode.prev().type === 'combinator') {
return this.#inventory
}
return this.currentResults
}
// combinators need information about previously filtered items along
// with info of the items parsed / retrieved from the selector right
// past the combinator, for this reason combinators are stored and
// only ran as the last part of each selector logic
processPendingCombinator (nextResults) {
if (this.#pendingCombinator) {
const res = this.#pendingCombinator(this.currentResults, nextResults)
this.#pendingCombinator = null
this.currentResults = res
} else {
this.currentResults = nextResults
}
}
// when collecting results to a root astNode, we traverse the list of child
// selector nodes and collect all of their resulting arborist nodes into a
// single/flat Set of items, this ensures we also deduplicate items
collect (rootAstNode) {
return new Set(rootAstNode.nodes.flatMap(n => this.#results.get(n)))
}
// selector types map to the '.type' property of the ast nodes via `${astNode.type}Type`
//
// attribute selector [name=value], etc
attributeType () {
const nextResults = this.initialItems.filter(node =>
attributeMatch(this.currentAstNode, node.package)
)
this.processPendingCombinator(nextResults)
}
// dependency type selector (i.e. .prod, .dev, etc)
// css calls this class, we interpret is as dependency type
classType () {
const depTypeFn = depTypes[String(this.currentAstNode)]
if (!depTypeFn) {
throw Object.assign(
new Error(`\`${String(this.currentAstNode)}\` is not a supported dependency type.`),
{ code: 'EQUERYNODEPTYPE' }
)
}
const nextResults = depTypeFn(this.initialItems)
this.processPendingCombinator(nextResults)
}
// combinators (i.e. '>', ' ', '~')
combinatorType () {
this.#pendingCombinator = combinators[String(this.currentAstNode)]
}
// name selectors (i.e. #foo)
// css calls this id, we interpret it as name
idType () {
const name = this.currentAstNode.value
const nextResults = this.initialItems.filter(node =>
(name === node.name) || (name === node.package.name)
)
this.processPendingCombinator(nextResults)
}
// pseudo selectors (prefixed with :)
async pseudoType () {
const pseudoFn = `${this.currentAstNode.value.slice(1)}Pseudo`
if (!this[pseudoFn]) {
throw Object.assign(
new Error(`\`${this.currentAstNode.value
}\` is not a supported pseudo selector.`),
{ code: 'EQUERYNOPSEUDO' }
)
}
const nextResults = await this[pseudoFn]()
this.processPendingCombinator(nextResults)
}
selectorType () {
this.#currentAstSelector = this.currentAstNode
// starts a new array in which resulting items
// can be stored for each given ast selector
if (!this.currentResults) {
this.currentResults = []
}
}
universalType () {
this.processPendingCombinator(this.initialItems)
}
// pseudo selectors map to the 'value' property of the pseudo selectors in the ast nodes
// via selectors via `${value.slice(1)}Pseudo`
attrPseudo () {
const { lookupProperties, attributeMatcher } = this.currentAstNode
return this.initialItems.filter(node => {
let objs = [node.package]
for (const prop of lookupProperties) {
// if an isArray symbol is found that means we'll need to iterate
// over the previous found array to basically make sure we traverse
// all its indexes testing for possible objects that may eventually
// hold more keys specified in a selector
if (prop === arrayDelimiter) {
objs = objs.flat()
continue
}
// otherwise just maps all currently found objs
// to the next prop from the lookup properties list,
// filters out any empty key lookup
objs = objs.flatMap(obj => obj[prop] || [])
// in case there's no property found in the lookup
// just filters that item out
const noAttr = objs.every(obj => !obj)
if (noAttr) {
return false
}
}
// if any of the potential object matches
// that item should be in the final result
return objs.some(obj => attributeMatch(attributeMatcher, obj))
})
}
emptyPseudo () {
return this.initialItems.filter(node => node.edgesOut.size === 0)
}
extraneousPseudo () {
return this.initialItems.filter(node => node.extraneous)
}
async hasPseudo () {
const found = []
for (const item of this.initialItems) {
// This is the one time initialItems differs from inventory
const res = await retrieveNodesFromParsedAst({
flatOptions: this.flatOptions,
initialItems: [item],
inventory: this.#inventory,
rootAstNode: this.currentAstNode.nestedNode,
targetNode: item,
})
if (res.size > 0) {
found.push(item)
}
}
return found
}
invalidPseudo () {
const found = []
for (const node of this.initialItems) {
for (const edge of node.edgesIn) {
if (edge.invalid) {
found.push(node)
break
}
}
}
return found
}
async isPseudo () {
const res = await retrieveNodesFromParsedAst({
flatOptions: this.flatOptions,
initialItems: this.initialItems,
inventory: this.#inventory,
rootAstNode: this.currentAstNode.nestedNode,
targetNode: this.currentAstNode,
})
return [...res]
}
linkPseudo () {
return this.initialItems.filter(node => node.isLink || (node.isTop && !node.isRoot))
}
missingPseudo () {
return this.#inventory.reduce((res, node) => {
for (const edge of node.edgesOut.values()) {
if (edge.missing) {
const pkg = { name: edge.name, version: edge.spec }
res.push(new this.#targetNode.constructor({ pkg }))
}
}
return res
}, [])
}
async notPseudo () {
const res = await retrieveNodesFromParsedAst({
flatOptions: this.flatOptions,
initialItems: this.initialItems,
inventory: this.#inventory,
rootAstNode: this.currentAstNode.nestedNode,
targetNode: this.currentAstNode,
})
const internalSelector = new Set(res)
return this.initialItems.filter(node =>
!internalSelector.has(node))
}
overriddenPseudo () {
return this.initialItems.filter(node => node.overridden)
}
pathPseudo () {
return this.initialItems.filter(node => {
if (!this.currentAstNode.pathValue) {
return true
}
return minimatch(
node.realpath.replace(/\\+/g, '/'),
resolve(node.root.realpath, this.currentAstNode.pathValue).replace(/\\+/g, '/')
)
})
}
privatePseudo () {
return this.initialItems.filter(node => node.package.private)
}
rootPseudo () {
return this.initialItems.filter(node => node === this.#targetNode.root)
}
scopePseudo () {
return this.initialItems.filter(node => node === this.#targetNode)
}
semverPseudo () {
const {
attributeMatcher,
lookupProperties,
semverFunc = 'infer',
semverValue,
} = this.currentAstNode
const { qualifiedAttribute } = attributeMatcher
if (!semverValue) {
// DEPRECATED: remove this warning and throw an error as part of @npmcli/arborist@6
log.warn('query', 'usage of :semver() with no parameters is deprecated')
return this.initialItems
}
if (!semver.valid(semverValue) && !semver.validRange(semverValue)) {
throw Object.assign(
new Error(`\`${semverValue}\` is not a valid semver version or range`),
{ code: 'EQUERYINVALIDSEMVER' })
}
const valueIsVersion = !!semver.valid(semverValue)
const nodeMatches = (node, obj) => {
// if we already have an operator, the user provided some test as part of the selector
// we evaluate that first because if it fails we don't want this node anyway
if (attributeMatcher.operator) {
if (!attributeMatch(attributeMatcher, obj)) {
// if the initial operator doesn't match, we're done
return false
}
}
const attrValue = obj[qualifiedAttribute]
// both valid and validRange return null for undefined, so this will skip both nodes that
// do not have the attribute defined as well as those where the attribute value is invalid
// and those where the value from the package.json is not a string
if ((!semver.valid(attrValue) && !semver.validRange(attrValue)) ||
typeof attrValue !== 'string') {
return false
}
const attrIsVersion = !!semver.valid(attrValue)
let actualFunc = semverFunc
// if we're asked to infer, we examine outputs to make a best guess
if (actualFunc === 'infer') {
if (valueIsVersion && attrIsVersion) {
// two versions -> semver.eq
actualFunc = 'eq'
} else if (!valueIsVersion && !attrIsVersion) {
// two ranges -> semver.intersects
actualFunc = 'intersects'
} else {
// anything else -> semver.satisfies
actualFunc = 'satisfies'
}
}
if (['eq', 'neq', 'gt', 'gte', 'lt', 'lte'].includes(actualFunc)) {
// both sides must be versions, but one is not
if (!valueIsVersion || !attrIsVersion) {
return false
}
return semver[actualFunc](attrValue, semverValue)
} else if (['gtr', 'ltr', 'satisfies'].includes(actualFunc)) {
// at least one side must be a version, but neither is
if (!valueIsVersion && !attrIsVersion) {
return false
}
return valueIsVersion
? semver[actualFunc](semverValue, attrValue)
: semver[actualFunc](attrValue, semverValue)
} else if (['intersects', 'subset'].includes(actualFunc)) {
// these accept two ranges and since a version is also a range, anything goes
return semver[actualFunc](attrValue, semverValue)
} else {
// user provided a function we don't know about, throw an error
throw Object.assign(new Error(`\`semver.${actualFunc}\` is not a supported operator.`),
{ code: 'EQUERYINVALIDOPERATOR' })
}
}
return this.initialItems.filter((node) => {
// no lookupProperties just means its a top level property, see if it matches
if (!lookupProperties.length) {
return nodeMatches(node, node.package)
}
// this code is mostly duplicated from attrPseudo to traverse into the package until we get
// to our deepest requested object
let objs = [node.package]
for (const prop of lookupProperties) {
if (prop === arrayDelimiter) {
objs = objs.flat()
continue
}
objs = objs.flatMap(obj => obj[prop] || [])
const noAttr = objs.every(obj => !obj)
if (noAttr) {
return false
}
return objs.some(obj => nodeMatches(node, obj))
}
})
}
typePseudo () {
if (!this.currentAstNode.typeValue) {
return this.initialItems
}
return this.initialItems
.flatMap(node => {
const found = []
for (const edge of node.edgesIn) {
if (npa(`${edge.name}@${edge.spec}`).type === this.currentAstNode.typeValue) {
found.push(edge.to)
}
}
return found
})
}
dedupedPseudo () {
return this.initialItems.filter(node => node.target.edgesIn.size > 1)
}
async outdatedPseudo () {
const { outdatedKind = 'any' } = this.currentAstNode
// filter the initialItems
// NOTE: this uses a Promise.all around a map without in-line concurrency handling
// since the only async action taken is retrieving the packument, which is limited
// based on the max-sockets config in make-fetch-happen
const initialResults = await Promise.all(this.initialItems.map(async (node) => {
// the root can't be outdated, skip it
if (node.isProjectRoot) {
return false
}
// we cache the promise representing the full versions list, this helps reduce the
// number of requests we send by keeping population of the cache in a single tick
// making it less likely that multiple requests for the same package will be inflight
if (!this.#outdatedCache.has(node.name)) {
this.#outdatedCache.set(node.name, getPackageVersions(node.name, this.flatOptions))
}
const availableVersions = await this.#outdatedCache.get(node.name)
// we attach _all_ versions to the queryContext to allow consumers to do their own
// filtering and comparisons
node.queryContext.versions = availableVersions
// next we further reduce the set to versions that are greater than the current one
const greaterVersions = availableVersions.filter((available) => {
return semver.gt(available, node.version)
})
// no newer versions than the current one, drop this node from the result set
if (!greaterVersions.length) {
return false
}
// if we got here, we know that newer versions exist, if the kind is 'any' we're done
if (outdatedKind === 'any') {
return node
}
// look for newer versions that differ from current by a specific part of the semver version
if (['major', 'minor', 'patch'].includes(outdatedKind)) {
// filter the versions greater than our current one based on semver.diff
const filteredVersions = greaterVersions.filter((version) => {
return semver.diff(node.version, version) === outdatedKind
})
// no available versions are of the correct diff type
if (!filteredVersions.length) {
return false
}
return node
}
// look for newer versions that satisfy at least one edgeIn to this node
if (outdatedKind === 'in-range') {
const inRangeContext = []
for (const edge of node.edgesIn) {
const inRangeVersions = greaterVersions.filter((version) => {
return semver.satisfies(version, edge.spec)
})
// this edge has no in-range candidates, just move on
if (!inRangeVersions.length) {
continue
}
inRangeContext.push({
from: edge.from.location,
versions: inRangeVersions,
})
}
// if we didn't find at least one match, drop this node
if (!inRangeContext.length) {
return false
}
// now add to the context each version that is in-range for each edgeIn
node.queryContext.outdated = {
...node.queryContext.outdated,
inRange: inRangeContext,
}
return node
}
// look for newer versions that _do not_ satisfy at least one edgeIn
if (outdatedKind === 'out-of-range') {
const outOfRangeContext = []
for (const edge of node.edgesIn) {
const outOfRangeVersions = greaterVersions.filter((version) => {
return !semver.satisfies(version, edge.spec)
})
// this edge has no out-of-range candidates, skip it
if (!outOfRangeVersions.length) {
continue
}
outOfRangeContext.push({
from: edge.from.location,
versions: outOfRangeVersions,
})
}
// if we didn't add at least one thing to the context, this node is not a match
if (!outOfRangeContext.length) {
return false
}
// attach the out-of-range context to the node
node.queryContext.outdated = {
...node.queryContext.outdated,
outOfRange: outOfRangeContext,
}
return node
}
// any other outdatedKind is unknown and will never match
return false
}))
// return an array with the holes for non-matching nodes removed
return initialResults.filter(Boolean)
}
}
// operators for attribute selectors
const attributeOperators = {
// attribute value is equivalent
'=' ({ attr, value, insensitive }) {
return attr === value
},
// attribute value contains word
'~=' ({ attr, value, insensitive }) {
return (attr.match(/\w+/g) || []).includes(value)
},
// attribute value contains string
'*=' ({ attr, value, insensitive }) {
return attr.includes(value)
},
// attribute value is equal or starts with
'|=' ({ attr, value, insensitive }) {
return attr.startsWith(`${value}-`)
},
// attribute value starts with
'^=' ({ attr, value, insensitive }) {
return attr.startsWith(value)
},
// attribute value ends with
'$=' ({ attr, value, insensitive }) {
return attr.endsWith(value)
},
}
const attributeOperator = ({ attr, value, insensitive, operator }) => {
if (typeof attr === 'number') {
attr = String(attr)
}
if (typeof attr !== 'string') {
// It's an object or an array, bail
return false
}
if (insensitive) {
attr = attr.toLowerCase()
}
return attributeOperators[operator]({
attr,
insensitive,
value,
})
}
const attributeMatch = (matcher, obj) => {
const insensitive = !!matcher.insensitive
const operator = matcher.operator || ''
const attribute = matcher.qualifiedAttribute
let value = matcher.value || ''
// return early if checking existence
if (operator === '') {
return Boolean(obj[attribute])
}
if (insensitive) {
value = value.toLowerCase()
}
// in case the current object is an array
// then we try to match every item in the array
if (Array.isArray(obj[attribute])) {
return obj[attribute].find((i, index) => {
const attr = obj[attribute][index] || ''
return attributeOperator({ attr, value, insensitive, operator })
})
} else {
const attr = obj[attribute] || ''
return attributeOperator({ attr, value, insensitive, operator })
}
}
const edgeIsType = (node, type, seen = new Set()) => {
for (const edgeIn of node.edgesIn) {
// TODO Need a test with an infinite loop
if (seen.has(edgeIn)) {
continue
}
seen.add(edgeIn)
if (edgeIn.type === type || edgeIn.from[type] || edgeIsType(edgeIn.from, type, seen)) {
return true
}
}
return false
}
const filterByType = (nodes, type) => {
const found = []
for (const node of nodes) {
if (node[type] || edgeIsType(node, type)) {
found.push(node)
}
}
return found
}
const depTypes = {
// dependency
'.prod' (prevResults) {
const found = []
for (const node of prevResults) {
if (!node.dev) {
found.push(node)
}
}
return found
},
// devDependency
'.dev' (prevResults) {
return filterByType(prevResults, 'dev')
},
// optionalDependency
'.optional' (prevResults) {
return filterByType(prevResults, 'optional')
},
// peerDependency
'.peer' (prevResults) {
return filterByType(prevResults, 'peer')
},
// workspace
'.workspace' (prevResults) {
return prevResults.filter(node => node.isWorkspace)
},
// bundledDependency
'.bundled' (prevResults) {
return prevResults.filter(node => node.inBundle)
},
}
// checks if a given node has a direct parent in any of the nodes provided in
// the compare nodes array
const hasParent = (node, compareNodes) => {
// All it takes is one so we loop and return on the first hit
for (const compareNode of compareNodes) {
// follows logical parent for link anscestors
if (node.isTop && (node.resolveParent === compareNode)) {
return true
}
// follows edges-in to check if they match a possible parent
for (const edge of node.edgesIn) {
if (edge && edge.from === compareNode) {
return true
}
}
}
return false
}
// checks if a given node is a descendant of any of the nodes provided in the
// compareNodes array
const hasAscendant = (node, compareNodes, seen = new Set()) => {
// TODO (future) loop over ancestry property
if (hasParent(node, compareNodes)) {
return true
}
if (node.isTop && node.resolveParent) {
return hasAscendant(node.resolveParent, compareNodes)
}
for (const edge of node.edgesIn) {
// TODO Need a test with an infinite loop
if (seen.has(edge)) {
continue
}
seen.add(edge)
if (edge && edge.from && hasAscendant(edge.from, compareNodes, seen)) {
return true
}
}
return false
}
const combinators = {
// direct descendant
'>' (prevResults, nextResults) {
return nextResults.filter(node => hasParent(node, prevResults))
},
// any descendant
' ' (prevResults, nextResults) {
return nextResults.filter(node => hasAscendant(node, prevResults))
},
// sibling
'~' (prevResults, nextResults) {
// Return any node in nextResults that is a sibling of (aka shares a
// parent with) a node in prevResults
const parentNodes = new Set() // Parents of everything in prevResults
for (const node of prevResults) {
for (const edge of node.edgesIn) {
// edge.from always exists cause it's from another node's edgesIn
parentNodes.add(edge.from)
}
}
return nextResults.filter(node =>
!prevResults.includes(node) && hasParent(node, [...parentNodes])
)
},
}
// get a list of available versions of a package filtered to respect --before
// NOTE: this runs over each node and should not throw
const getPackageVersions = async (name, opts) => {
let packument
try {
packument = await pacote.packument(name, {
...opts,
fullMetadata: false, // we only need the corgi
})
} catch (err) {
// if the fetch fails, log a warning and pretend there are no versions
log.warn('query', `could not retrieve packument for ${name}: ${err.message}`)
return []
}
// start with a sorted list of all versions (lowest first)
let candidates = Object.keys(packument.versions).sort(semver.compare)
// if the packument has a time property, and the user passed a before flag, then
// we filter this list down to only those versions that existed before the specified date
if (packument.time && opts.before) {
candidates = candidates.filter((version) => {
// this version isn't found in the times at all, drop it
if (!packument.time[version]) {
return false
}
return Date.parse(packument.time[version]) <= opts.before
})
}
return candidates
}
const retrieveNodesFromParsedAst = async (opts) => {
// when we first call this it's the parsed query. all other times it's
// results.currentNode.nestedNode
const rootAstNode = opts.rootAstNode
if (!rootAstNode.nodes) {
return new Set()
}
const results = new Results(opts)
const astNodeQueue = new Set()
// walk is sync, so we have to build up our async functions and then await them later
rootAstNode.walk((nextAstNode) => {
astNodeQueue.add(nextAstNode)
})
for (const nextAstNode of astNodeQueue) {
// This is the only place we reset currentAstNode
results.currentAstNode = nextAstNode
const updateFn = `${results.currentAstNode.type}Type`
if (typeof results[updateFn] !== 'function') {
throw Object.assign(
new Error(`\`${results.currentAstNode.type}\` is not a supported selector.`),
{ code: 'EQUERYNOSELECTOR' }
)
}
await results[updateFn]()
}
return results.collect(rootAstNode)
}
// We are keeping this async in the event that we do add async operators, we
// won't have to have a breaking change on this function signature.
const querySelectorAll = async (targetNode, query, flatOptions) => {
// This never changes ever we just pass it around. But we can't scope it to
// this whole file if we ever want to support concurrent calls to this
// function.
const inventory = [...targetNode.root.inventory.values()]
// res is a Set of items returned for each parsed css ast selector
const res = await retrieveNodesFromParsedAst({
initialItems: inventory,
inventory,
flatOptions,
rootAstNode: parser(query),
targetNode,
})
// returns nodes ordered by realpath
return [...res].sort((a, b) => localeCompare(a.location, b.location))
}
module.exports = querySelectorAll
|