odin-js-fundamentals-part-5/node_modules/eslint/lib/rules/no-irregular-whitespace.js

277 lines
9.9 KiB
JavaScript
Raw Permalink Normal View History

2024-01-11 08:52:05 +00:00
/**
* @fileoverview Rule to disallow whitespace that is not a tab or space, whitespace inside strings and comments are allowed
* @author Jonathan Kingston
* @author Christophe Porteneuve
*/
"use strict";
//------------------------------------------------------------------------------
// Requirements
//------------------------------------------------------------------------------
const astUtils = require("./utils/ast-utils");
//------------------------------------------------------------------------------
// Constants
//------------------------------------------------------------------------------
const ALL_IRREGULARS = /[\f\v\u0085\ufeff\u00a0\u1680\u180e\u2000\u2001\u2002\u2003\u2004\u2005\u2006\u2007\u2008\u2009\u200a\u200b\u202f\u205f\u3000\u2028\u2029]/u;
const IRREGULAR_WHITESPACE = /[\f\v\u0085\ufeff\u00a0\u1680\u180e\u2000\u2001\u2002\u2003\u2004\u2005\u2006\u2007\u2008\u2009\u200a\u200b\u202f\u205f\u3000]+/mgu;
const IRREGULAR_LINE_TERMINATORS = /[\u2028\u2029]/mgu;
const LINE_BREAK = astUtils.createGlobalLinebreakMatcher();
//------------------------------------------------------------------------------
// Rule Definition
//------------------------------------------------------------------------------
/** @type {import('../shared/types').Rule} */
module.exports = {
meta: {
type: "problem",
docs: {
description: "Disallow irregular whitespace",
recommended: true,
url: "https://eslint.org/docs/latest/rules/no-irregular-whitespace"
},
schema: [
{
type: "object",
properties: {
skipComments: {
type: "boolean",
default: false
},
skipStrings: {
type: "boolean",
default: true
},
skipTemplates: {
type: "boolean",
default: false
},
skipRegExps: {
type: "boolean",
default: false
},
skipJSXText: {
type: "boolean",
default: false
}
},
additionalProperties: false
}
],
messages: {
noIrregularWhitespace: "Irregular whitespace not allowed."
}
},
create(context) {
// Module store of errors that we have found
let errors = [];
// Lookup the `skipComments` option, which defaults to `false`.
const options = context.options[0] || {};
const skipComments = !!options.skipComments;
const skipStrings = options.skipStrings !== false;
const skipRegExps = !!options.skipRegExps;
const skipTemplates = !!options.skipTemplates;
const skipJSXText = !!options.skipJSXText;
const sourceCode = context.sourceCode;
const commentNodes = sourceCode.getAllComments();
/**
* Removes errors that occur inside the given node
* @param {ASTNode} node to check for matching errors.
* @returns {void}
* @private
*/
function removeWhitespaceError(node) {
const locStart = node.loc.start;
const locEnd = node.loc.end;
errors = errors.filter(({ loc: { start: errorLocStart } }) => (
errorLocStart.line < locStart.line ||
errorLocStart.line === locStart.line && errorLocStart.column < locStart.column ||
errorLocStart.line === locEnd.line && errorLocStart.column >= locEnd.column ||
errorLocStart.line > locEnd.line
));
}
/**
* Checks literal nodes for errors that we are choosing to ignore and calls the relevant methods to remove the errors
* @param {ASTNode} node to check for matching errors.
* @returns {void}
* @private
*/
function removeInvalidNodeErrorsInLiteral(node) {
const shouldCheckStrings = skipStrings && (typeof node.value === "string");
const shouldCheckRegExps = skipRegExps && Boolean(node.regex);
if (shouldCheckStrings || shouldCheckRegExps) {
// If we have irregular characters remove them from the errors list
if (ALL_IRREGULARS.test(node.raw)) {
removeWhitespaceError(node);
}
}
}
/**
* Checks template string literal nodes for errors that we are choosing to ignore and calls the relevant methods to remove the errors
* @param {ASTNode} node to check for matching errors.
* @returns {void}
* @private
*/
function removeInvalidNodeErrorsInTemplateLiteral(node) {
if (typeof node.value.raw === "string") {
if (ALL_IRREGULARS.test(node.value.raw)) {
removeWhitespaceError(node);
}
}
}
/**
* Checks comment nodes for errors that we are choosing to ignore and calls the relevant methods to remove the errors
* @param {ASTNode} node to check for matching errors.
* @returns {void}
* @private
*/
function removeInvalidNodeErrorsInComment(node) {
if (ALL_IRREGULARS.test(node.value)) {
removeWhitespaceError(node);
}
}
/**
* Checks JSX nodes for errors that we are choosing to ignore and calls the relevant methods to remove the errors
* @param {ASTNode} node to check for matching errors.
* @returns {void}
* @private
*/
function removeInvalidNodeErrorsInJSXText(node) {
if (ALL_IRREGULARS.test(node.raw)) {
removeWhitespaceError(node);
}
}
/**
* Checks the program source for irregular whitespace
* @param {ASTNode} node The program node
* @returns {void}
* @private
*/
function checkForIrregularWhitespace(node) {
const sourceLines = sourceCode.lines;
sourceLines.forEach((sourceLine, lineIndex) => {
const lineNumber = lineIndex + 1;
let match;
while ((match = IRREGULAR_WHITESPACE.exec(sourceLine)) !== null) {
errors.push({
node,
messageId: "noIrregularWhitespace",
loc: {
start: {
line: lineNumber,
column: match.index
},
end: {
line: lineNumber,
column: match.index + match[0].length
}
}
});
}
});
}
/**
* Checks the program source for irregular line terminators
* @param {ASTNode} node The program node
* @returns {void}
* @private
*/
function checkForIrregularLineTerminators(node) {
const source = sourceCode.getText(),
sourceLines = sourceCode.lines,
linebreaks = source.match(LINE_BREAK);
let lastLineIndex = -1,
match;
while ((match = IRREGULAR_LINE_TERMINATORS.exec(source)) !== null) {
const lineIndex = linebreaks.indexOf(match[0], lastLineIndex + 1) || 0;
errors.push({
node,
messageId: "noIrregularWhitespace",
loc: {
start: {
line: lineIndex + 1,
column: sourceLines[lineIndex].length
},
end: {
line: lineIndex + 2,
column: 0
}
}
});
lastLineIndex = lineIndex;
}
}
/**
* A no-op function to act as placeholder for comment accumulation when the `skipComments` option is `false`.
* @returns {void}
* @private
*/
function noop() {}
const nodes = {};
if (ALL_IRREGULARS.test(sourceCode.getText())) {
nodes.Program = function(node) {
/*
* As we can easily fire warnings for all white space issues with
* all the source its simpler to fire them here.
* This means we can check all the application code without having
* to worry about issues caused in the parser tokens.
* When writing this code also evaluating per node was missing out
* connecting tokens in some cases.
* We can later filter the errors when they are found to be not an
* issue in nodes we don't care about.
*/
checkForIrregularWhitespace(node);
checkForIrregularLineTerminators(node);
};
nodes.Literal = removeInvalidNodeErrorsInLiteral;
nodes.TemplateElement = skipTemplates ? removeInvalidNodeErrorsInTemplateLiteral : noop;
nodes.JSXText = skipJSXText ? removeInvalidNodeErrorsInJSXText : noop;
nodes["Program:exit"] = function() {
if (skipComments) {
// First strip errors occurring in comment nodes.
commentNodes.forEach(removeInvalidNodeErrorsInComment);
}
// If we have any errors remaining report on them
errors.forEach(error => context.report(error));
};
} else {
nodes.Program = noop;
}
return nodes;
}
};