GoScrobble/web/node_modules/eslint/lib/rules/no-extra-parens.js

1227 lines
47 KiB
JavaScript

/**
* @fileoverview Disallow parenthesising higher precedence subexpressions.
* @author Michael Ficarra
*/
"use strict";
//------------------------------------------------------------------------------
// Rule Definition
//------------------------------------------------------------------------------
const { isParenthesized: isParenthesizedRaw } = require("eslint-utils");
const astUtils = require("./utils/ast-utils.js");
module.exports = {
meta: {
type: "layout",
docs: {
description: "disallow unnecessary parentheses",
category: "Possible Errors",
recommended: false,
url: "https://eslint.org/docs/rules/no-extra-parens"
},
fixable: "code",
schema: {
anyOf: [
{
type: "array",
items: [
{
enum: ["functions"]
}
],
minItems: 0,
maxItems: 1
},
{
type: "array",
items: [
{
enum: ["all"]
},
{
type: "object",
properties: {
conditionalAssign: { type: "boolean" },
nestedBinaryExpressions: { type: "boolean" },
returnAssign: { type: "boolean" },
ignoreJSX: { enum: ["none", "all", "single-line", "multi-line"] },
enforceForArrowConditionals: { type: "boolean" },
enforceForSequenceExpressions: { type: "boolean" },
enforceForNewInMemberExpressions: { type: "boolean" },
enforceForFunctionPrototypeMethods: { type: "boolean" }
},
additionalProperties: false
}
],
minItems: 0,
maxItems: 2
}
]
},
messages: {
unexpected: "Unnecessary parentheses around expression."
}
},
create(context) {
const sourceCode = context.getSourceCode();
const tokensToIgnore = new WeakSet();
const precedence = astUtils.getPrecedence;
const ALL_NODES = context.options[0] !== "functions";
const EXCEPT_COND_ASSIGN = ALL_NODES && context.options[1] && context.options[1].conditionalAssign === false;
const NESTED_BINARY = ALL_NODES && context.options[1] && context.options[1].nestedBinaryExpressions === false;
const EXCEPT_RETURN_ASSIGN = ALL_NODES && context.options[1] && context.options[1].returnAssign === false;
const IGNORE_JSX = ALL_NODES && context.options[1] && context.options[1].ignoreJSX;
const IGNORE_ARROW_CONDITIONALS = ALL_NODES && context.options[1] &&
context.options[1].enforceForArrowConditionals === false;
const IGNORE_SEQUENCE_EXPRESSIONS = ALL_NODES && context.options[1] &&
context.options[1].enforceForSequenceExpressions === false;
const IGNORE_NEW_IN_MEMBER_EXPR = ALL_NODES && context.options[1] &&
context.options[1].enforceForNewInMemberExpressions === false;
const IGNORE_FUNCTION_PROTOTYPE_METHODS = ALL_NODES && context.options[1] &&
context.options[1].enforceForFunctionPrototypeMethods === false;
const PRECEDENCE_OF_ASSIGNMENT_EXPR = precedence({ type: "AssignmentExpression" });
const PRECEDENCE_OF_UPDATE_EXPR = precedence({ type: "UpdateExpression" });
let reportsBuffer;
/**
* Determines whether the given node is a `call` or `apply` method call, invoked directly on a `FunctionExpression` node.
* Example: function(){}.call()
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is an immediate `call` or `apply` method call.
* @private
*/
function isImmediateFunctionPrototypeMethodCall(node) {
const callNode = astUtils.skipChainExpression(node);
if (callNode.type !== "CallExpression") {
return false;
}
const callee = astUtils.skipChainExpression(callNode.callee);
return (
callee.type === "MemberExpression" &&
callee.object.type === "FunctionExpression" &&
["call", "apply"].includes(astUtils.getStaticPropertyName(callee))
);
}
/**
* Determines if this rule should be enforced for a node given the current configuration.
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the rule should be enforced for this node.
* @private
*/
function ruleApplies(node) {
if (node.type === "JSXElement" || node.type === "JSXFragment") {
const isSingleLine = node.loc.start.line === node.loc.end.line;
switch (IGNORE_JSX) {
// Exclude this JSX element from linting
case "all":
return false;
// Exclude this JSX element if it is multi-line element
case "multi-line":
return isSingleLine;
// Exclude this JSX element if it is single-line element
case "single-line":
return !isSingleLine;
// Nothing special to be done for JSX elements
case "none":
break;
// no default
}
}
if (node.type === "SequenceExpression" && IGNORE_SEQUENCE_EXPRESSIONS) {
return false;
}
if (isImmediateFunctionPrototypeMethodCall(node) && IGNORE_FUNCTION_PROTOTYPE_METHODS) {
return false;
}
return ALL_NODES || node.type === "FunctionExpression" || node.type === "ArrowFunctionExpression";
}
/**
* Determines if a node is surrounded by parentheses.
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is parenthesised.
* @private
*/
function isParenthesised(node) {
return isParenthesizedRaw(1, node, sourceCode);
}
/**
* Determines if a node is surrounded by parentheses twice.
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is doubly parenthesised.
* @private
*/
function isParenthesisedTwice(node) {
return isParenthesizedRaw(2, node, sourceCode);
}
/**
* Determines if a node is surrounded by (potentially) invalid parentheses.
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is incorrectly parenthesised.
* @private
*/
function hasExcessParens(node) {
return ruleApplies(node) && isParenthesised(node);
}
/**
* Determines if a node that is expected to be parenthesised is surrounded by
* (potentially) invalid extra parentheses.
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is has an unexpected extra pair of parentheses.
* @private
*/
function hasDoubleExcessParens(node) {
return ruleApplies(node) && isParenthesisedTwice(node);
}
/**
* Determines if a node that is expected to be parenthesised is surrounded by
* (potentially) invalid extra parentheses with considering precedence level of the node.
* If the preference level of the node is not higher or equal to precedence lower limit, it also checks
* whether the node is surrounded by parentheses twice or not.
* @param {ASTNode} node The node to be checked.
* @param {number} precedenceLowerLimit The lower limit of precedence.
* @returns {boolean} True if the node is has an unexpected extra pair of parentheses.
* @private
*/
function hasExcessParensWithPrecedence(node, precedenceLowerLimit) {
if (ruleApplies(node) && isParenthesised(node)) {
if (
precedence(node) >= precedenceLowerLimit ||
isParenthesisedTwice(node)
) {
return true;
}
}
return false;
}
/**
* Determines if a node test expression is allowed to have a parenthesised assignment
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the assignment can be parenthesised.
* @private
*/
function isCondAssignException(node) {
return EXCEPT_COND_ASSIGN && node.test.type === "AssignmentExpression";
}
/**
* Determines if a node is in a return statement
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is in a return statement.
* @private
*/
function isInReturnStatement(node) {
for (let currentNode = node; currentNode; currentNode = currentNode.parent) {
if (
currentNode.type === "ReturnStatement" ||
(currentNode.type === "ArrowFunctionExpression" && currentNode.body.type !== "BlockStatement")
) {
return true;
}
}
return false;
}
/**
* Determines if a constructor function is newed-up with parens
* @param {ASTNode} newExpression The NewExpression node to be checked.
* @returns {boolean} True if the constructor is called with parens.
* @private
*/
function isNewExpressionWithParens(newExpression) {
const lastToken = sourceCode.getLastToken(newExpression);
const penultimateToken = sourceCode.getTokenBefore(lastToken);
return newExpression.arguments.length > 0 ||
(
// The expression should end with its own parens, e.g., new new foo() is not a new expression with parens
astUtils.isOpeningParenToken(penultimateToken) &&
astUtils.isClosingParenToken(lastToken) &&
newExpression.callee.range[1] < newExpression.range[1]
);
}
/**
* Determines if a node is or contains an assignment expression
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is or contains an assignment expression.
* @private
*/
function containsAssignment(node) {
if (node.type === "AssignmentExpression") {
return true;
}
if (node.type === "ConditionalExpression" &&
(node.consequent.type === "AssignmentExpression" || node.alternate.type === "AssignmentExpression")) {
return true;
}
if ((node.left && node.left.type === "AssignmentExpression") ||
(node.right && node.right.type === "AssignmentExpression")) {
return true;
}
return false;
}
/**
* Determines if a node is contained by or is itself a return statement and is allowed to have a parenthesised assignment
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the assignment can be parenthesised.
* @private
*/
function isReturnAssignException(node) {
if (!EXCEPT_RETURN_ASSIGN || !isInReturnStatement(node)) {
return false;
}
if (node.type === "ReturnStatement") {
return node.argument && containsAssignment(node.argument);
}
if (node.type === "ArrowFunctionExpression" && node.body.type !== "BlockStatement") {
return containsAssignment(node.body);
}
return containsAssignment(node);
}
/**
* Determines if a node following a [no LineTerminator here] restriction is
* surrounded by (potentially) invalid extra parentheses.
* @param {Token} token The token preceding the [no LineTerminator here] restriction.
* @param {ASTNode} node The node to be checked.
* @returns {boolean} True if the node is incorrectly parenthesised.
* @private
*/
function hasExcessParensNoLineTerminator(token, node) {
if (token.loc.end.line === node.loc.start.line) {
return hasExcessParens(node);
}
return hasDoubleExcessParens(node);
}
/**
* Determines whether a node should be preceded by an additional space when removing parens
* @param {ASTNode} node node to evaluate; must be surrounded by parentheses
* @returns {boolean} `true` if a space should be inserted before the node
* @private
*/
function requiresLeadingSpace(node) {
const leftParenToken = sourceCode.getTokenBefore(node);
const tokenBeforeLeftParen = sourceCode.getTokenBefore(leftParenToken, { includeComments: true });
const tokenAfterLeftParen = sourceCode.getTokenAfter(leftParenToken, { includeComments: true });
return tokenBeforeLeftParen &&
tokenBeforeLeftParen.range[1] === leftParenToken.range[0] &&
leftParenToken.range[1] === tokenAfterLeftParen.range[0] &&
!astUtils.canTokensBeAdjacent(tokenBeforeLeftParen, tokenAfterLeftParen);
}
/**
* Determines whether a node should be followed by an additional space when removing parens
* @param {ASTNode} node node to evaluate; must be surrounded by parentheses
* @returns {boolean} `true` if a space should be inserted after the node
* @private
*/
function requiresTrailingSpace(node) {
const nextTwoTokens = sourceCode.getTokensAfter(node, { count: 2 });
const rightParenToken = nextTwoTokens[0];
const tokenAfterRightParen = nextTwoTokens[1];
const tokenBeforeRightParen = sourceCode.getLastToken(node);
return rightParenToken && tokenAfterRightParen &&
!sourceCode.isSpaceBetweenTokens(rightParenToken, tokenAfterRightParen) &&
!astUtils.canTokensBeAdjacent(tokenBeforeRightParen, tokenAfterRightParen);
}
/**
* Determines if a given expression node is an IIFE
* @param {ASTNode} node The node to check
* @returns {boolean} `true` if the given node is an IIFE
*/
function isIIFE(node) {
const maybeCallNode = astUtils.skipChainExpression(node);
return maybeCallNode.type === "CallExpression" && maybeCallNode.callee.type === "FunctionExpression";
}
/**
* Determines if the given node can be the assignment target in destructuring or the LHS of an assignment.
* This is to avoid an autofix that could change behavior because parsers mistakenly allow invalid syntax,
* such as `(a = b) = c` and `[(a = b) = c] = []`. Ideally, this function shouldn't be necessary.
* @param {ASTNode} [node] The node to check
* @returns {boolean} `true` if the given node can be a valid assignment target
*/
function canBeAssignmentTarget(node) {
return node && (node.type === "Identifier" || node.type === "MemberExpression");
}
/**
* Report the node
* @param {ASTNode} node node to evaluate
* @returns {void}
* @private
*/
function report(node) {
const leftParenToken = sourceCode.getTokenBefore(node);
const rightParenToken = sourceCode.getTokenAfter(node);
if (!isParenthesisedTwice(node)) {
if (tokensToIgnore.has(sourceCode.getFirstToken(node))) {
return;
}
if (isIIFE(node) && !isParenthesised(node.callee)) {
return;
}
}
/**
* Finishes reporting
* @returns {void}
* @private
*/
function finishReport() {
context.report({
node,
loc: leftParenToken.loc,
messageId: "unexpected",
fix(fixer) {
const parenthesizedSource = sourceCode.text.slice(leftParenToken.range[1], rightParenToken.range[0]);
return fixer.replaceTextRange([
leftParenToken.range[0],
rightParenToken.range[1]
], (requiresLeadingSpace(node) ? " " : "") + parenthesizedSource + (requiresTrailingSpace(node) ? " " : ""));
}
});
}
if (reportsBuffer) {
reportsBuffer.reports.push({ node, finishReport });
return;
}
finishReport();
}
/**
* Evaluate a argument of the node.
* @param {ASTNode} node node to evaluate
* @returns {void}
* @private
*/
function checkArgumentWithPrecedence(node) {
if (hasExcessParensWithPrecedence(node.argument, precedence(node))) {
report(node.argument);
}
}
/**
* Check if a member expression contains a call expression
* @param {ASTNode} node MemberExpression node to evaluate
* @returns {boolean} true if found, false if not
*/
function doesMemberExpressionContainCallExpression(node) {
let currentNode = node.object;
let currentNodeType = node.object.type;
while (currentNodeType === "MemberExpression") {
currentNode = currentNode.object;
currentNodeType = currentNode.type;
}
return currentNodeType === "CallExpression";
}
/**
* Evaluate a new call
* @param {ASTNode} node node to evaluate
* @returns {void}
* @private
*/
function checkCallNew(node) {
const callee = node.callee;
if (hasExcessParensWithPrecedence(callee, precedence(node))) {
if (
hasDoubleExcessParens(callee) ||
!(
isIIFE(node) ||
// (new A)(); new (new A)();
(
callee.type === "NewExpression" &&
!isNewExpressionWithParens(callee) &&
!(
node.type === "NewExpression" &&
!isNewExpressionWithParens(node)
)
) ||
// new (a().b)(); new (a.b().c);
(
node.type === "NewExpression" &&
callee.type === "MemberExpression" &&
doesMemberExpressionContainCallExpression(callee)
) ||
// (a?.b)(); (a?.())();
(
!node.optional &&
callee.type === "ChainExpression"
)
)
) {
report(node.callee);
}
}
node.arguments
.filter(arg => hasExcessParensWithPrecedence(arg, PRECEDENCE_OF_ASSIGNMENT_EXPR))
.forEach(report);
}
/**
* Evaluate binary logicals
* @param {ASTNode} node node to evaluate
* @returns {void}
* @private
*/
function checkBinaryLogical(node) {
const prec = precedence(node);
const leftPrecedence = precedence(node.left);
const rightPrecedence = precedence(node.right);
const isExponentiation = node.operator === "**";
const shouldSkipLeft = NESTED_BINARY && (node.left.type === "BinaryExpression" || node.left.type === "LogicalExpression");
const shouldSkipRight = NESTED_BINARY && (node.right.type === "BinaryExpression" || node.right.type === "LogicalExpression");
if (!shouldSkipLeft && hasExcessParens(node.left)) {
if (
!(["AwaitExpression", "UnaryExpression"].includes(node.left.type) && isExponentiation) &&
!astUtils.isMixedLogicalAndCoalesceExpressions(node.left, node) &&
(leftPrecedence > prec || (leftPrecedence === prec && !isExponentiation)) ||
isParenthesisedTwice(node.left)
) {
report(node.left);
}
}
if (!shouldSkipRight && hasExcessParens(node.right)) {
if (
!astUtils.isMixedLogicalAndCoalesceExpressions(node.right, node) &&
(rightPrecedence > prec || (rightPrecedence === prec && isExponentiation)) ||
isParenthesisedTwice(node.right)
) {
report(node.right);
}
}
}
/**
* Check the parentheses around the super class of the given class definition.
* @param {ASTNode} node The node of class declarations to check.
* @returns {void}
*/
function checkClass(node) {
if (!node.superClass) {
return;
}
/*
* If `node.superClass` is a LeftHandSideExpression, parentheses are extra.
* Otherwise, parentheses are needed.
*/
const hasExtraParens = precedence(node.superClass) > PRECEDENCE_OF_UPDATE_EXPR
? hasExcessParens(node.superClass)
: hasDoubleExcessParens(node.superClass);
if (hasExtraParens) {
report(node.superClass);
}
}
/**
* Check the parentheses around the argument of the given spread operator.
* @param {ASTNode} node The node of spread elements/properties to check.
* @returns {void}
*/
function checkSpreadOperator(node) {
if (hasExcessParensWithPrecedence(node.argument, PRECEDENCE_OF_ASSIGNMENT_EXPR)) {
report(node.argument);
}
}
/**
* Checks the parentheses for an ExpressionStatement or ExportDefaultDeclaration
* @param {ASTNode} node The ExpressionStatement.expression or ExportDefaultDeclaration.declaration node
* @returns {void}
*/
function checkExpressionOrExportStatement(node) {
const firstToken = isParenthesised(node) ? sourceCode.getTokenBefore(node) : sourceCode.getFirstToken(node);
const secondToken = sourceCode.getTokenAfter(firstToken, astUtils.isNotOpeningParenToken);
const thirdToken = secondToken ? sourceCode.getTokenAfter(secondToken) : null;
const tokenAfterClosingParens = secondToken ? sourceCode.getTokenAfter(secondToken, astUtils.isNotClosingParenToken) : null;
if (
astUtils.isOpeningParenToken(firstToken) &&
(
astUtils.isOpeningBraceToken(secondToken) ||
secondToken.type === "Keyword" && (
secondToken.value === "function" ||
secondToken.value === "class" ||
secondToken.value === "let" &&
tokenAfterClosingParens &&
(
astUtils.isOpeningBracketToken(tokenAfterClosingParens) ||
tokenAfterClosingParens.type === "Identifier"
)
) ||
secondToken && secondToken.type === "Identifier" && secondToken.value === "async" && thirdToken && thirdToken.type === "Keyword" && thirdToken.value === "function"
)
) {
tokensToIgnore.add(secondToken);
}
const hasExtraParens = node.parent.type === "ExportDefaultDeclaration"
? hasExcessParensWithPrecedence(node, PRECEDENCE_OF_ASSIGNMENT_EXPR)
: hasExcessParens(node);
if (hasExtraParens) {
report(node);
}
}
/**
* Finds the path from the given node to the specified ancestor.
* @param {ASTNode} node First node in the path.
* @param {ASTNode} ancestor Last node in the path.
* @returns {ASTNode[]} Path, including both nodes.
* @throws {Error} If the given node does not have the specified ancestor.
*/
function pathToAncestor(node, ancestor) {
const path = [node];
let currentNode = node;
while (currentNode !== ancestor) {
currentNode = currentNode.parent;
/* istanbul ignore if */
if (currentNode === null) {
throw new Error("Nodes are not in the ancestor-descendant relationship.");
}
path.push(currentNode);
}
return path;
}
/**
* Finds the path from the given node to the specified descendant.
* @param {ASTNode} node First node in the path.
* @param {ASTNode} descendant Last node in the path.
* @returns {ASTNode[]} Path, including both nodes.
* @throws {Error} If the given node does not have the specified descendant.
*/
function pathToDescendant(node, descendant) {
return pathToAncestor(descendant, node).reverse();
}
/**
* Checks whether the syntax of the given ancestor of an 'in' expression inside a for-loop initializer
* is preventing the 'in' keyword from being interpreted as a part of an ill-formed for-in loop.
* @param {ASTNode} node Ancestor of an 'in' expression.
* @param {ASTNode} child Child of the node, ancestor of the same 'in' expression or the 'in' expression itself.
* @returns {boolean} True if the keyword 'in' would be interpreted as the 'in' operator, without any parenthesis.
*/
function isSafelyEnclosingInExpression(node, child) {
switch (node.type) {
case "ArrayExpression":
case "ArrayPattern":
case "BlockStatement":
case "ObjectExpression":
case "ObjectPattern":
case "TemplateLiteral":
return true;
case "ArrowFunctionExpression":
case "FunctionExpression":
return node.params.includes(child);
case "CallExpression":
case "NewExpression":
return node.arguments.includes(child);
case "MemberExpression":
return node.computed && node.property === child;
case "ConditionalExpression":
return node.consequent === child;
default:
return false;
}
}
/**
* Starts a new reports buffering. Warnings will be stored in a buffer instead of being reported immediately.
* An additional logic that requires multiple nodes (e.g. a whole subtree) may dismiss some of the stored warnings.
* @returns {void}
*/
function startNewReportsBuffering() {
reportsBuffer = {
upper: reportsBuffer,
inExpressionNodes: [],
reports: []
};
}
/**
* Ends the current reports buffering.
* @returns {void}
*/
function endCurrentReportsBuffering() {
const { upper, inExpressionNodes, reports } = reportsBuffer;
if (upper) {
upper.inExpressionNodes.push(...inExpressionNodes);
upper.reports.push(...reports);
} else {
// flush remaining reports
reports.forEach(({ finishReport }) => finishReport());
}
reportsBuffer = upper;
}
/**
* Checks whether the given node is in the current reports buffer.
* @param {ASTNode} node Node to check.
* @returns {boolean} True if the node is in the current buffer, false otherwise.
*/
function isInCurrentReportsBuffer(node) {
return reportsBuffer.reports.some(r => r.node === node);
}
/**
* Removes the given node from the current reports buffer.
* @param {ASTNode} node Node to remove.
* @returns {void}
*/
function removeFromCurrentReportsBuffer(node) {
reportsBuffer.reports = reportsBuffer.reports.filter(r => r.node !== node);
}
/**
* Checks whether a node is a MemberExpression at NewExpression's callee.
* @param {ASTNode} node node to check.
* @returns {boolean} True if the node is a MemberExpression at NewExpression's callee. false otherwise.
*/
function isMemberExpInNewCallee(node) {
if (node.type === "MemberExpression") {
return node.parent.type === "NewExpression" && node.parent.callee === node
? true
: node.parent.object === node && isMemberExpInNewCallee(node.parent);
}
return false;
}
return {
ArrayExpression(node) {
node.elements
.filter(e => e && hasExcessParensWithPrecedence(e, PRECEDENCE_OF_ASSIGNMENT_EXPR))
.forEach(report);
},
ArrayPattern(node) {
node.elements
.filter(e => canBeAssignmentTarget(e) && hasExcessParens(e))
.forEach(report);
},
ArrowFunctionExpression(node) {
if (isReturnAssignException(node)) {
return;
}
if (node.body.type === "ConditionalExpression" &&
IGNORE_ARROW_CONDITIONALS
) {
return;
}
if (node.body.type !== "BlockStatement") {
const firstBodyToken = sourceCode.getFirstToken(node.body, astUtils.isNotOpeningParenToken);
const tokenBeforeFirst = sourceCode.getTokenBefore(firstBodyToken);
if (astUtils.isOpeningParenToken(tokenBeforeFirst) && astUtils.isOpeningBraceToken(firstBodyToken)) {
tokensToIgnore.add(firstBodyToken);
}
if (hasExcessParensWithPrecedence(node.body, PRECEDENCE_OF_ASSIGNMENT_EXPR)) {
report(node.body);
}
}
},
AssignmentExpression(node) {
if (canBeAssignmentTarget(node.left) && hasExcessParens(node.left)) {
report(node.left);
}
if (!isReturnAssignException(node) && hasExcessParensWithPrecedence(node.right, precedence(node))) {
report(node.right);
}
},
BinaryExpression(node) {
if (reportsBuffer && node.operator === "in") {
reportsBuffer.inExpressionNodes.push(node);
}
checkBinaryLogical(node);
},
CallExpression: checkCallNew,
ClassBody(node) {
node.body
.filter(member => member.type === "MethodDefinition" && member.computed && member.key)
.filter(member => hasExcessParensWithPrecedence(member.key, PRECEDENCE_OF_ASSIGNMENT_EXPR))
.forEach(member => report(member.key));
},
ConditionalExpression(node) {
if (isReturnAssignException(node)) {
return;
}
if (
!isCondAssignException(node) &&
hasExcessParensWithPrecedence(node.test, precedence({ type: "LogicalExpression", operator: "||" }))
) {
report(node.test);
}
if (hasExcessParensWithPrecedence(node.consequent, PRECEDENCE_OF_ASSIGNMENT_EXPR)) {
report(node.consequent);
}
if (hasExcessParensWithPrecedence(node.alternate, PRECEDENCE_OF_ASSIGNMENT_EXPR)) {
report(node.alternate);
}
},
DoWhileStatement(node) {
if (hasExcessParens(node.test) && !isCondAssignException(node)) {
report(node.test);
}
},
ExportDefaultDeclaration: node => checkExpressionOrExportStatement(node.declaration),
ExpressionStatement: node => checkExpressionOrExportStatement(node.expression),
ForInStatement(node) {
if (node.left.type !== "VariableDeclaration") {
const firstLeftToken = sourceCode.getFirstToken(node.left, astUtils.isNotOpeningParenToken);
if (
firstLeftToken.value === "let" &&
astUtils.isOpeningBracketToken(
sourceCode.getTokenAfter(firstLeftToken, astUtils.isNotClosingParenToken)
)
) {
// ForInStatement#left expression cannot start with `let[`.
tokensToIgnore.add(firstLeftToken);
}
}
if (hasExcessParens(node.left)) {
report(node.left);
}
if (hasExcessParens(node.right)) {
report(node.right);
}
},
ForOfStatement(node) {
if (node.left.type !== "VariableDeclaration") {
const firstLeftToken = sourceCode.getFirstToken(node.left, astUtils.isNotOpeningParenToken);
if (firstLeftToken.value === "let") {
// ForOfStatement#left expression cannot start with `let`.
tokensToIgnore.add(firstLeftToken);
}
}
if (hasExcessParens(node.left)) {
report(node.left);
}
if (hasExcessParensWithPrecedence(node.right, PRECEDENCE_OF_ASSIGNMENT_EXPR)) {
report(node.right);
}
},
ForStatement(node) {
if (node.test && hasExcessParens(node.test) && !isCondAssignException(node)) {
report(node.test);
}
if (node.update && hasExcessParens(node.update)) {
report(node.update);
}
if (node.init) {
if (node.init.type !== "VariableDeclaration") {
const firstToken = sourceCode.getFirstToken(node.init, astUtils.isNotOpeningParenToken);
if (
firstToken.value === "let" &&
astUtils.isOpeningBracketToken(
sourceCode.getTokenAfter(firstToken, astUtils.isNotClosingParenToken)
)
) {
// ForStatement#init expression cannot start with `let[`.
tokensToIgnore.add(firstToken);
}
}
startNewReportsBuffering();
if (hasExcessParens(node.init)) {
report(node.init);
}
}
},
"ForStatement > *.init:exit"(node) {
/*
* Removing parentheses around `in` expressions might change semantics and cause errors.
*
* For example, this valid for loop:
* for (let a = (b in c); ;);
* after removing parentheses would be treated as an invalid for-in loop:
* for (let a = b in c; ;);
*/
if (reportsBuffer.reports.length) {
reportsBuffer.inExpressionNodes.forEach(inExpressionNode => {
const path = pathToDescendant(node, inExpressionNode);
let nodeToExclude;
for (let i = 0; i < path.length; i++) {
const pathNode = path[i];
if (i < path.length - 1) {
const nextPathNode = path[i + 1];
if (isSafelyEnclosingInExpression(pathNode, nextPathNode)) {
// The 'in' expression in safely enclosed by the syntax of its ancestor nodes (e.g. by '{}' or '[]').
return;
}
}
if (isParenthesised(pathNode)) {
if (isInCurrentReportsBuffer(pathNode)) {
// This node was supposed to be reported, but parentheses might be necessary.
if (isParenthesisedTwice(pathNode)) {
/*
* This node is parenthesised twice, it certainly has at least one pair of `extra` parentheses.
* If the --fix option is on, the current fixing iteration will remove only one pair of parentheses.
* The remaining pair is safely enclosing the 'in' expression.
*/
return;
}
// Exclude the outermost node only.
if (!nodeToExclude) {
nodeToExclude = pathNode;
}
// Don't break the loop here, there might be some safe nodes or parentheses that will stay inside.
} else {
// This node will stay parenthesised, the 'in' expression in safely enclosed by '()'.
return;
}
}
}
// Exclude the node from the list (i.e. treat parentheses as necessary)
removeFromCurrentReportsBuffer(nodeToExclude);
});
}
endCurrentReportsBuffering();
},
IfStatement(node) {
if (hasExcessParens(node.test) && !isCondAssignException(node)) {
report(node.test);
}
},
ImportExpression(node) {
const { source } = node;
if (source.type === "SequenceExpression") {
if (hasDoubleExcessParens(source)) {
report(source);
}
} else if (hasExcessParens(source)) {
report(source);
}
},
LogicalExpression: checkBinaryLogical,
MemberExpression(node) {
const shouldAllowWrapOnce = isMemberExpInNewCallee(node) &&
doesMemberExpressionContainCallExpression(node);
const nodeObjHasExcessParens = shouldAllowWrapOnce
? hasDoubleExcessParens(node.object)
: hasExcessParens(node.object) &&
!(
isImmediateFunctionPrototypeMethodCall(node.parent) &&
node.parent.callee === node &&
IGNORE_FUNCTION_PROTOTYPE_METHODS
);
if (
nodeObjHasExcessParens &&
precedence(node.object) >= precedence(node) &&
(
node.computed ||
!(
astUtils.isDecimalInteger(node.object) ||
// RegExp literal is allowed to have parens (#1589)
(node.object.type === "Literal" && node.object.regex)
)
)
) {
report(node.object);
}
if (nodeObjHasExcessParens &&
node.object.type === "CallExpression"
) {
report(node.object);
}
if (nodeObjHasExcessParens &&
!IGNORE_NEW_IN_MEMBER_EXPR &&
node.object.type === "NewExpression" &&
isNewExpressionWithParens(node.object)) {
report(node.object);
}
if (nodeObjHasExcessParens &&
node.optional &&
node.object.type === "ChainExpression"
) {
report(node.object);
}
if (node.computed && hasExcessParens(node.property)) {
report(node.property);
}
},
NewExpression: checkCallNew,
ObjectExpression(node) {
node.properties
.filter(property => property.value && hasExcessParensWithPrecedence(property.value, PRECEDENCE_OF_ASSIGNMENT_EXPR))
.forEach(property => report(property.value));
},
ObjectPattern(node) {
node.properties
.filter(property => {
const value = property.value;
return canBeAssignmentTarget(value) && hasExcessParens(value);
}).forEach(property => report(property.value));
},
Property(node) {
if (node.computed) {
const { key } = node;
if (key && hasExcessParensWithPrecedence(key, PRECEDENCE_OF_ASSIGNMENT_EXPR)) {
report(key);
}
}
},
RestElement(node) {
const argument = node.argument;
if (canBeAssignmentTarget(argument) && hasExcessParens(argument)) {
report(argument);
}
},
ReturnStatement(node) {
const returnToken = sourceCode.getFirstToken(node);
if (isReturnAssignException(node)) {
return;
}
if (node.argument &&
hasExcessParensNoLineTerminator(returnToken, node.argument) &&
// RegExp literal is allowed to have parens (#1589)
!(node.argument.type === "Literal" && node.argument.regex)) {
report(node.argument);
}
},
SequenceExpression(node) {
const precedenceOfNode = precedence(node);
node.expressions
.filter(e => hasExcessParensWithPrecedence(e, precedenceOfNode))
.forEach(report);
},
SwitchCase(node) {
if (node.test && hasExcessParens(node.test)) {
report(node.test);
}
},
SwitchStatement(node) {
if (hasExcessParens(node.discriminant)) {
report(node.discriminant);
}
},
ThrowStatement(node) {
const throwToken = sourceCode.getFirstToken(node);
if (hasExcessParensNoLineTerminator(throwToken, node.argument)) {
report(node.argument);
}
},
UnaryExpression: checkArgumentWithPrecedence,
UpdateExpression(node) {
if (node.prefix) {
checkArgumentWithPrecedence(node);
} else {
const { argument } = node;
const operatorToken = sourceCode.getLastToken(node);
if (argument.loc.end.line === operatorToken.loc.start.line) {
checkArgumentWithPrecedence(node);
} else {
if (hasDoubleExcessParens(argument)) {
report(argument);
}
}
}
},
AwaitExpression: checkArgumentWithPrecedence,
VariableDeclarator(node) {
if (
node.init && hasExcessParensWithPrecedence(node.init, PRECEDENCE_OF_ASSIGNMENT_EXPR) &&
// RegExp literal is allowed to have parens (#1589)
!(node.init.type === "Literal" && node.init.regex)
) {
report(node.init);
}
},
WhileStatement(node) {
if (hasExcessParens(node.test) && !isCondAssignException(node)) {
report(node.test);
}
},
WithStatement(node) {
if (hasExcessParens(node.object)) {
report(node.object);
}
},
YieldExpression(node) {
if (node.argument) {
const yieldToken = sourceCode.getFirstToken(node);
if ((precedence(node.argument) >= precedence(node) &&
hasExcessParensNoLineTerminator(yieldToken, node.argument)) ||
hasDoubleExcessParens(node.argument)) {
report(node.argument);
}
}
},
ClassDeclaration: checkClass,
ClassExpression: checkClass,
SpreadElement: checkSpreadOperator,
SpreadProperty: checkSpreadOperator,
ExperimentalSpreadProperty: checkSpreadOperator,
TemplateLiteral(node) {
node.expressions
.filter(e => e && hasExcessParens(e))
.forEach(report);
},
AssignmentPattern(node) {
const { left, right } = node;
if (canBeAssignmentTarget(left) && hasExcessParens(left)) {
report(left);
}
if (right && hasExcessParensWithPrecedence(right, PRECEDENCE_OF_ASSIGNMENT_EXPR)) {
report(right);
}
}
};
}
};