printer-notifications/node_modules/eslint/lib/rules/lines-around-comment.js

472 lines
18 KiB
JavaScript
Raw Permalink Normal View History

2023-11-13 21:10:04 +00:00
/**
* @fileoverview Enforces empty lines around comments.
* @author Jamund Ferguson
* @deprecated in ESLint v8.53.0
*/
"use strict";
//------------------------------------------------------------------------------
// Requirements
//------------------------------------------------------------------------------
const astUtils = require("./utils/ast-utils");
//------------------------------------------------------------------------------
// Helpers
//------------------------------------------------------------------------------
/**
* Return an array with any line numbers that are empty.
* @param {Array} lines An array of each line of the file.
* @returns {Array} An array of line numbers.
*/
function getEmptyLineNums(lines) {
const emptyLines = lines.map((line, i) => ({
code: line.trim(),
num: i + 1
})).filter(line => !line.code).map(line => line.num);
return emptyLines;
}
/**
* Return an array with any line numbers that contain comments.
* @param {Array} comments An array of comment tokens.
* @returns {Array} An array of line numbers.
*/
function getCommentLineNums(comments) {
const lines = [];
comments.forEach(token => {
const start = token.loc.start.line;
const end = token.loc.end.line;
lines.push(start, end);
});
return lines;
}
//------------------------------------------------------------------------------
// Rule Definition
//------------------------------------------------------------------------------
/** @type {import('../shared/types').Rule} */
module.exports = {
meta: {
deprecated: true,
replacedBy: [],
type: "layout",
docs: {
description: "Require empty lines around comments",
recommended: false,
url: "https://eslint.org/docs/latest/rules/lines-around-comment"
},
fixable: "whitespace",
schema: [
{
type: "object",
properties: {
beforeBlockComment: {
type: "boolean",
default: true
},
afterBlockComment: {
type: "boolean",
default: false
},
beforeLineComment: {
type: "boolean",
default: false
},
afterLineComment: {
type: "boolean",
default: false
},
allowBlockStart: {
type: "boolean",
default: false
},
allowBlockEnd: {
type: "boolean",
default: false
},
allowClassStart: {
type: "boolean"
},
allowClassEnd: {
type: "boolean"
},
allowObjectStart: {
type: "boolean"
},
allowObjectEnd: {
type: "boolean"
},
allowArrayStart: {
type: "boolean"
},
allowArrayEnd: {
type: "boolean"
},
ignorePattern: {
type: "string"
},
applyDefaultIgnorePatterns: {
type: "boolean"
},
afterHashbangComment: {
type: "boolean",
default: false
}
},
additionalProperties: false
}
],
messages: {
after: "Expected line after comment.",
before: "Expected line before comment."
}
},
create(context) {
const options = Object.assign({}, context.options[0]);
const ignorePattern = options.ignorePattern;
const defaultIgnoreRegExp = astUtils.COMMENTS_IGNORE_PATTERN;
const customIgnoreRegExp = new RegExp(ignorePattern, "u");
const applyDefaultIgnorePatterns = options.applyDefaultIgnorePatterns !== false;
options.beforeBlockComment = typeof options.beforeBlockComment !== "undefined" ? options.beforeBlockComment : true;
const sourceCode = context.sourceCode;
const lines = sourceCode.lines,
numLines = lines.length + 1,
comments = sourceCode.getAllComments(),
commentLines = getCommentLineNums(comments),
emptyLines = getEmptyLineNums(lines),
commentAndEmptyLines = new Set(commentLines.concat(emptyLines));
/**
* Returns whether or not comments are on lines starting with or ending with code
* @param {token} token The comment token to check.
* @returns {boolean} True if the comment is not alone.
*/
function codeAroundComment(token) {
let currentToken = token;
do {
currentToken = sourceCode.getTokenBefore(currentToken, { includeComments: true });
} while (currentToken && astUtils.isCommentToken(currentToken));
if (currentToken && astUtils.isTokenOnSameLine(currentToken, token)) {
return true;
}
currentToken = token;
do {
currentToken = sourceCode.getTokenAfter(currentToken, { includeComments: true });
} while (currentToken && astUtils.isCommentToken(currentToken));
if (currentToken && astUtils.isTokenOnSameLine(token, currentToken)) {
return true;
}
return false;
}
/**
* Returns whether or not comments are inside a node type or not.
* @param {ASTNode} parent The Comment parent node.
* @param {string} nodeType The parent type to check against.
* @returns {boolean} True if the comment is inside nodeType.
*/
function isParentNodeType(parent, nodeType) {
return parent.type === nodeType ||
(parent.body && parent.body.type === nodeType) ||
(parent.consequent && parent.consequent.type === nodeType);
}
/**
* Returns the parent node that contains the given token.
* @param {token} token The token to check.
* @returns {ASTNode|null} The parent node that contains the given token.
*/
function getParentNodeOfToken(token) {
const node = sourceCode.getNodeByRangeIndex(token.range[0]);
/*
* For the purpose of this rule, the comment token is in a `StaticBlock` node only
* if it's inside the braces of that `StaticBlock` node.
*
* Example where this function returns `null`:
*
* static
* // comment
* {
* }
*
* Example where this function returns `StaticBlock` node:
*
* static
* {
* // comment
* }
*
*/
if (node && node.type === "StaticBlock") {
const openingBrace = sourceCode.getFirstToken(node, { skip: 1 }); // skip the `static` token
return token.range[0] >= openingBrace.range[0]
? node
: null;
}
return node;
}
/**
* Returns whether or not comments are at the parent start or not.
* @param {token} token The Comment token.
* @param {string} nodeType The parent type to check against.
* @returns {boolean} True if the comment is at parent start.
*/
function isCommentAtParentStart(token, nodeType) {
const parent = getParentNodeOfToken(token);
if (parent && isParentNodeType(parent, nodeType)) {
let parentStartNodeOrToken = parent;
if (parent.type === "StaticBlock") {
parentStartNodeOrToken = sourceCode.getFirstToken(parent, { skip: 1 }); // opening brace of the static block
} else if (parent.type === "SwitchStatement") {
parentStartNodeOrToken = sourceCode.getTokenAfter(parent.discriminant, {
filter: astUtils.isOpeningBraceToken
}); // opening brace of the switch statement
}
return token.loc.start.line - parentStartNodeOrToken.loc.start.line === 1;
}
return false;
}
/**
* Returns whether or not comments are at the parent end or not.
* @param {token} token The Comment token.
* @param {string} nodeType The parent type to check against.
* @returns {boolean} True if the comment is at parent end.
*/
function isCommentAtParentEnd(token, nodeType) {
const parent = getParentNodeOfToken(token);
return !!parent && isParentNodeType(parent, nodeType) &&
parent.loc.end.line - token.loc.end.line === 1;
}
/**
* Returns whether or not comments are at the block start or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at block start.
*/
function isCommentAtBlockStart(token) {
return (
isCommentAtParentStart(token, "ClassBody") ||
isCommentAtParentStart(token, "BlockStatement") ||
isCommentAtParentStart(token, "StaticBlock") ||
isCommentAtParentStart(token, "SwitchCase") ||
isCommentAtParentStart(token, "SwitchStatement")
);
}
/**
* Returns whether or not comments are at the block end or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at block end.
*/
function isCommentAtBlockEnd(token) {
return (
isCommentAtParentEnd(token, "ClassBody") ||
isCommentAtParentEnd(token, "BlockStatement") ||
isCommentAtParentEnd(token, "StaticBlock") ||
isCommentAtParentEnd(token, "SwitchCase") ||
isCommentAtParentEnd(token, "SwitchStatement")
);
}
/**
* Returns whether or not comments are at the class start or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at class start.
*/
function isCommentAtClassStart(token) {
return isCommentAtParentStart(token, "ClassBody");
}
/**
* Returns whether or not comments are at the class end or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at class end.
*/
function isCommentAtClassEnd(token) {
return isCommentAtParentEnd(token, "ClassBody");
}
/**
* Returns whether or not comments are at the object start or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at object start.
*/
function isCommentAtObjectStart(token) {
return isCommentAtParentStart(token, "ObjectExpression") || isCommentAtParentStart(token, "ObjectPattern");
}
/**
* Returns whether or not comments are at the object end or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at object end.
*/
function isCommentAtObjectEnd(token) {
return isCommentAtParentEnd(token, "ObjectExpression") || isCommentAtParentEnd(token, "ObjectPattern");
}
/**
* Returns whether or not comments are at the array start or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at array start.
*/
function isCommentAtArrayStart(token) {
return isCommentAtParentStart(token, "ArrayExpression") || isCommentAtParentStart(token, "ArrayPattern");
}
/**
* Returns whether or not comments are at the array end or not.
* @param {token} token The Comment token.
* @returns {boolean} True if the comment is at array end.
*/
function isCommentAtArrayEnd(token) {
return isCommentAtParentEnd(token, "ArrayExpression") || isCommentAtParentEnd(token, "ArrayPattern");
}
/**
* Checks if a comment token has lines around it (ignores inline comments)
* @param {token} token The Comment token.
* @param {Object} opts Options to determine the newline.
* @param {boolean} opts.after Should have a newline after this line.
* @param {boolean} opts.before Should have a newline before this line.
* @returns {void}
*/
function checkForEmptyLine(token, opts) {
if (applyDefaultIgnorePatterns && defaultIgnoreRegExp.test(token.value)) {
return;
}
if (ignorePattern && customIgnoreRegExp.test(token.value)) {
return;
}
let after = opts.after,
before = opts.before;
const prevLineNum = token.loc.start.line - 1,
nextLineNum = token.loc.end.line + 1,
commentIsNotAlone = codeAroundComment(token);
const blockStartAllowed = options.allowBlockStart &&
isCommentAtBlockStart(token) &&
!(options.allowClassStart === false &&
isCommentAtClassStart(token)),
blockEndAllowed = options.allowBlockEnd && isCommentAtBlockEnd(token) && !(options.allowClassEnd === false && isCommentAtClassEnd(token)),
classStartAllowed = options.allowClassStart && isCommentAtClassStart(token),
classEndAllowed = options.allowClassEnd && isCommentAtClassEnd(token),
objectStartAllowed = options.allowObjectStart && isCommentAtObjectStart(token),
objectEndAllowed = options.allowObjectEnd && isCommentAtObjectEnd(token),
arrayStartAllowed = options.allowArrayStart && isCommentAtArrayStart(token),
arrayEndAllowed = options.allowArrayEnd && isCommentAtArrayEnd(token);
const exceptionStartAllowed = blockStartAllowed || classStartAllowed || objectStartAllowed || arrayStartAllowed;
const exceptionEndAllowed = blockEndAllowed || classEndAllowed || objectEndAllowed || arrayEndAllowed;
// ignore top of the file and bottom of the file
if (prevLineNum < 1) {
before = false;
}
if (nextLineNum >= numLines) {
after = false;
}
// we ignore all inline comments
if (commentIsNotAlone) {
return;
}
const previousTokenOrComment = sourceCode.getTokenBefore(token, { includeComments: true });
const nextTokenOrComment = sourceCode.getTokenAfter(token, { includeComments: true });
// check for newline before
if (!exceptionStartAllowed && before && !commentAndEmptyLines.has(prevLineNum) &&
!(astUtils.isCommentToken(previousTokenOrComment) && astUtils.isTokenOnSameLine(previousTokenOrComment, token))) {
const lineStart = token.range[0] - token.loc.start.column;
const range = [lineStart, lineStart];
context.report({
node: token,
messageId: "before",
fix(fixer) {
return fixer.insertTextBeforeRange(range, "\n");
}
});
}
// check for newline after
if (!exceptionEndAllowed && after && !commentAndEmptyLines.has(nextLineNum) &&
!(astUtils.isCommentToken(nextTokenOrComment) && astUtils.isTokenOnSameLine(token, nextTokenOrComment))) {
context.report({
node: token,
messageId: "after",
fix(fixer) {
return fixer.insertTextAfter(token, "\n");
}
});
}
}
//--------------------------------------------------------------------------
// Public
//--------------------------------------------------------------------------
return {
Program() {
comments.forEach(token => {
if (token.type === "Line") {
if (options.beforeLineComment || options.afterLineComment) {
checkForEmptyLine(token, {
after: options.afterLineComment,
before: options.beforeLineComment
});
}
} else if (token.type === "Block") {
if (options.beforeBlockComment || options.afterBlockComment) {
checkForEmptyLine(token, {
after: options.afterBlockComment,
before: options.beforeBlockComment
});
}
} else if (token.type === "Shebang") {
if (options.afterHashbangComment) {
checkForEmptyLine(token, {
after: options.afterHashbangComment,
before: false
});
}
}
});
}
};
}
};