|
|
/** * @fileoverview Rule that warns about used warning comments * @author Alexander Schmidt <https://github.com/lxanders>
*/
"use strict";
const { escapeRegExp } = require("lodash"); const astUtils = require("./utils/ast-utils");
//------------------------------------------------------------------------------
// Rule Definition
//------------------------------------------------------------------------------
module.exports = { meta: { type: "suggestion",
docs: { description: "disallow specified warning terms in comments", category: "Best Practices", recommended: false, url: "https://eslint.org/docs/rules/no-warning-comments" },
schema: [ { type: "object", properties: { terms: { type: "array", items: { type: "string" } }, location: { enum: ["start", "anywhere"] } }, additionalProperties: false } ],
messages: { unexpectedComment: "Unexpected '{{matchedTerm}}' comment." } },
create(context) {
const sourceCode = context.getSourceCode(), configuration = context.options[0] || {}, warningTerms = configuration.terms || ["todo", "fixme", "xxx"], location = configuration.location || "start", selfConfigRegEx = /\bno-warning-comments\b/u;
/** * Convert a warning term into a RegExp which will match a comment containing that whole word in the specified * location ("start" or "anywhere"). If the term starts or ends with non word characters, then the match will not * require word boundaries on that side. * @param {string} term A term to convert to a RegExp * @returns {RegExp} The term converted to a RegExp */ function convertToRegExp(term) { const escaped = escapeRegExp(term); const wordBoundary = "\\b"; const eitherOrWordBoundary = `|${wordBoundary}`; let prefix;
/* * If the term ends in a word character (a-z0-9_), ensure a word * boundary at the end, so that substrings do not get falsely * matched. eg "todo" in a string such as "mastodon". * If the term ends in a non-word character, then \b won't match on * the boundary to the next non-word character, which would likely * be a space. For example `/\bFIX!\b/.test('FIX! blah') === false`. * In these cases, use no bounding match. Same applies for the * prefix, handled below. */ const suffix = /\w$/u.test(term) ? "\\b" : "";
if (location === "start") {
/* * When matching at the start, ignore leading whitespace, and * there's no need to worry about word boundaries. */ prefix = "^\\s*"; } else if (/^\w/u.test(term)) { prefix = wordBoundary; } else { prefix = ""; }
if (location === "start") {
/* * For location "start" the regex should be * ^\s*TERM\b. This checks the word boundary * at the beginning of the comment. */ return new RegExp(prefix + escaped + suffix, "iu"); }
/* * For location "anywhere" the regex should be * \bTERM\b|\bTERM\b, this checks the entire comment * for the term. */ return new RegExp(prefix + escaped + suffix + eitherOrWordBoundary + term + wordBoundary, "iu"); }
const warningRegExps = warningTerms.map(convertToRegExp);
/** * Checks the specified comment for matches of the configured warning terms and returns the matches. * @param {string} comment The comment which is checked. * @returns {Array} All matched warning terms for this comment. */ function commentContainsWarningTerm(comment) { const matches = [];
warningRegExps.forEach((regex, index) => { if (regex.test(comment)) { matches.push(warningTerms[index]); } });
return matches; }
/** * Checks the specified node for matching warning comments and reports them. * @param {ASTNode} node The AST node being checked. * @returns {void} undefined. */ function checkComment(node) { if (astUtils.isDirectiveComment(node) && selfConfigRegEx.test(node.value)) { return; }
const matches = commentContainsWarningTerm(node.value);
matches.forEach(matchedTerm => { context.report({ node, messageId: "unexpectedComment", data: { matchedTerm } }); }); }
return { Program() { const comments = sourceCode.getAllComments();
comments.filter(token => token.type !== "Shebang").forEach(checkComment); } }; } };
|