12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706 |
- /**
- * @fileoverview This rule sets a specific indentation style and width for your code
- *
- * @author Teddy Katz
- * @author Vitaly Puzrin
- * @author Gyandeep Singh
- */
- "use strict";
- //------------------------------------------------------------------------------
- // Requirements
- //------------------------------------------------------------------------------
- const lodash = require("lodash");
- const astUtils = require("./utils/ast-utils");
- const createTree = require("functional-red-black-tree");
- //------------------------------------------------------------------------------
- // Rule Definition
- //------------------------------------------------------------------------------
- const KNOWN_NODES = new Set([
- "AssignmentExpression",
- "AssignmentPattern",
- "ArrayExpression",
- "ArrayPattern",
- "ArrowFunctionExpression",
- "AwaitExpression",
- "BlockStatement",
- "BinaryExpression",
- "BreakStatement",
- "CallExpression",
- "CatchClause",
- "ChainExpression",
- "ClassBody",
- "ClassDeclaration",
- "ClassExpression",
- "ConditionalExpression",
- "ContinueStatement",
- "DoWhileStatement",
- "DebuggerStatement",
- "EmptyStatement",
- "ExperimentalRestProperty",
- "ExperimentalSpreadProperty",
- "ExpressionStatement",
- "ForStatement",
- "ForInStatement",
- "ForOfStatement",
- "FunctionDeclaration",
- "FunctionExpression",
- "Identifier",
- "IfStatement",
- "Literal",
- "LabeledStatement",
- "LogicalExpression",
- "MemberExpression",
- "MetaProperty",
- "MethodDefinition",
- "NewExpression",
- "ObjectExpression",
- "ObjectPattern",
- "Program",
- "Property",
- "RestElement",
- "ReturnStatement",
- "SequenceExpression",
- "SpreadElement",
- "Super",
- "SwitchCase",
- "SwitchStatement",
- "TaggedTemplateExpression",
- "TemplateElement",
- "TemplateLiteral",
- "ThisExpression",
- "ThrowStatement",
- "TryStatement",
- "UnaryExpression",
- "UpdateExpression",
- "VariableDeclaration",
- "VariableDeclarator",
- "WhileStatement",
- "WithStatement",
- "YieldExpression",
- "JSXFragment",
- "JSXOpeningFragment",
- "JSXClosingFragment",
- "JSXIdentifier",
- "JSXNamespacedName",
- "JSXMemberExpression",
- "JSXEmptyExpression",
- "JSXExpressionContainer",
- "JSXElement",
- "JSXClosingElement",
- "JSXOpeningElement",
- "JSXAttribute",
- "JSXSpreadAttribute",
- "JSXText",
- "ExportDefaultDeclaration",
- "ExportNamedDeclaration",
- "ExportAllDeclaration",
- "ExportSpecifier",
- "ImportDeclaration",
- "ImportSpecifier",
- "ImportDefaultSpecifier",
- "ImportNamespaceSpecifier",
- "ImportExpression"
- ]);
- /*
- * General rule strategy:
- * 1. An OffsetStorage instance stores a map of desired offsets, where each token has a specified offset from another
- * specified token or to the first column.
- * 2. As the AST is traversed, modify the desired offsets of tokens accordingly. For example, when entering a
- * BlockStatement, offset all of the tokens in the BlockStatement by 1 indent level from the opening curly
- * brace of the BlockStatement.
- * 3. After traversing the AST, calculate the expected indentation levels of every token according to the
- * OffsetStorage container.
- * 4. For each line, compare the expected indentation of the first token to the actual indentation in the file,
- * and report the token if the two values are not equal.
- */
- /**
- * A mutable balanced binary search tree that stores (key, value) pairs. The keys are numeric, and must be unique.
- * This is intended to be a generic wrapper around a balanced binary search tree library, so that the underlying implementation
- * can easily be swapped out.
- */
- class BinarySearchTree {
- /**
- * Creates an empty tree
- */
- constructor() {
- this._rbTree = createTree();
- }
- /**
- * Inserts an entry into the tree.
- * @param {number} key The entry's key
- * @param {*} value The entry's value
- * @returns {void}
- */
- insert(key, value) {
- const iterator = this._rbTree.find(key);
- if (iterator.valid) {
- this._rbTree = iterator.update(value);
- } else {
- this._rbTree = this._rbTree.insert(key, value);
- }
- }
- /**
- * Finds the entry with the largest key less than or equal to the provided key
- * @param {number} key The provided key
- * @returns {{key: number, value: *}|null} The found entry, or null if no such entry exists.
- */
- findLe(key) {
- const iterator = this._rbTree.le(key);
- return iterator && { key: iterator.key, value: iterator.value };
- }
- /**
- * Deletes all of the keys in the interval [start, end)
- * @param {number} start The start of the range
- * @param {number} end The end of the range
- * @returns {void}
- */
- deleteRange(start, end) {
- // Exit without traversing the tree if the range has zero size.
- if (start === end) {
- return;
- }
- const iterator = this._rbTree.ge(start);
- while (iterator.valid && iterator.key < end) {
- this._rbTree = this._rbTree.remove(iterator.key);
- iterator.next();
- }
- }
- }
- /**
- * A helper class to get token-based info related to indentation
- */
- class TokenInfo {
- // eslint-disable-next-line jsdoc/require-description
- /**
- * @param {SourceCode} sourceCode A SourceCode object
- */
- constructor(sourceCode) {
- this.sourceCode = sourceCode;
- this.firstTokensByLineNumber = sourceCode.tokensAndComments.reduce((map, token) => {
- if (!map.has(token.loc.start.line)) {
- map.set(token.loc.start.line, token);
- }
- if (!map.has(token.loc.end.line) && sourceCode.text.slice(token.range[1] - token.loc.end.column, token.range[1]).trim()) {
- map.set(token.loc.end.line, token);
- }
- return map;
- }, new Map());
- }
- /**
- * Gets the first token on a given token's line
- * @param {Token|ASTNode} token a node or token
- * @returns {Token} The first token on the given line
- */
- getFirstTokenOfLine(token) {
- return this.firstTokensByLineNumber.get(token.loc.start.line);
- }
- /**
- * Determines whether a token is the first token in its line
- * @param {Token} token The token
- * @returns {boolean} `true` if the token is the first on its line
- */
- isFirstTokenOfLine(token) {
- return this.getFirstTokenOfLine(token) === token;
- }
- /**
- * Get the actual indent of a token
- * @param {Token} token Token to examine. This should be the first token on its line.
- * @returns {string} The indentation characters that precede the token
- */
- getTokenIndent(token) {
- return this.sourceCode.text.slice(token.range[0] - token.loc.start.column, token.range[0]);
- }
- }
- /**
- * A class to store information on desired offsets of tokens from each other
- */
- class OffsetStorage {
- // eslint-disable-next-line jsdoc/require-description
- /**
- * @param {TokenInfo} tokenInfo a TokenInfo instance
- * @param {number} indentSize The desired size of each indentation level
- * @param {string} indentType The indentation character
- */
- constructor(tokenInfo, indentSize, indentType) {
- this._tokenInfo = tokenInfo;
- this._indentSize = indentSize;
- this._indentType = indentType;
- this._tree = new BinarySearchTree();
- this._tree.insert(0, { offset: 0, from: null, force: false });
- this._lockedFirstTokens = new WeakMap();
- this._desiredIndentCache = new WeakMap();
- this._ignoredTokens = new WeakSet();
- }
- _getOffsetDescriptor(token) {
- return this._tree.findLe(token.range[0]).value;
- }
- /**
- * Sets the offset column of token B to match the offset column of token A.
- * **WARNING**: This matches a *column*, even if baseToken is not the first token on its line. In
- * most cases, `setDesiredOffset` should be used instead.
- * @param {Token} baseToken The first token
- * @param {Token} offsetToken The second token, whose offset should be matched to the first token
- * @returns {void}
- */
- matchOffsetOf(baseToken, offsetToken) {
- /*
- * lockedFirstTokens is a map from a token whose indentation is controlled by the "first" option to
- * the token that it depends on. For example, with the `ArrayExpression: first` option, the first
- * token of each element in the array after the first will be mapped to the first token of the first
- * element. The desired indentation of each of these tokens is computed based on the desired indentation
- * of the "first" element, rather than through the normal offset mechanism.
- */
- this._lockedFirstTokens.set(offsetToken, baseToken);
- }
- /**
- * Sets the desired offset of a token.
- *
- * This uses a line-based offset collapsing behavior to handle tokens on the same line.
- * For example, consider the following two cases:
- *
- * (
- * [
- * bar
- * ]
- * )
- *
- * ([
- * bar
- * ])
- *
- * Based on the first case, it's clear that the `bar` token needs to have an offset of 1 indent level (4 spaces) from
- * the `[` token, and the `[` token has to have an offset of 1 indent level from the `(` token. Since the `(` token is
- * the first on its line (with an indent of 0 spaces), the `bar` token needs to be offset by 2 indent levels (8 spaces)
- * from the start of its line.
- *
- * However, in the second case `bar` should only be indented by 4 spaces. This is because the offset of 1 indent level
- * between the `(` and the `[` tokens gets "collapsed" because the two tokens are on the same line. As a result, the
- * `(` token is mapped to the `[` token with an offset of 0, and the rule correctly decides that `bar` should be indented
- * by 1 indent level from the start of the line.
- *
- * This is useful because rule listeners can usually just call `setDesiredOffset` for all the tokens in the node,
- * without needing to check which lines those tokens are on.
- *
- * Note that since collapsing only occurs when two tokens are on the same line, there are a few cases where non-intuitive
- * behavior can occur. For example, consider the following cases:
- *
- * foo(
- * ).
- * bar(
- * baz
- * )
- *
- * foo(
- * ).bar(
- * baz
- * )
- *
- * Based on the first example, it would seem that `bar` should be offset by 1 indent level from `foo`, and `baz`
- * should be offset by 1 indent level from `bar`. However, this is not correct, because it would result in `baz`
- * being indented by 2 indent levels in the second case (since `foo`, `bar`, and `baz` are all on separate lines, no
- * collapsing would occur).
- *
- * Instead, the correct way would be to offset `baz` by 1 level from `bar`, offset `bar` by 1 level from the `)`, and
- * offset the `)` by 0 levels from `foo`. This ensures that the offset between `bar` and the `)` are correctly collapsed
- * in the second case.
- * @param {Token} token The token
- * @param {Token} fromToken The token that `token` should be offset from
- * @param {number} offset The desired indent level
- * @returns {void}
- */
- setDesiredOffset(token, fromToken, offset) {
- return this.setDesiredOffsets(token.range, fromToken, offset);
- }
- /**
- * Sets the desired offset of all tokens in a range
- * It's common for node listeners in this file to need to apply the same offset to a large, contiguous range of tokens.
- * Moreover, the offset of any given token is usually updated multiple times (roughly once for each node that contains
- * it). This means that the offset of each token is updated O(AST depth) times.
- * It would not be performant to store and update the offsets for each token independently, because the rule would end
- * up having a time complexity of O(number of tokens * AST depth), which is quite slow for large files.
- *
- * Instead, the offset tree is represented as a collection of contiguous offset ranges in a file. For example, the following
- * list could represent the state of the offset tree at a given point:
- *
- * * Tokens starting in the interval [0, 15) are aligned with the beginning of the file
- * * Tokens starting in the interval [15, 30) are offset by 1 indent level from the `bar` token
- * * Tokens starting in the interval [30, 43) are offset by 1 indent level from the `foo` token
- * * Tokens starting in the interval [43, 820) are offset by 2 indent levels from the `bar` token
- * * Tokens starting in the interval [820, ∞) are offset by 1 indent level from the `baz` token
- *
- * The `setDesiredOffsets` methods inserts ranges like the ones above. The third line above would be inserted by using:
- * `setDesiredOffsets([30, 43], fooToken, 1);`
- * @param {[number, number]} range A [start, end] pair. All tokens with range[0] <= token.start < range[1] will have the offset applied.
- * @param {Token} fromToken The token that this is offset from
- * @param {number} offset The desired indent level
- * @param {boolean} force `true` if this offset should not use the normal collapsing behavior. This should almost always be false.
- * @returns {void}
- */
- setDesiredOffsets(range, fromToken, offset, force) {
- /*
- * Offset ranges are stored as a collection of nodes, where each node maps a numeric key to an offset
- * descriptor. The tree for the example above would have the following nodes:
- *
- * * key: 0, value: { offset: 0, from: null }
- * * key: 15, value: { offset: 1, from: barToken }
- * * key: 30, value: { offset: 1, from: fooToken }
- * * key: 43, value: { offset: 2, from: barToken }
- * * key: 820, value: { offset: 1, from: bazToken }
- *
- * To find the offset descriptor for any given token, one needs to find the node with the largest key
- * which is <= token.start. To make this operation fast, the nodes are stored in a balanced binary
- * search tree indexed by key.
- */
- const descriptorToInsert = { offset, from: fromToken, force };
- const descriptorAfterRange = this._tree.findLe(range[1]).value;
- const fromTokenIsInRange = fromToken && fromToken.range[0] >= range[0] && fromToken.range[1] <= range[1];
- const fromTokenDescriptor = fromTokenIsInRange && this._getOffsetDescriptor(fromToken);
- // First, remove any existing nodes in the range from the tree.
- this._tree.deleteRange(range[0] + 1, range[1]);
- // Insert a new node into the tree for this range
- this._tree.insert(range[0], descriptorToInsert);
- /*
- * To avoid circular offset dependencies, keep the `fromToken` token mapped to whatever it was mapped to previously,
- * even if it's in the current range.
- */
- if (fromTokenIsInRange) {
- this._tree.insert(fromToken.range[0], fromTokenDescriptor);
- this._tree.insert(fromToken.range[1], descriptorToInsert);
- }
- /*
- * To avoid modifying the offset of tokens after the range, insert another node to keep the offset of the following
- * tokens the same as it was before.
- */
- this._tree.insert(range[1], descriptorAfterRange);
- }
- /**
- * Gets the desired indent of a token
- * @param {Token} token The token
- * @returns {string} The desired indent of the token
- */
- getDesiredIndent(token) {
- if (!this._desiredIndentCache.has(token)) {
- if (this._ignoredTokens.has(token)) {
- /*
- * If the token is ignored, use the actual indent of the token as the desired indent.
- * This ensures that no errors are reported for this token.
- */
- this._desiredIndentCache.set(
- token,
- this._tokenInfo.getTokenIndent(token)
- );
- } else if (this._lockedFirstTokens.has(token)) {
- const firstToken = this._lockedFirstTokens.get(token);
- this._desiredIndentCache.set(
- token,
- // (indentation for the first element's line)
- this.getDesiredIndent(this._tokenInfo.getFirstTokenOfLine(firstToken)) +
- // (space between the start of the first element's line and the first element)
- this._indentType.repeat(firstToken.loc.start.column - this._tokenInfo.getFirstTokenOfLine(firstToken).loc.start.column)
- );
- } else {
- const offsetInfo = this._getOffsetDescriptor(token);
- const offset = (
- offsetInfo.from &&
- offsetInfo.from.loc.start.line === token.loc.start.line &&
- !/^\s*?\n/u.test(token.value) &&
- !offsetInfo.force
- ) ? 0 : offsetInfo.offset * this._indentSize;
- this._desiredIndentCache.set(
- token,
- (offsetInfo.from ? this.getDesiredIndent(offsetInfo.from) : "") + this._indentType.repeat(offset)
- );
- }
- }
- return this._desiredIndentCache.get(token);
- }
- /**
- * Ignores a token, preventing it from being reported.
- * @param {Token} token The token
- * @returns {void}
- */
- ignoreToken(token) {
- if (this._tokenInfo.isFirstTokenOfLine(token)) {
- this._ignoredTokens.add(token);
- }
- }
- /**
- * Gets the first token that the given token's indentation is dependent on
- * @param {Token} token The token
- * @returns {Token} The token that the given token depends on, or `null` if the given token is at the top level
- */
- getFirstDependency(token) {
- return this._getOffsetDescriptor(token).from;
- }
- }
- const ELEMENT_LIST_SCHEMA = {
- oneOf: [
- {
- type: "integer",
- minimum: 0
- },
- {
- enum: ["first", "off"]
- }
- ]
- };
- module.exports = {
- meta: {
- type: "layout",
- docs: {
- description: "enforce consistent indentation",
- category: "Stylistic Issues",
- recommended: false,
- url: "https://eslint.org/docs/rules/indent"
- },
- fixable: "whitespace",
- schema: [
- {
- oneOf: [
- {
- enum: ["tab"]
- },
- {
- type: "integer",
- minimum: 0
- }
- ]
- },
- {
- type: "object",
- properties: {
- SwitchCase: {
- type: "integer",
- minimum: 0,
- default: 0
- },
- VariableDeclarator: {
- oneOf: [
- ELEMENT_LIST_SCHEMA,
- {
- type: "object",
- properties: {
- var: ELEMENT_LIST_SCHEMA,
- let: ELEMENT_LIST_SCHEMA,
- const: ELEMENT_LIST_SCHEMA
- },
- additionalProperties: false
- }
- ]
- },
- outerIIFEBody: {
- oneOf: [
- {
- type: "integer",
- minimum: 0
- },
- {
- enum: ["off"]
- }
- ]
- },
- MemberExpression: {
- oneOf: [
- {
- type: "integer",
- minimum: 0
- },
- {
- enum: ["off"]
- }
- ]
- },
- FunctionDeclaration: {
- type: "object",
- properties: {
- parameters: ELEMENT_LIST_SCHEMA,
- body: {
- type: "integer",
- minimum: 0
- }
- },
- additionalProperties: false
- },
- FunctionExpression: {
- type: "object",
- properties: {
- parameters: ELEMENT_LIST_SCHEMA,
- body: {
- type: "integer",
- minimum: 0
- }
- },
- additionalProperties: false
- },
- CallExpression: {
- type: "object",
- properties: {
- arguments: ELEMENT_LIST_SCHEMA
- },
- additionalProperties: false
- },
- ArrayExpression: ELEMENT_LIST_SCHEMA,
- ObjectExpression: ELEMENT_LIST_SCHEMA,
- ImportDeclaration: ELEMENT_LIST_SCHEMA,
- flatTernaryExpressions: {
- type: "boolean",
- default: false
- },
- offsetTernaryExpressions: {
- type: "boolean",
- default: false
- },
- ignoredNodes: {
- type: "array",
- items: {
- type: "string",
- not: {
- pattern: ":exit$"
- }
- }
- },
- ignoreComments: {
- type: "boolean",
- default: false
- }
- },
- additionalProperties: false
- }
- ],
- messages: {
- wrongIndentation: "Expected indentation of {{expected}} but found {{actual}}."
- }
- },
- create(context) {
- const DEFAULT_VARIABLE_INDENT = 1;
- const DEFAULT_PARAMETER_INDENT = 1;
- const DEFAULT_FUNCTION_BODY_INDENT = 1;
- let indentType = "space";
- let indentSize = 4;
- const options = {
- SwitchCase: 0,
- VariableDeclarator: {
- var: DEFAULT_VARIABLE_INDENT,
- let: DEFAULT_VARIABLE_INDENT,
- const: DEFAULT_VARIABLE_INDENT
- },
- outerIIFEBody: 1,
- FunctionDeclaration: {
- parameters: DEFAULT_PARAMETER_INDENT,
- body: DEFAULT_FUNCTION_BODY_INDENT
- },
- FunctionExpression: {
- parameters: DEFAULT_PARAMETER_INDENT,
- body: DEFAULT_FUNCTION_BODY_INDENT
- },
- CallExpression: {
- arguments: DEFAULT_PARAMETER_INDENT
- },
- MemberExpression: 1,
- ArrayExpression: 1,
- ObjectExpression: 1,
- ImportDeclaration: 1,
- flatTernaryExpressions: false,
- ignoredNodes: [],
- ignoreComments: false
- };
- if (context.options.length) {
- if (context.options[0] === "tab") {
- indentSize = 1;
- indentType = "tab";
- } else {
- indentSize = context.options[0];
- indentType = "space";
- }
- if (context.options[1]) {
- Object.assign(options, context.options[1]);
- if (typeof options.VariableDeclarator === "number" || options.VariableDeclarator === "first") {
- options.VariableDeclarator = {
- var: options.VariableDeclarator,
- let: options.VariableDeclarator,
- const: options.VariableDeclarator
- };
- }
- }
- }
- const sourceCode = context.getSourceCode();
- const tokenInfo = new TokenInfo(sourceCode);
- const offsets = new OffsetStorage(tokenInfo, indentSize, indentType === "space" ? " " : "\t");
- const parameterParens = new WeakSet();
- /**
- * Creates an error message for a line, given the expected/actual indentation.
- * @param {int} expectedAmount The expected amount of indentation characters for this line
- * @param {int} actualSpaces The actual number of indentation spaces that were found on this line
- * @param {int} actualTabs The actual number of indentation tabs that were found on this line
- * @returns {string} An error message for this line
- */
- function createErrorMessageData(expectedAmount, actualSpaces, actualTabs) {
- const expectedStatement = `${expectedAmount} ${indentType}${expectedAmount === 1 ? "" : "s"}`; // e.g. "2 tabs"
- const foundSpacesWord = `space${actualSpaces === 1 ? "" : "s"}`; // e.g. "space"
- const foundTabsWord = `tab${actualTabs === 1 ? "" : "s"}`; // e.g. "tabs"
- let foundStatement;
- if (actualSpaces > 0) {
- /*
- * Abbreviate the message if the expected indentation is also spaces.
- * e.g. 'Expected 4 spaces but found 2' rather than 'Expected 4 spaces but found 2 spaces'
- */
- foundStatement = indentType === "space" ? actualSpaces : `${actualSpaces} ${foundSpacesWord}`;
- } else if (actualTabs > 0) {
- foundStatement = indentType === "tab" ? actualTabs : `${actualTabs} ${foundTabsWord}`;
- } else {
- foundStatement = "0";
- }
- return {
- expected: expectedStatement,
- actual: foundStatement
- };
- }
- /**
- * Reports a given indent violation
- * @param {Token} token Token violating the indent rule
- * @param {string} neededIndent Expected indentation string
- * @returns {void}
- */
- function report(token, neededIndent) {
- const actualIndent = Array.from(tokenInfo.getTokenIndent(token));
- const numSpaces = actualIndent.filter(char => char === " ").length;
- const numTabs = actualIndent.filter(char => char === "\t").length;
- context.report({
- node: token,
- messageId: "wrongIndentation",
- data: createErrorMessageData(neededIndent.length, numSpaces, numTabs),
- loc: {
- start: { line: token.loc.start.line, column: 0 },
- end: { line: token.loc.start.line, column: token.loc.start.column }
- },
- fix(fixer) {
- const range = [token.range[0] - token.loc.start.column, token.range[0]];
- const newText = neededIndent;
- return fixer.replaceTextRange(range, newText);
- }
- });
- }
- /**
- * Checks if a token's indentation is correct
- * @param {Token} token Token to examine
- * @param {string} desiredIndent Desired indentation of the string
- * @returns {boolean} `true` if the token's indentation is correct
- */
- function validateTokenIndent(token, desiredIndent) {
- const indentation = tokenInfo.getTokenIndent(token);
- return indentation === desiredIndent ||
- // To avoid conflicts with no-mixed-spaces-and-tabs, don't report mixed spaces and tabs.
- indentation.includes(" ") && indentation.includes("\t");
- }
- /**
- * Check to see if the node is a file level IIFE
- * @param {ASTNode} node The function node to check.
- * @returns {boolean} True if the node is the outer IIFE
- */
- function isOuterIIFE(node) {
- /*
- * Verify that the node is an IIFE
- */
- if (!node.parent || node.parent.type !== "CallExpression" || node.parent.callee !== node) {
- return false;
- }
- /*
- * Navigate legal ancestors to determine whether this IIFE is outer.
- * A "legal ancestor" is an expression or statement that causes the function to get executed immediately.
- * For example, `!(function(){})()` is an outer IIFE even though it is preceded by a ! operator.
- */
- let statement = node.parent && node.parent.parent;
- while (
- statement.type === "UnaryExpression" && ["!", "~", "+", "-"].indexOf(statement.operator) > -1 ||
- statement.type === "AssignmentExpression" ||
- statement.type === "LogicalExpression" ||
- statement.type === "SequenceExpression" ||
- statement.type === "VariableDeclarator"
- ) {
- statement = statement.parent;
- }
- return (statement.type === "ExpressionStatement" || statement.type === "VariableDeclaration") && statement.parent.type === "Program";
- }
- /**
- * Counts the number of linebreaks that follow the last non-whitespace character in a string
- * @param {string} string The string to check
- * @returns {number} The number of JavaScript linebreaks that follow the last non-whitespace character,
- * or the total number of linebreaks if the string is all whitespace.
- */
- function countTrailingLinebreaks(string) {
- const trailingWhitespace = string.match(/\s*$/u)[0];
- const linebreakMatches = trailingWhitespace.match(astUtils.createGlobalLinebreakMatcher());
- return linebreakMatches === null ? 0 : linebreakMatches.length;
- }
- /**
- * Check indentation for lists of elements (arrays, objects, function params)
- * @param {ASTNode[]} elements List of elements that should be offset
- * @param {Token} startToken The start token of the list that element should be aligned against, e.g. '['
- * @param {Token} endToken The end token of the list, e.g. ']'
- * @param {number|string} offset The amount that the elements should be offset
- * @returns {void}
- */
- function addElementListIndent(elements, startToken, endToken, offset) {
- /**
- * Gets the first token of a given element, including surrounding parentheses.
- * @param {ASTNode} element A node in the `elements` list
- * @returns {Token} The first token of this element
- */
- function getFirstToken(element) {
- let token = sourceCode.getTokenBefore(element);
- while (astUtils.isOpeningParenToken(token) && token !== startToken) {
- token = sourceCode.getTokenBefore(token);
- }
- return sourceCode.getTokenAfter(token);
- }
- // Run through all the tokens in the list, and offset them by one indent level (mainly for comments, other things will end up overridden)
- offsets.setDesiredOffsets(
- [startToken.range[1], endToken.range[0]],
- startToken,
- typeof offset === "number" ? offset : 1
- );
- offsets.setDesiredOffset(endToken, startToken, 0);
- // If the preference is "first" but there is no first element (e.g. sparse arrays w/ empty first slot), fall back to 1 level.
- if (offset === "first" && elements.length && !elements[0]) {
- return;
- }
- elements.forEach((element, index) => {
- if (!element) {
- // Skip holes in arrays
- return;
- }
- if (offset === "off") {
- // Ignore the first token of every element if the "off" option is used
- offsets.ignoreToken(getFirstToken(element));
- }
- // Offset the following elements correctly relative to the first element
- if (index === 0) {
- return;
- }
- if (offset === "first" && tokenInfo.isFirstTokenOfLine(getFirstToken(element))) {
- offsets.matchOffsetOf(getFirstToken(elements[0]), getFirstToken(element));
- } else {
- const previousElement = elements[index - 1];
- const firstTokenOfPreviousElement = previousElement && getFirstToken(previousElement);
- const previousElementLastToken = previousElement && sourceCode.getLastToken(previousElement);
- if (
- previousElement &&
- previousElementLastToken.loc.end.line - countTrailingLinebreaks(previousElementLastToken.value) > startToken.loc.end.line
- ) {
- offsets.setDesiredOffsets(
- [previousElement.range[1], element.range[1]],
- firstTokenOfPreviousElement,
- 0
- );
- }
- }
- });
- }
- /**
- * Check and decide whether to check for indentation for blockless nodes
- * Scenarios are for or while statements without braces around them
- * @param {ASTNode} node node to examine
- * @returns {void}
- */
- function addBlocklessNodeIndent(node) {
- if (node.type !== "BlockStatement") {
- const lastParentToken = sourceCode.getTokenBefore(node, astUtils.isNotOpeningParenToken);
- let firstBodyToken = sourceCode.getFirstToken(node);
- let lastBodyToken = sourceCode.getLastToken(node);
- while (
- astUtils.isOpeningParenToken(sourceCode.getTokenBefore(firstBodyToken)) &&
- astUtils.isClosingParenToken(sourceCode.getTokenAfter(lastBodyToken))
- ) {
- firstBodyToken = sourceCode.getTokenBefore(firstBodyToken);
- lastBodyToken = sourceCode.getTokenAfter(lastBodyToken);
- }
- offsets.setDesiredOffsets([firstBodyToken.range[0], lastBodyToken.range[1]], lastParentToken, 1);
- /*
- * For blockless nodes with semicolon-first style, don't indent the semicolon.
- * e.g.
- * if (foo) bar()
- * ; [1, 2, 3].map(foo)
- */
- const lastToken = sourceCode.getLastToken(node);
- if (node.type !== "EmptyStatement" && astUtils.isSemicolonToken(lastToken)) {
- offsets.setDesiredOffset(lastToken, lastParentToken, 0);
- }
- }
- }
- /**
- * Checks the indentation for nodes that are like function calls (`CallExpression` and `NewExpression`)
- * @param {ASTNode} node A CallExpression or NewExpression node
- * @returns {void}
- */
- function addFunctionCallIndent(node) {
- let openingParen;
- if (node.arguments.length) {
- openingParen = sourceCode.getFirstTokenBetween(node.callee, node.arguments[0], astUtils.isOpeningParenToken);
- } else {
- openingParen = sourceCode.getLastToken(node, 1);
- }
- const closingParen = sourceCode.getLastToken(node);
- parameterParens.add(openingParen);
- parameterParens.add(closingParen);
- /*
- * If `?.` token exists, set desired offset for that.
- * This logic is copied from `MemberExpression`'s.
- */
- if (node.optional) {
- const dotToken = sourceCode.getTokenAfter(node.callee, astUtils.isQuestionDotToken);
- const calleeParenCount = sourceCode.getTokensBetween(node.callee, dotToken, { filter: astUtils.isClosingParenToken }).length;
- const firstTokenOfCallee = calleeParenCount
- ? sourceCode.getTokenBefore(node.callee, { skip: calleeParenCount - 1 })
- : sourceCode.getFirstToken(node.callee);
- const lastTokenOfCallee = sourceCode.getTokenBefore(dotToken);
- const offsetBase = lastTokenOfCallee.loc.end.line === openingParen.loc.start.line
- ? lastTokenOfCallee
- : firstTokenOfCallee;
- offsets.setDesiredOffset(dotToken, offsetBase, 1);
- }
- const offsetAfterToken = node.callee.type === "TaggedTemplateExpression" ? sourceCode.getFirstToken(node.callee.quasi) : openingParen;
- const offsetToken = sourceCode.getTokenBefore(offsetAfterToken);
- offsets.setDesiredOffset(openingParen, offsetToken, 0);
- addElementListIndent(node.arguments, openingParen, closingParen, options.CallExpression.arguments);
- }
- /**
- * Checks the indentation of parenthesized values, given a list of tokens in a program
- * @param {Token[]} tokens A list of tokens
- * @returns {void}
- */
- function addParensIndent(tokens) {
- const parenStack = [];
- const parenPairs = [];
- tokens.forEach(nextToken => {
- // Accumulate a list of parenthesis pairs
- if (astUtils.isOpeningParenToken(nextToken)) {
- parenStack.push(nextToken);
- } else if (astUtils.isClosingParenToken(nextToken)) {
- parenPairs.unshift({ left: parenStack.pop(), right: nextToken });
- }
- });
- parenPairs.forEach(pair => {
- const leftParen = pair.left;
- const rightParen = pair.right;
- // We only want to handle parens around expressions, so exclude parentheses that are in function parameters and function call arguments.
- if (!parameterParens.has(leftParen) && !parameterParens.has(rightParen)) {
- const parenthesizedTokens = new Set(sourceCode.getTokensBetween(leftParen, rightParen));
- parenthesizedTokens.forEach(token => {
- if (!parenthesizedTokens.has(offsets.getFirstDependency(token))) {
- offsets.setDesiredOffset(token, leftParen, 1);
- }
- });
- }
- offsets.setDesiredOffset(rightParen, leftParen, 0);
- });
- }
- /**
- * Ignore all tokens within an unknown node whose offset do not depend
- * on another token's offset within the unknown node
- * @param {ASTNode} node Unknown Node
- * @returns {void}
- */
- function ignoreNode(node) {
- const unknownNodeTokens = new Set(sourceCode.getTokens(node, { includeComments: true }));
- unknownNodeTokens.forEach(token => {
- if (!unknownNodeTokens.has(offsets.getFirstDependency(token))) {
- const firstTokenOfLine = tokenInfo.getFirstTokenOfLine(token);
- if (token === firstTokenOfLine) {
- offsets.ignoreToken(token);
- } else {
- offsets.setDesiredOffset(token, firstTokenOfLine, 0);
- }
- }
- });
- }
- /**
- * Check whether the given token is on the first line of a statement.
- * @param {Token} token The token to check.
- * @param {ASTNode} leafNode The expression node that the token belongs directly.
- * @returns {boolean} `true` if the token is on the first line of a statement.
- */
- function isOnFirstLineOfStatement(token, leafNode) {
- let node = leafNode;
- while (node.parent && !node.parent.type.endsWith("Statement") && !node.parent.type.endsWith("Declaration")) {
- node = node.parent;
- }
- node = node.parent;
- return !node || node.loc.start.line === token.loc.start.line;
- }
- /**
- * Check whether there are any blank (whitespace-only) lines between
- * two tokens on separate lines.
- * @param {Token} firstToken The first token.
- * @param {Token} secondToken The second token.
- * @returns {boolean} `true` if the tokens are on separate lines and
- * there exists a blank line between them, `false` otherwise.
- */
- function hasBlankLinesBetween(firstToken, secondToken) {
- const firstTokenLine = firstToken.loc.end.line;
- const secondTokenLine = secondToken.loc.start.line;
- if (firstTokenLine === secondTokenLine || firstTokenLine === secondTokenLine - 1) {
- return false;
- }
- for (let line = firstTokenLine + 1; line < secondTokenLine; ++line) {
- if (!tokenInfo.firstTokensByLineNumber.has(line)) {
- return true;
- }
- }
- return false;
- }
- const ignoredNodeFirstTokens = new Set();
- const baseOffsetListeners = {
- "ArrayExpression, ArrayPattern"(node) {
- const openingBracket = sourceCode.getFirstToken(node);
- const closingBracket = sourceCode.getTokenAfter(lodash.findLast(node.elements) || openingBracket, astUtils.isClosingBracketToken);
- addElementListIndent(node.elements, openingBracket, closingBracket, options.ArrayExpression);
- },
- "ObjectExpression, ObjectPattern"(node) {
- const openingCurly = sourceCode.getFirstToken(node);
- const closingCurly = sourceCode.getTokenAfter(
- node.properties.length ? node.properties[node.properties.length - 1] : openingCurly,
- astUtils.isClosingBraceToken
- );
- addElementListIndent(node.properties, openingCurly, closingCurly, options.ObjectExpression);
- },
- ArrowFunctionExpression(node) {
- const maybeOpeningParen = sourceCode.getFirstToken(node, { skip: node.async ? 1 : 0 });
- if (astUtils.isOpeningParenToken(maybeOpeningParen)) {
- const openingParen = maybeOpeningParen;
- const closingParen = sourceCode.getTokenBefore(node.body, astUtils.isClosingParenToken);
- parameterParens.add(openingParen);
- parameterParens.add(closingParen);
- addElementListIndent(node.params, openingParen, closingParen, options.FunctionExpression.parameters);
- }
- addBlocklessNodeIndent(node.body);
- },
- AssignmentExpression(node) {
- const operator = sourceCode.getFirstTokenBetween(node.left, node.right, token => token.value === node.operator);
- offsets.setDesiredOffsets([operator.range[0], node.range[1]], sourceCode.getLastToken(node.left), 1);
- offsets.ignoreToken(operator);
- offsets.ignoreToken(sourceCode.getTokenAfter(operator));
- },
- "BinaryExpression, LogicalExpression"(node) {
- const operator = sourceCode.getFirstTokenBetween(node.left, node.right, token => token.value === node.operator);
- /*
- * For backwards compatibility, don't check BinaryExpression indents, e.g.
- * var foo = bar &&
- * baz;
- */
- const tokenAfterOperator = sourceCode.getTokenAfter(operator);
- offsets.ignoreToken(operator);
- offsets.ignoreToken(tokenAfterOperator);
- offsets.setDesiredOffset(tokenAfterOperator, operator, 0);
- },
- "BlockStatement, ClassBody"(node) {
- let blockIndentLevel;
- if (node.parent && isOuterIIFE(node.parent)) {
- blockIndentLevel = options.outerIIFEBody;
- } else if (node.parent && (node.parent.type === "FunctionExpression" || node.parent.type === "ArrowFunctionExpression")) {
- blockIndentLevel = options.FunctionExpression.body;
- } else if (node.parent && node.parent.type === "FunctionDeclaration") {
- blockIndentLevel = options.FunctionDeclaration.body;
- } else {
- blockIndentLevel = 1;
- }
- /*
- * For blocks that aren't lone statements, ensure that the opening curly brace
- * is aligned with the parent.
- */
- if (!astUtils.STATEMENT_LIST_PARENTS.has(node.parent.type)) {
- offsets.setDesiredOffset(sourceCode.getFirstToken(node), sourceCode.getFirstToken(node.parent), 0);
- }
- addElementListIndent(node.body, sourceCode.getFirstToken(node), sourceCode.getLastToken(node), blockIndentLevel);
- },
- CallExpression: addFunctionCallIndent,
- "ClassDeclaration[superClass], ClassExpression[superClass]"(node) {
- const classToken = sourceCode.getFirstToken(node);
- const extendsToken = sourceCode.getTokenBefore(node.superClass, astUtils.isNotOpeningParenToken);
- offsets.setDesiredOffsets([extendsToken.range[0], node.body.range[0]], classToken, 1);
- },
- ConditionalExpression(node) {
- const firstToken = sourceCode.getFirstToken(node);
- // `flatTernaryExpressions` option is for the following style:
- // var a =
- // foo > 0 ? bar :
- // foo < 0 ? baz :
- // /*else*/ qiz ;
- if (!options.flatTernaryExpressions ||
- !astUtils.isTokenOnSameLine(node.test, node.consequent) ||
- isOnFirstLineOfStatement(firstToken, node)
- ) {
- const questionMarkToken = sourceCode.getFirstTokenBetween(node.test, node.consequent, token => token.type === "Punctuator" && token.value === "?");
- const colonToken = sourceCode.getFirstTokenBetween(node.consequent, node.alternate, token => token.type === "Punctuator" && token.value === ":");
- const firstConsequentToken = sourceCode.getTokenAfter(questionMarkToken);
- const lastConsequentToken = sourceCode.getTokenBefore(colonToken);
- const firstAlternateToken = sourceCode.getTokenAfter(colonToken);
- offsets.setDesiredOffset(questionMarkToken, firstToken, 1);
- offsets.setDesiredOffset(colonToken, firstToken, 1);
- offsets.setDesiredOffset(firstConsequentToken, firstToken,
- options.offsetTernaryExpressions ? 2 : 1);
- /*
- * The alternate and the consequent should usually have the same indentation.
- * If they share part of a line, align the alternate against the first token of the consequent.
- * This allows the alternate to be indented correctly in cases like this:
- * foo ? (
- * bar
- * ) : ( // this '(' is aligned with the '(' above, so it's considered to be aligned with `foo`
- * baz // as a result, `baz` is offset by 1 rather than 2
- * )
- */
- if (lastConsequentToken.loc.end.line === firstAlternateToken.loc.start.line) {
- offsets.setDesiredOffset(firstAlternateToken, firstConsequentToken, 0);
- } else {
- /**
- * If the alternate and consequent do not share part of a line, offset the alternate from the first
- * token of the conditional expression. For example:
- * foo ? bar
- * : baz
- *
- * If `baz` were aligned with `bar` rather than being offset by 1 from `foo`, `baz` would end up
- * having no expected indentation.
- */
- offsets.setDesiredOffset(firstAlternateToken, firstToken,
- firstAlternateToken.type === "Punctuator" &&
- options.offsetTernaryExpressions ? 2 : 1);
- }
- }
- },
- "DoWhileStatement, WhileStatement, ForInStatement, ForOfStatement": node => addBlocklessNodeIndent(node.body),
- ExportNamedDeclaration(node) {
- if (node.declaration === null) {
- const closingCurly = sourceCode.getLastToken(node, astUtils.isClosingBraceToken);
- // Indent the specifiers in `export {foo, bar, baz}`
- addElementListIndent(node.specifiers, sourceCode.getFirstToken(node, { skip: 1 }), closingCurly, 1);
- if (node.source) {
- // Indent everything after and including the `from` token in `export {foo, bar, baz} from 'qux'`
- offsets.setDesiredOffsets([closingCurly.range[1], node.range[1]], sourceCode.getFirstToken(node), 1);
- }
- }
- },
- ForStatement(node) {
- const forOpeningParen = sourceCode.getFirstToken(node, 1);
- if (node.init) {
- offsets.setDesiredOffsets(node.init.range, forOpeningParen, 1);
- }
- if (node.test) {
- offsets.setDesiredOffsets(node.test.range, forOpeningParen, 1);
- }
- if (node.update) {
- offsets.setDesiredOffsets(node.update.range, forOpeningParen, 1);
- }
- addBlocklessNodeIndent(node.body);
- },
- "FunctionDeclaration, FunctionExpression"(node) {
- const closingParen = sourceCode.getTokenBefore(node.body);
- const openingParen = sourceCode.getTokenBefore(node.params.length ? node.params[0] : closingParen);
- parameterParens.add(openingParen);
- parameterParens.add(closingParen);
- addElementListIndent(node.params, openingParen, closingParen, options[node.type].parameters);
- },
- IfStatement(node) {
- addBlocklessNodeIndent(node.consequent);
- if (node.alternate && node.alternate.type !== "IfStatement") {
- addBlocklessNodeIndent(node.alternate);
- }
- },
- ImportDeclaration(node) {
- if (node.specifiers.some(specifier => specifier.type === "ImportSpecifier")) {
- const openingCurly = sourceCode.getFirstToken(node, astUtils.isOpeningBraceToken);
- const closingCurly = sourceCode.getLastToken(node, astUtils.isClosingBraceToken);
- addElementListIndent(node.specifiers.filter(specifier => specifier.type === "ImportSpecifier"), openingCurly, closingCurly, options.ImportDeclaration);
- }
- const fromToken = sourceCode.getLastToken(node, token => token.type === "Identifier" && token.value === "from");
- const sourceToken = sourceCode.getLastToken(node, token => token.type === "String");
- const semiToken = sourceCode.getLastToken(node, token => token.type === "Punctuator" && token.value === ";");
- if (fromToken) {
- const end = semiToken && semiToken.range[1] === sourceToken.range[1] ? node.range[1] : sourceToken.range[1];
- offsets.setDesiredOffsets([fromToken.range[0], end], sourceCode.getFirstToken(node), 1);
- }
- },
- ImportExpression(node) {
- const openingParen = sourceCode.getFirstToken(node, 1);
- const closingParen = sourceCode.getLastToken(node);
- parameterParens.add(openingParen);
- parameterParens.add(closingParen);
- offsets.setDesiredOffset(openingParen, sourceCode.getTokenBefore(openingParen), 0);
- addElementListIndent([node.source], openingParen, closingParen, options.CallExpression.arguments);
- },
- "MemberExpression, JSXMemberExpression, MetaProperty"(node) {
- const object = node.type === "MetaProperty" ? node.meta : node.object;
- const firstNonObjectToken = sourceCode.getFirstTokenBetween(object, node.property, astUtils.isNotClosingParenToken);
- const secondNonObjectToken = sourceCode.getTokenAfter(firstNonObjectToken);
- const objectParenCount = sourceCode.getTokensBetween(object, node.property, { filter: astUtils.isClosingParenToken }).length;
- const firstObjectToken = objectParenCount
- ? sourceCode.getTokenBefore(object, { skip: objectParenCount - 1 })
- : sourceCode.getFirstToken(object);
- const lastObjectToken = sourceCode.getTokenBefore(firstNonObjectToken);
- const firstPropertyToken = node.computed ? firstNonObjectToken : secondNonObjectToken;
- if (node.computed) {
- // For computed MemberExpressions, match the closing bracket with the opening bracket.
- offsets.setDesiredOffset(sourceCode.getLastToken(node), firstNonObjectToken, 0);
- offsets.setDesiredOffsets(node.property.range, firstNonObjectToken, 1);
- }
- /*
- * If the object ends on the same line that the property starts, match against the last token
- * of the object, to ensure that the MemberExpression is not indented.
- *
- * Otherwise, match against the first token of the object, e.g.
- * foo
- * .bar
- * .baz // <-- offset by 1 from `foo`
- */
- const offsetBase = lastObjectToken.loc.end.line === firstPropertyToken.loc.start.line
- ? lastObjectToken
- : firstObjectToken;
- if (typeof options.MemberExpression === "number") {
- // Match the dot (for non-computed properties) or the opening bracket (for computed properties) against the object.
- offsets.setDesiredOffset(firstNonObjectToken, offsetBase, options.MemberExpression);
- /*
- * For computed MemberExpressions, match the first token of the property against the opening bracket.
- * Otherwise, match the first token of the property against the object.
- */
- offsets.setDesiredOffset(secondNonObjectToken, node.computed ? firstNonObjectToken : offsetBase, options.MemberExpression);
- } else {
- // If the MemberExpression option is off, ignore the dot and the first token of the property.
- offsets.ignoreToken(firstNonObjectToken);
- offsets.ignoreToken(secondNonObjectToken);
- // To ignore the property indentation, ensure that the property tokens depend on the ignored tokens.
- offsets.setDesiredOffset(firstNonObjectToken, offsetBase, 0);
- offsets.setDesiredOffset(secondNonObjectToken, firstNonObjectToken, 0);
- }
- },
- NewExpression(node) {
- // Only indent the arguments if the NewExpression has parens (e.g. `new Foo(bar)` or `new Foo()`, but not `new Foo`
- if (node.arguments.length > 0 ||
- astUtils.isClosingParenToken(sourceCode.getLastToken(node)) &&
- astUtils.isOpeningParenToken(sourceCode.getLastToken(node, 1))) {
- addFunctionCallIndent(node);
- }
- },
- Property(node) {
- if (!node.shorthand && !node.method && node.kind === "init") {
- const colon = sourceCode.getFirstTokenBetween(node.key, node.value, astUtils.isColonToken);
- offsets.ignoreToken(sourceCode.getTokenAfter(colon));
- }
- },
- SwitchStatement(node) {
- const openingCurly = sourceCode.getTokenAfter(node.discriminant, astUtils.isOpeningBraceToken);
- const closingCurly = sourceCode.getLastToken(node);
- offsets.setDesiredOffsets([openingCurly.range[1], closingCurly.range[0]], openingCurly, options.SwitchCase);
- if (node.cases.length) {
- sourceCode.getTokensBetween(
- node.cases[node.cases.length - 1],
- closingCurly,
- { includeComments: true, filter: astUtils.isCommentToken }
- ).forEach(token => offsets.ignoreToken(token));
- }
- },
- SwitchCase(node) {
- if (!(node.consequent.length === 1 && node.consequent[0].type === "BlockStatement")) {
- const caseKeyword = sourceCode.getFirstToken(node);
- const tokenAfterCurrentCase = sourceCode.getTokenAfter(node);
- offsets.setDesiredOffsets([caseKeyword.range[1], tokenAfterCurrentCase.range[0]], caseKeyword, 1);
- }
- },
- TemplateLiteral(node) {
- node.expressions.forEach((expression, index) => {
- const previousQuasi = node.quasis[index];
- const nextQuasi = node.quasis[index + 1];
- const tokenToAlignFrom = previousQuasi.loc.start.line === previousQuasi.loc.end.line
- ? sourceCode.getFirstToken(previousQuasi)
- : null;
- offsets.setDesiredOffsets([previousQuasi.range[1], nextQuasi.range[0]], tokenToAlignFrom, 1);
- offsets.setDesiredOffset(sourceCode.getFirstToken(nextQuasi), tokenToAlignFrom, 0);
- });
- },
- VariableDeclaration(node) {
- let variableIndent = Object.prototype.hasOwnProperty.call(options.VariableDeclarator, node.kind)
- ? options.VariableDeclarator[node.kind]
- : DEFAULT_VARIABLE_INDENT;
- const firstToken = sourceCode.getFirstToken(node),
- lastToken = sourceCode.getLastToken(node);
- if (options.VariableDeclarator[node.kind] === "first") {
- if (node.declarations.length > 1) {
- addElementListIndent(
- node.declarations,
- firstToken,
- lastToken,
- "first"
- );
- return;
- }
- variableIndent = DEFAULT_VARIABLE_INDENT;
- }
- if (node.declarations[node.declarations.length - 1].loc.start.line > node.loc.start.line) {
- /*
- * VariableDeclarator indentation is a bit different from other forms of indentation, in that the
- * indentation of an opening bracket sometimes won't match that of a closing bracket. For example,
- * the following indentations are correct:
- *
- * var foo = {
- * ok: true
- * };
- *
- * var foo = {
- * ok: true,
- * },
- * bar = 1;
- *
- * Account for when exiting the AST (after indentations have already been set for the nodes in
- * the declaration) by manually increasing the indentation level of the tokens in this declarator
- * on the same line as the start of the declaration, provided that there are declarators that
- * follow this one.
- */
- offsets.setDesiredOffsets(node.range, firstToken, variableIndent, true);
- } else {
- offsets.setDesiredOffsets(node.range, firstToken, variableIndent);
- }
- if (astUtils.isSemicolonToken(lastToken)) {
- offsets.ignoreToken(lastToken);
- }
- },
- VariableDeclarator(node) {
- if (node.init) {
- const equalOperator = sourceCode.getTokenBefore(node.init, astUtils.isNotOpeningParenToken);
- const tokenAfterOperator = sourceCode.getTokenAfter(equalOperator);
- offsets.ignoreToken(equalOperator);
- offsets.ignoreToken(tokenAfterOperator);
- offsets.setDesiredOffsets([tokenAfterOperator.range[0], node.range[1]], equalOperator, 1);
- offsets.setDesiredOffset(equalOperator, sourceCode.getLastToken(node.id), 0);
- }
- },
- "JSXAttribute[value]"(node) {
- const equalsToken = sourceCode.getFirstTokenBetween(node.name, node.value, token => token.type === "Punctuator" && token.value === "=");
- offsets.setDesiredOffsets([equalsToken.range[0], node.value.range[1]], sourceCode.getFirstToken(node.name), 1);
- },
- JSXElement(node) {
- if (node.closingElement) {
- addElementListIndent(node.children, sourceCode.getFirstToken(node.openingElement), sourceCode.getFirstToken(node.closingElement), 1);
- }
- },
- JSXOpeningElement(node) {
- const firstToken = sourceCode.getFirstToken(node);
- let closingToken;
- if (node.selfClosing) {
- closingToken = sourceCode.getLastToken(node, { skip: 1 });
- offsets.setDesiredOffset(sourceCode.getLastToken(node), closingToken, 0);
- } else {
- closingToken = sourceCode.getLastToken(node);
- }
- offsets.setDesiredOffsets(node.name.range, sourceCode.getFirstToken(node));
- addElementListIndent(node.attributes, firstToken, closingToken, 1);
- },
- JSXClosingElement(node) {
- const firstToken = sourceCode.getFirstToken(node);
- offsets.setDesiredOffsets(node.name.range, firstToken, 1);
- },
- JSXFragment(node) {
- const firstOpeningToken = sourceCode.getFirstToken(node.openingFragment);
- const firstClosingToken = sourceCode.getFirstToken(node.closingFragment);
- addElementListIndent(node.children, firstOpeningToken, firstClosingToken, 1);
- },
- JSXOpeningFragment(node) {
- const firstToken = sourceCode.getFirstToken(node);
- const closingToken = sourceCode.getLastToken(node);
- offsets.setDesiredOffsets(node.range, firstToken, 1);
- offsets.matchOffsetOf(firstToken, closingToken);
- },
- JSXClosingFragment(node) {
- const firstToken = sourceCode.getFirstToken(node);
- const slashToken = sourceCode.getLastToken(node, { skip: 1 });
- const closingToken = sourceCode.getLastToken(node);
- const tokenToMatch = astUtils.isTokenOnSameLine(slashToken, closingToken) ? slashToken : closingToken;
- offsets.setDesiredOffsets(node.range, firstToken, 1);
- offsets.matchOffsetOf(firstToken, tokenToMatch);
- },
- JSXExpressionContainer(node) {
- const openingCurly = sourceCode.getFirstToken(node);
- const closingCurly = sourceCode.getLastToken(node);
- offsets.setDesiredOffsets(
- [openingCurly.range[1], closingCurly.range[0]],
- openingCurly,
- 1
- );
- },
- JSXSpreadAttribute(node) {
- const openingCurly = sourceCode.getFirstToken(node);
- const closingCurly = sourceCode.getLastToken(node);
- offsets.setDesiredOffsets(
- [openingCurly.range[1], closingCurly.range[0]],
- openingCurly,
- 1
- );
- },
- "*"(node) {
- const firstToken = sourceCode.getFirstToken(node);
- // Ensure that the children of every node are indented at least as much as the first token.
- if (firstToken && !ignoredNodeFirstTokens.has(firstToken)) {
- offsets.setDesiredOffsets(node.range, firstToken, 0);
- }
- }
- };
- const listenerCallQueue = [];
- /*
- * To ignore the indentation of a node:
- * 1. Don't call the node's listener when entering it (if it has a listener)
- * 2. Don't set any offsets against the first token of the node.
- * 3. Call `ignoreNode` on the node sometime after exiting it and before validating offsets.
- */
- const offsetListeners = lodash.mapValues(
- baseOffsetListeners,
- /*
- * Offset listener calls are deferred until traversal is finished, and are called as
- * part of the final `Program:exit` listener. This is necessary because a node might
- * be matched by multiple selectors.
- *
- * Example: Suppose there is an offset listener for `Identifier`, and the user has
- * specified in configuration that `MemberExpression > Identifier` should be ignored.
- * Due to selector specificity rules, the `Identifier` listener will get called first. However,
- * if a given Identifier node is supposed to be ignored, then the `Identifier` offset listener
- * should not have been called at all. Without doing extra selector matching, we don't know
- * whether the Identifier matches the `MemberExpression > Identifier` selector until the
- * `MemberExpression > Identifier` listener is called.
- *
- * To avoid this, the `Identifier` listener isn't called until traversal finishes and all
- * ignored nodes are known.
- */
- listener =>
- node =>
- listenerCallQueue.push({ listener, node })
- );
- // For each ignored node selector, set up a listener to collect it into the `ignoredNodes` set.
- const ignoredNodes = new Set();
- /**
- * Ignores a node
- * @param {ASTNode} node The node to ignore
- * @returns {void}
- */
- function addToIgnoredNodes(node) {
- ignoredNodes.add(node);
- ignoredNodeFirstTokens.add(sourceCode.getFirstToken(node));
- }
- const ignoredNodeListeners = options.ignoredNodes.reduce(
- (listeners, ignoredSelector) => Object.assign(listeners, { [ignoredSelector]: addToIgnoredNodes }),
- {}
- );
- /*
- * Join the listeners, and add a listener to verify that all tokens actually have the correct indentation
- * at the end.
- *
- * Using Object.assign will cause some offset listeners to be overwritten if the same selector also appears
- * in `ignoredNodeListeners`. This isn't a problem because all of the matching nodes will be ignored,
- * so those listeners wouldn't be called anyway.
- */
- return Object.assign(
- offsetListeners,
- ignoredNodeListeners,
- {
- "*:exit"(node) {
- // If a node's type is nonstandard, we can't tell how its children should be offset, so ignore it.
- if (!KNOWN_NODES.has(node.type)) {
- addToIgnoredNodes(node);
- }
- },
- "Program:exit"() {
- // If ignoreComments option is enabled, ignore all comment tokens.
- if (options.ignoreComments) {
- sourceCode.getAllComments()
- .forEach(comment => offsets.ignoreToken(comment));
- }
- // Invoke the queued offset listeners for the nodes that aren't ignored.
- listenerCallQueue
- .filter(nodeInfo => !ignoredNodes.has(nodeInfo.node))
- .forEach(nodeInfo => nodeInfo.listener(nodeInfo.node));
- // Update the offsets for ignored nodes to prevent their child tokens from being reported.
- ignoredNodes.forEach(ignoreNode);
- addParensIndent(sourceCode.ast.tokens);
- /*
- * Create a Map from (tokenOrComment) => (precedingToken).
- * This is necessary because sourceCode.getTokenBefore does not handle a comment as an argument correctly.
- */
- const precedingTokens = sourceCode.ast.comments.reduce((commentMap, comment) => {
- const tokenOrCommentBefore = sourceCode.getTokenBefore(comment, { includeComments: true });
- return commentMap.set(comment, commentMap.has(tokenOrCommentBefore) ? commentMap.get(tokenOrCommentBefore) : tokenOrCommentBefore);
- }, new WeakMap());
- sourceCode.lines.forEach((line, lineIndex) => {
- const lineNumber = lineIndex + 1;
- if (!tokenInfo.firstTokensByLineNumber.has(lineNumber)) {
- // Don't check indentation on blank lines
- return;
- }
- const firstTokenOfLine = tokenInfo.firstTokensByLineNumber.get(lineNumber);
- if (firstTokenOfLine.loc.start.line !== lineNumber) {
- // Don't check the indentation of multi-line tokens (e.g. template literals or block comments) twice.
- return;
- }
- if (astUtils.isCommentToken(firstTokenOfLine)) {
- const tokenBefore = precedingTokens.get(firstTokenOfLine);
- const tokenAfter = tokenBefore ? sourceCode.getTokenAfter(tokenBefore) : sourceCode.ast.tokens[0];
- const mayAlignWithBefore = tokenBefore && !hasBlankLinesBetween(tokenBefore, firstTokenOfLine);
- const mayAlignWithAfter = tokenAfter && !hasBlankLinesBetween(firstTokenOfLine, tokenAfter);
- /*
- * If a comment precedes a line that begins with a semicolon token, align to that token, i.e.
- *
- * let foo
- * // comment
- * ;(async () => {})()
- */
- if (tokenAfter && astUtils.isSemicolonToken(tokenAfter) && !astUtils.isTokenOnSameLine(firstTokenOfLine, tokenAfter)) {
- offsets.setDesiredOffset(firstTokenOfLine, tokenAfter, 0);
- }
- // If a comment matches the expected indentation of the token immediately before or after, don't report it.
- if (
- mayAlignWithBefore && validateTokenIndent(firstTokenOfLine, offsets.getDesiredIndent(tokenBefore)) ||
- mayAlignWithAfter && validateTokenIndent(firstTokenOfLine, offsets.getDesiredIndent(tokenAfter))
- ) {
- return;
- }
- }
- // If the token matches the expected indentation, don't report it.
- if (validateTokenIndent(firstTokenOfLine, offsets.getDesiredIndent(firstTokenOfLine))) {
- return;
- }
- // Otherwise, report the token/comment.
- report(firstTokenOfLine, offsets.getDesiredIndent(firstTokenOfLine));
- });
- }
- }
- );
- }
- };
|