使用eslint和stylelint之类的工具扫描前端代码现在已经基本成为前端同学的标配。但是，业务这么复杂，指望eslint等提供的工具完全解决业务中遇到的代码问题还是不太现实的。我们一线业务同学也要有自己的写规则的能力。eslint是构建在AST Parser基础上的规则扫描器，缺省情况下使用espree作为AST解析器。rules写好对于AST事件的回调，linter处理源代码之后会根据相应的事件来回调rules中的处理函数。
另外，在进入细节之前，请思考一下：eslint的边界在哪里？哪些功能是通过eslint写规则可以做到的，哪些是用eslint无法做到的？
一 先学会如何写规则测试
兵马未动，测试先行。规则写出来，如何用实际代码进行测试呢？所幸非常简单，直接写个json串把代码写进来就好了。我们来看个no-console的例子，就是不允许代码中出现console.*语句的规则。首先把规则和测试运行对象ruleTester引进来：
const rule = require("../../../lib/rules/no-console"), { RuleTester } = require("../../../lib/rule-tester");
const ruleTester = new RuleTester();
然后我们就直接调用ruleTester的run函数就好了。有效的样例放在valid下面，无效的样例放在invalid下面，是不是很简单。ruleTester.run("no-console", rule, { valid: [ "Console.info(foo)",
{ code: "console.info(foo)", options: [{ allow: ["info"] }] }, { code: "console.warn(foo)", options: [{ allow: ["warn"] }] }, { code: "console.error(foo)", options: [{ allow: ["error"] }] }, { code: "console.log(foo)", options: [{ allow: ["log"] }] },
{ code: "console.info(foo)", options: [{ allow: ["warn", "info"] }] }, { code: "console.warn(foo)", options: [{ allow: ["error", "warn"] }] }, { code: "console.error(foo)", options: [{ allow: ["log", "error"] }] }, { code: "console.log(foo)", options: [{ allow: ["info", "log", "warn"] }] },
"var console = require('myconsole'); console.log(foo)" ],
能通过的情况比较容易，我们就直接给代码和选项就好。 invalid: [
{ code: "console.log(foo)", errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.error(foo)", errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.info(foo)", errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.warn(foo)", errors: [{ messageId: "unexpected", type: "MemberExpression" }] },
{ code: "console.log(foo)", options: [{ allow: ["error"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.error(foo)", options: [{ allow: ["warn"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.info(foo)", options: [{ allow: ["log"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.warn(foo)", options: [{ allow: ["error"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] },
{ code: "console.log(foo)", options: [{ allow: ["warn", "info"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.error(foo)", options: [{ allow: ["warn", "info", "log"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.info(foo)", options: [{ allow: ["warn", "error", "log"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] }, { code: "console.warn(foo)", options: [{ allow: ["info", "log"] }], errors: [{ messageId: "unexpected", type: "MemberExpression" }] },
{ code: "console.log(foo)", env: { node: true }, errors: [{ messageId: "unexpected", type: "MemberExpression" }] } ]});
./node_modules/.bin/mocha tests/lib/rules/no-console.js
no-console valid ✓ Console.info(foo) ✓ console.info(foo) ✓ console.warn(foo) ✓ console.error(foo) ✓ console.log(foo) ✓ console.info(foo) ✓ console.warn(foo) ✓ console.error(foo) ✓ console.log(foo) ✓ var console = require('myconsole'); console.log(foo) invalid ✓ console.log(foo) ✓ console.error(foo) ✓ console.info(foo) ✓ console.warn(foo) ✓ console.log(foo) ✓ console.error(foo) ✓ console.info(foo) ✓ console.warn(foo) ✓ console.log(foo) ✓ console.error(foo) ✓ console.info(foo) ✓ console.warn(foo) ✓ console.log(foo)
23 passing (83ms)
如果在valid里面放一个不能通过的，则会报错，比如我们加一个：ruleTester.run("no-console", rule, { valid: [ "Console.info(foo)",
// single array item { code: "console.log('Hello,World')", options: [] },
1 failing
1) no-console valid console.log('Hello,World'):
AssertionError [ERR_ASSERTION]: Should have no errors but had 1: [ { ruleId: 'no-console', severity: 1, message: 'Unexpected console statement.', line: 1, column: 1, nodeType: 'MemberExpression', messageId: 'unexpected', endLine: 1, endColumn: 12 }] + expected - actual
-1 +0 at testValidTemplate (lib/rule-tester/rule-tester.js:697:20) at Context.<anonymous> (lib/rule-tester/rule-tester.js:972:29) at processImmediate (node:internal/timers:464:21)
说明我们刚加的console是会报一个messageId为unexpected，而nodeType为MemberExpression的错误。 invalid: [
{ code: "console.log('Hello,World')", errors: [{ messageId: "unexpected", type: "MemberExpression" }] },
invalid ✓ console.log('Hello,World')
我们先看下规则的模板，其实主要要提供meta对象和create方法：module.exports = { meta: { type: "规则类型，如suggestion",
docs: { description: "规则描述", category: "规则分类：如Possible Errors", recommended: true, url: "说明规则的文档地址，如https://eslint.org/docs/rules/no-extra-semi" }, fixable: "是否可以修复，如code", schema: [] }, create: function(context) { return { }; }};
总体来说，一个eslint规则所能做的事情，就是写事件回调函数，在回调函数中使用context中获取的AST等信息进行分析。代码信息类主要我们使用getScope获取作用域的信息，getAncestors获取上一级AST节点，getDeclaredVariables获取变量表。最后的绝招是直接获取源代码getSourceCode自己分析去。markVariableAsUsed用于跨文件分析，用于分析变量的使用情况。report函数用于输出分析结果，比如报错信息、修改建议和自动修复的代码等。还以no-console为例，我们先看meta部分，这部分不涉及逻辑代码，都是一些配置： meta: { type: "suggestion",
docs: { description: "disallow the use of `console`", recommended: false, url: "https://eslint.org/docs/rules/no-console" },
schema: [ { type: "object", properties: { allow: { type: "array", items: { type: "string" }, minItems: 1, uniqueItems: true } }, additionalProperties: false } ],
messages: { unexpected: "Unexpected console statement." } },
我们再看no-console的回调函数，只处理一处Program:exit, 这是程序退出的事件： return { "Program:exit"() { const scope = context.getScope(); const consoleVar = astUtils.getVariableByName(scope, "console"); const shadowed = consoleVar && consoleVar.defs.length > 0;
const references = consoleVar ? consoleVar.references : scope.through.filter(isConsole);
if (!shadowed) { references .filter(isMemberAccessExceptAllowed) .forEach(report); } } };
1 获取作用域和AST信息
我们首先通过context.getScope()获取作用域信息。作用域与AST的对应关系如下图：
我们前面的console语句的例子，首先拿到的都是全局作用域，举例如下：<ref *1> GlobalScope { type: 'global', set: Map(38) { 'Array' => Variable { name: 'Array', identifiers: [], references: [], defs: [], tainted: false, stack: true, scope: [Circular *1], eslintImplicitGlobalSetting: 'readonly', eslintExplicitGlobal: false, eslintExplicitGlobalComments: undefined, writeable: false }, 'Boolean' => Variable { name: 'Boolean', identifiers: [], references: [], defs: [], tainted: false, stack: true, scope: [Circular *1], eslintImplicitGlobalSetting: 'readonly', eslintExplicitGlobal: false, eslintExplicitGlobalComments: undefined, writeable: false }, 'constructor' => Variable { name: 'constructor', identifiers: [], references: [], defs: [], tainted: false, stack: true, scope: [Circular *1], eslintImplicitGlobalSetting: 'readonly', eslintExplicitGlobal: false, eslintExplicitGlobalComments: undefined, writeable: false },...
具体看一下38个全局变量，复习下Javascript基础吧： set: Map(38) { 'Array' => [Variable], 'Boolean' => [Variable], 'constructor' => [Variable], 'Date' => [Variable], 'decodeURI' => [Variable], 'decodeURIComponent' => [Variable], 'encodeURI' => [Variable], 'encodeURIComponent' => [Variable], 'Error' => [Variable], 'escape' => [Variable], 'eval' => [Variable], 'EvalError' => [Variable], 'Function' => [Variable], 'hasOwnProperty' => [Variable], 'Infinity' => [Variable], 'isFinite' => [Variable], 'isNaN' => [Variable], 'isPrototypeOf' => [Variable], 'JSON' => [Variable], 'Math' => [Variable], 'NaN' => [Variable], 'Number' => [Variable], 'Object' => [Variable], 'parseFloat' => [Variable], 'parseInt' => [Variable], 'propertyIsEnumerable' => [Variable], 'RangeError' => [Variable], 'ReferenceError' => [Variable], 'RegExp' => [Variable], 'String' => [Variable], 'SyntaxError' => [Variable], 'toLocaleString' => [Variable], 'toString' => [Variable], 'TypeError' => [Variable], 'undefined' => [Variable], 'unescape' => [Variable], 'URIError' => [Variable], 'valueOf' => [Variable] },
我们看到，所有的变量，都以一个名为set的Map中，这样我们就可以以遍历获取所有的变量。针对no-console的规则，我们主要是要查找是否有叫console的变量名。于是可以这么写： getVariableByName(initScope, name) { let scope = initScope;
while (scope) { const variable = scope.set.get(name);
if (variable) { return variable; }
scope = scope.upper; }
return null; },
我们可以在刚才列出的38个变量中发现，console是并没有定义的变量，所以const consoleVar = astUtils.getVariableByName(scope, "console");
于是我们要去查找未定义的变量，这部分是在scope.through中，果然找到了name是console的节点：[ Reference { identifier: Node { type: 'Identifier', loc: [SourceLocation], range: [Array], name: 'console', parent: [Node] }, from: <ref *2> GlobalScope { type: 'global', set: [Map], taints: Map(0) {}, dynamic: true, block: [Node], through: [Circular *1], variables: [Array], references: [Array], variableScope: [Circular *2], functionExpressionScope: false, directCallToEvalScope: false, thisFound: false, __left: null, upper: null, isStrict: false, childScopes: [], __declaredVariables: [WeakMap], implicit: [Object] }, tainted: false, resolved: null, flag: 1, __maybeImplicitGlobal: undefined }]
这样我们就可以写个检查reference的名字是不是console的函数就好： function isConsole(reference) { const id = reference.identifier;
return id && id.name === "console"; }
然后用这个函数去filter scope.though中的所有未定义的变量：scope.through.filter(isConsole);
最后一步是输出报告，针对过滤出的reference进行报告： references .filter(isMemberAccessExceptAllowed) .forEach(report);
function report(reference) { const node = reference.identifier.parent;
context.report({ node, loc: node.loc, messageId: "unexpected" }); }
2 处理特定类型的语句
no-console从规则书写上并不是最容易的，我们以其为例主要是这类问题最多。下面我们举一反三，看看针对其它不应该出现的语句该如何处理。其中最简单的就是针对一类语句统统报错，比如no-continue规则，就是遇到ContinueStatement就报错：module.exports = { meta: { type: "suggestion",
docs: { description: "disallow `continue` statements", recommended: false, url: "https://eslint.org/docs/rules/no-continue" },
schema: [],
messages: { unexpected: "Unexpected use of continue statement." } },
create(context) {
return { ContinueStatement(node) { context.report({ node, messageId: "unexpected" }); } };
}};
不允许使用debugger的no-debugger规则： create(context) {
return { DebuggerStatement(node) { context.report({ node, messageId: "unexpected" }); } };
}
create(context) {
return { WithStatement(node) { context.report({ node, messageId: "unexpectedWith" }); } };
}
create(context) { function isLexicalDeclaration(node) { switch (node.type) { case "FunctionDeclaration": case "ClassDeclaration": return true; case "VariableDeclaration": return node.kind !== "var"; default: return false; } }
return { SwitchCase(node) { for (let i = 0; i < node.consequent.length; i++) { const statement = node.consequent[i];
if (isLexicalDeclaration(statement)) { context.report({ node: statement, messageId: "unexpected" }); } } } };
}
function check(node) { if ( node.arguments.length !== 1 && node.callee.type === "Identifier" && node.callee.name === "Array" ) { context.report({ node, messageId: "preferLiteral" }); } }
return { CallExpression: check, NewExpression: check };
create(context) { function checkVariable(variable) { astUtils.getModifyingReferences(variable.references).forEach(reference => { context.report({ node: reference.identifier, messageId: "class", data: { name: reference.identifier.name } });
}); }
function checkForClass(node) { context.getDeclaredVariables(node).forEach(checkVariable); }
return { ClassDeclaration: checkForClass, ClassExpression: checkForClass };
}
create(context) {
function isParameter(def) { return def.type === "Parameter"; }
function checkParams(node) { const variables = context.getDeclaredVariables(node);
for (let i = 0; i < variables.length; ++i) { const variable = variables[i];
const defs = variable.defs.filter(isParameter);
if (defs.length >= 2) { context.report({ node, messageId: "unexpected", data: { name: variable.name } }); } } }
return { FunctionDeclaration: checkParams, FunctionExpression: checkParams };
}
如果事件太多的话，可以写成一个数组，这被称为选择器数组：const allLoopTypes = ["WhileStatement", "DoWhileStatement", "ForStatement", "ForInStatement", "ForOfStatement"];... [loopSelector](node) { if (currentCodePath.currentSegments.some(segment => segment.reachable)) { loopsToReport.add(node); } },
除了直接处理语句类型，还可以针对类型加上一些额外的判断。 create(context) {
return {
UnaryExpression(node) { if (node.operator === "delete" && node.argument.type === "Identifier") { context.report({ node, messageId: "unexpected" }); } } };
}
create(context) {
return {
BinaryExpression(node) { const badOperator = node.operator === "==" || node.operator === "!=";
if (node.right.type === "Literal" && node.right.raw === "null" && badOperator || node.left.type === "Literal" && node.left.raw === "null" && badOperator) { context.report({ node, messageId: "unexpected" }); } } };
}
create(context) {
function isNegZero(node) { return node.type === "UnaryExpression" && node.operator === "-" && node.argument.type === "Literal" && node.argument.value === 0; } const OPERATORS_TO_CHECK = new Set([">", ">=", "<", "<=", "==", "===", "!=", "!=="]);
return { BinaryExpression(node) { if (OPERATORS_TO_CHECK.has(node.operator)) { if (isNegZero(node.left) || isNegZero(node.right)) { context.report({ node, messageId: "unexpected", data: { operator: node.operator } }); } } } }; }
create(context) { function checkVariable(variable) { astUtils.getModifyingReferences(variable.references).forEach(reference => { context.report({ node: reference.identifier, messageId: "const", data: { name: reference.identifier.name } }); }); }
return { VariableDeclaration(node) { if (node.kind === "const") { context.getDeclaredVariables(node).forEach(checkVariable); } } }; }
3 :exit - 语句结束事件
除了语句事件之外，eslint还提供了:exit事件。比如上面的例子我们使用了VariableDeclaration语句事件，我们下面看看如何使用VariableDeclaration结束时调用的VariableDeclaration:exit事件。 return { "VariableDeclaration:exit"(node) { if (node.kind === "var") { report(node); } } };
如果觉得进入和退出不好区分的话，我们来看一个不允许在非函数的块中使用var来定义变量的例子： BlockStatement: enterScope, "BlockStatement:exit": exitScope, ForStatement: enterScope, "ForStatement:exit": exitScope, ForInStatement: enterScope, "ForInStatement:exit": exitScope, ForOfStatement: enterScope, "ForOfStatement:exit": exitScope, SwitchStatement: enterScope, "SwitchStatement:exit": exitScope, CatchClause: enterScope, "CatchClause:exit": exitScope, StaticBlock: enterScope, "StaticBlock:exit": exitScope,
这些逻辑的作用是，进入语句块的时候调用enterScope，退出语句块的时候调用exitScope: function enterScope(node) { stack.push(node.range); }
function exitScope() { stack.pop(); }
4 直接使用文字信息 - Literal
比如不允许使用"-.7"这样省略了0的浮点数。此时使用Literal来处理纯文字信息。 create(context) { const sourceCode = context.getSourceCode();
return { Literal(node) {
if (typeof node.value === "number") { if (node.raw.startsWith(".")) { context.report({ node, messageId: "leading", fix(fixer) { const tokenBefore = sourceCode.getTokenBefore(node); const needsSpaceBefore = tokenBefore && tokenBefore.range[1] === node.range[0] && !astUtils.canTokensBeAdjacent(tokenBefore, `0${node.raw}`);
return fixer.insertTextBefore(node, needsSpaceBefore ? " 0" : "0"); } }); } if (node.raw.indexOf(".") === node.raw.length - 1) { context.report({ node, messageId: "trailing", fix: fixer => fixer.insertTextAfter(node, "0") }); } } } }; }
create(context) { return { Literal(node) { if (typeof node.value === "number" && /^0[0-9]/u.test(node.raw)) { context.report({ node, messageId: "noOcatal" }); } } }; }
三 代码路径分析
前面我们讨论的基本都是一个代码片段，现在我们把代码逻辑串起来，形成一条代码路径。
除了采用上面的事件处理方法之外，我们还可以针对CodePath事件进行处理：
事件onCodePathStart和onCodePathEnd用于整个路径的分析，而onCodePathSegmentStart, onCodePathSegmentEnd是CodePath中的一个片段，onCodePathSegmentLoop是循环片段。 create(context) { const ignoredLoopTypes = context.options[0] && context.options[0].ignore || [], loopTypesToCheck = getDifference(allLoopTypes, ignoredLoopTypes), loopSelector = loopTypesToCheck.join(","), loopsByTargetSegments = new Map(), loopsToReport = new Set();
let currentCodePath = null;
return { onCodePathStart(codePath) { currentCodePath = codePath; },
onCodePathEnd() { currentCodePath = currentCodePath.upper; },
[loopSelector](node) { if (currentCodePath.currentSegments.some(segment => segment.reachable)) { loopsToReport.add(node); } },
onCodePathSegmentStart(segment, node) { if (isLoopingTarget(node)) { const loop = node.parent;
loopsByTargetSegments.set(segment, loop); } },
onCodePathSegmentLoop(_, toSegment, node) { const loop = loopsByTargetSegments.get(toSegment);
if (node === loop || node.type === "ContinueStatement") { loopsToReport.delete(loop); } },
"Program:exit"() { loopsToReport.forEach( node => context.report({ node, messageId: "invalid" }) ); } }; }
四 提供问题自动修复的代码
我们之前报告问题都是使用context.report函数，自动修复代码也是通过这个接口返回给调用者。我们以将"=="和"!="替换成"==="和"!=="为例。这个fix没有多少技术含量哈，就是给原来发现问题的运算符多加一个"=":report(node, `${node.operator}=`);
最终实现时是调用了fixer的replaceText函数： fix(fixer) { if (isTypeOfBinary(node) || areLiteralsAndSameType(node)) { return fixer.replaceText(operatorToken, expectedOperator); } return null; }
function report(node, expectedOperator) { const operatorToken = sourceCode.getFirstTokenBetween( node.left, node.right, token => token.value === node.operator );
context.report({ node, loc: operatorToken.loc, messageId: "unexpected", data: { expectedOperator, actualOperator: node.operator }, fix(fixer) { if (isTypeOfBinary(node) || areLiteralsAndSameType(node)) { return fixer.replaceText(operatorToken, expectedOperator); } return null; } }); }
Fixer支持4个添加API，2个删除API，2个替换类的API：
五 高级话题
1 React JSX的支持
Facebook给我们封装好了框架，写起来也是蛮眼熟的。刚好之前没有举markVariableAsUsed的例子，正好一起看了：module.exports = { meta: { docs: { description: 'Prevent React to be marked as unused', category: 'Best Practices', recommended: true, url: docsUrl('jsx-uses-react'), }, schema: [], },
create(context) { const pragma = pragmaUtil.getFromContext(context); const fragment = pragmaUtil.getFragmentFromContext(context);
function handleOpeningElement() { context.markVariableAsUsed(pragma); }
return { JSXOpeningElement: handleOpeningElement, JSXOpeningFragment: handleOpeningElement, JSXFragment() { context.markVariableAsUsed(fragment); }, }; },};
JSX的特殊之处是增加了JSXOpenElement, JSXClosingElement, JSXOpenFragment, JSXClosingFragment等处理JSX的事件。2 TypeScript的支持
随着tslint合并到eslint中，TypeScript的lint功能由typescript-eslint承载。因为estree只支持javascript，typescript-eslint提供兼容estree格式的parser. 既然是ts的lint，自然是拥有了ts的支持，拥有了新的工具方法，其基本架构仍是和eslint一致的：import * as ts from 'typescript';import * as util from '../util';
export default util.createRule({ name: 'no-for-in-array', meta: { docs: { description: 'Disallow iterating over an array with a for-in loop', recommended: 'error', requiresTypeChecking: true, }, messages: { forInViolation: 'For-in loops over arrays are forbidden. Use for-of or array.forEach instead.', }, schema: [], type: 'problem', }, defaultOptions: [], create(context) { return { ForInStatement(node): void { const parserServices = util.getParserServices(context); const checker = parserServices.program.getTypeChecker(); const originalNode = parserServices.esTreeNodeToTSNodeMap.get(node);
const type = util.getConstrainedTypeAtLocation( checker, originalNode.expression, );
if ( util.isTypeArrayTypeOrUnionOfArrayTypes(type, checker) || (type.flags & ts.TypeFlags.StringLike) !== 0 ) { context.report({ node, messageId: 'forInViolation', }); } }, }; },});
3 更换ESLint的AST解析器
ESLint支持使用第三方AST解析器，刚好Babel也支持ESLint，于是我们就可以用@babel/eslint-parser来替换espree. 装好插件之后，修改.eslintrc.js即可：module.exports = { parser: "@babel/eslint-parser",};
六 StyleLint
说完了Eslint，我们再花一小点篇幅看下StyleLint。StyleLint与Eslint的架构思想一脉相承，都是对于AST的事件分析进行处理的工具。只不过css使用不同的AST Parser，比如Post CSS API, postcss-value-parser, postcss-selector-parser等。const rule = (primary) => { return (root, result) => { const validOptions = validateOptions(result, ruleName, { actual: primary });
if (!validOptions) { return; }
root.walkDecls((decl) => { const parsedValue = valueParser(getDeclarationValue(decl));
parsedValue.walk((node) => { if (isIgnoredFunction(node)) return false;
if (!isHexColor(node)) return;
report({ message: messages.rejected(node.value), node: decl, index: declarationValueIndex(decl) + node.sourceIndex, result, ruleName, }); }); }); };};
也是熟悉的report函数回报，也可以支持autofix的生成。
七 小结
以上，我们基本将eslint规则写法的大致框架梳理清楚了。当然，实际写规刚的过程中还需要对于AST以及语言细节有比较深的了解。预祝大家通过写出适合自己业务的检查器，写出更健壮的代码。
网站架构师（CUED）培训课程
网站架构师CUED(Cloud User Experience Design)，集项目经理、产品经理、原型设计师等多重身份于一身，帮助客户整理需求、内容及框架搭建工作，把客户的需求完整地在网站系统中实现 。需要网站架构师具备完整的逻辑能力，对行业有较深理解，协助用户完成网站的原型设计。点击阅读原文了解详情。
网友评论已有0条评论, 我也要评论