mirror of
https://github.com/idanoo/GoScrobble.git
synced 2024-11-23 00:45:16 +00:00
246 lines
6.8 KiB
JavaScript
246 lines
6.8 KiB
JavaScript
|
"use strict";
|
||
|
|
||
|
Object.defineProperty(exports, "__esModule", {
|
||
|
value: true
|
||
|
});
|
||
|
exports.default = void 0;
|
||
|
|
||
|
var _experimentalUtils = require("@typescript-eslint/experimental-utils");
|
||
|
|
||
|
var _utils = require("./utils");
|
||
|
|
||
|
const trimFXprefix = word => ['f', 'x'].includes(word.charAt(0)) ? word.substr(1) : word;
|
||
|
|
||
|
const doesBinaryExpressionContainStringNode = binaryExp => {
|
||
|
if ((0, _utils.isStringNode)(binaryExp.right)) {
|
||
|
return true;
|
||
|
}
|
||
|
|
||
|
if (binaryExp.left.type === _experimentalUtils.AST_NODE_TYPES.BinaryExpression) {
|
||
|
return doesBinaryExpressionContainStringNode(binaryExp.left);
|
||
|
}
|
||
|
|
||
|
return (0, _utils.isStringNode)(binaryExp.left);
|
||
|
};
|
||
|
|
||
|
const quoteStringValue = node => node.type === _experimentalUtils.AST_NODE_TYPES.TemplateLiteral ? `\`${node.quasis[0].value.raw}\`` : node.raw;
|
||
|
|
||
|
const compileMatcherPatterns = matchers => {
|
||
|
if (typeof matchers === 'string') {
|
||
|
const matcher = new RegExp(matchers, 'u');
|
||
|
return {
|
||
|
describe: matcher,
|
||
|
test: matcher,
|
||
|
it: matcher
|
||
|
};
|
||
|
}
|
||
|
|
||
|
return {
|
||
|
describe: matchers.describe ? new RegExp(matchers.describe, 'u') : null,
|
||
|
test: matchers.test ? new RegExp(matchers.test, 'u') : null,
|
||
|
it: matchers.it ? new RegExp(matchers.it, 'u') : null
|
||
|
};
|
||
|
};
|
||
|
|
||
|
var _default = (0, _utils.createRule)({
|
||
|
name: __filename,
|
||
|
meta: {
|
||
|
docs: {
|
||
|
category: 'Best Practices',
|
||
|
description: 'Enforce valid titles',
|
||
|
recommended: 'error'
|
||
|
},
|
||
|
messages: {
|
||
|
titleMustBeString: 'Title must be a string',
|
||
|
emptyTitle: '{{ jestFunctionName }} should not have an empty title',
|
||
|
duplicatePrefix: 'should not have duplicate prefix',
|
||
|
accidentalSpace: 'should not have leading or trailing spaces',
|
||
|
disallowedWord: '"{{ word }}" is not allowed in test titles.',
|
||
|
mustNotMatch: '{{ jestFunctionName }} should not match {{ pattern }}',
|
||
|
mustMatch: '{{ jestFunctionName }} should match {{ pattern }}'
|
||
|
},
|
||
|
type: 'suggestion',
|
||
|
schema: [{
|
||
|
type: 'object',
|
||
|
properties: {
|
||
|
ignoreTypeOfDescribeName: {
|
||
|
type: 'boolean',
|
||
|
default: false
|
||
|
},
|
||
|
disallowedWords: {
|
||
|
type: 'array',
|
||
|
items: {
|
||
|
type: 'string'
|
||
|
}
|
||
|
},
|
||
|
mustNotMatch: {
|
||
|
oneOf: [{
|
||
|
type: 'string'
|
||
|
}, {
|
||
|
type: 'object',
|
||
|
properties: {
|
||
|
describe: {
|
||
|
type: 'string'
|
||
|
},
|
||
|
test: {
|
||
|
type: 'string'
|
||
|
},
|
||
|
it: {
|
||
|
type: 'string'
|
||
|
}
|
||
|
},
|
||
|
additionalProperties: false
|
||
|
}]
|
||
|
},
|
||
|
mustMatch: {
|
||
|
oneOf: [{
|
||
|
type: 'string'
|
||
|
}, {
|
||
|
type: 'object',
|
||
|
properties: {
|
||
|
describe: {
|
||
|
type: 'string'
|
||
|
},
|
||
|
test: {
|
||
|
type: 'string'
|
||
|
},
|
||
|
it: {
|
||
|
type: 'string'
|
||
|
}
|
||
|
},
|
||
|
additionalProperties: false
|
||
|
}]
|
||
|
}
|
||
|
},
|
||
|
additionalProperties: false
|
||
|
}],
|
||
|
fixable: 'code'
|
||
|
},
|
||
|
defaultOptions: [{
|
||
|
ignoreTypeOfDescribeName: false,
|
||
|
disallowedWords: []
|
||
|
}],
|
||
|
|
||
|
create(context, [{
|
||
|
ignoreTypeOfDescribeName,
|
||
|
disallowedWords = [],
|
||
|
mustNotMatch,
|
||
|
mustMatch
|
||
|
}]) {
|
||
|
const disallowedWordsRegexp = new RegExp(`\\b(${disallowedWords.join('|')})\\b`, 'iu');
|
||
|
const mustNotMatchPatterns = compileMatcherPatterns(mustNotMatch !== null && mustNotMatch !== void 0 ? mustNotMatch : {});
|
||
|
const mustMatchPatterns = compileMatcherPatterns(mustMatch !== null && mustMatch !== void 0 ? mustMatch : {});
|
||
|
return {
|
||
|
CallExpression(node) {
|
||
|
if (!(0, _utils.isDescribeCall)(node) && !(0, _utils.isTestCaseCall)(node)) {
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
const [argument] = node.arguments;
|
||
|
|
||
|
if (!argument) {
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
if (!(0, _utils.isStringNode)(argument)) {
|
||
|
if (argument.type === _experimentalUtils.AST_NODE_TYPES.BinaryExpression && doesBinaryExpressionContainStringNode(argument)) {
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
if (argument.type !== _experimentalUtils.AST_NODE_TYPES.TemplateLiteral && !(ignoreTypeOfDescribeName && (0, _utils.isDescribeCall)(node))) {
|
||
|
context.report({
|
||
|
messageId: 'titleMustBeString',
|
||
|
loc: argument.loc
|
||
|
});
|
||
|
}
|
||
|
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
const title = (0, _utils.getStringValue)(argument);
|
||
|
|
||
|
if (!title) {
|
||
|
context.report({
|
||
|
messageId: 'emptyTitle',
|
||
|
data: {
|
||
|
jestFunctionName: (0, _utils.isDescribeCall)(node) ? _utils.DescribeAlias.describe : _utils.TestCaseName.test
|
||
|
},
|
||
|
node
|
||
|
});
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
if (disallowedWords.length > 0) {
|
||
|
const disallowedMatch = disallowedWordsRegexp.exec(title);
|
||
|
|
||
|
if (disallowedMatch) {
|
||
|
context.report({
|
||
|
data: {
|
||
|
word: disallowedMatch[1]
|
||
|
},
|
||
|
messageId: 'disallowedWord',
|
||
|
node: argument
|
||
|
});
|
||
|
return;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
if (title.trim().length !== title.length) {
|
||
|
context.report({
|
||
|
messageId: 'accidentalSpace',
|
||
|
node: argument,
|
||
|
fix: fixer => [fixer.replaceTextRange(argument.range, quoteStringValue(argument).replace(/^([`'"]) +?/u, '$1').replace(/ +?([`'"])$/u, '$1'))]
|
||
|
});
|
||
|
}
|
||
|
|
||
|
const nodeName = trimFXprefix((0, _utils.getNodeName)(node));
|
||
|
const [firstWord] = title.split(' ');
|
||
|
|
||
|
if (firstWord.toLowerCase() === nodeName) {
|
||
|
context.report({
|
||
|
messageId: 'duplicatePrefix',
|
||
|
node: argument,
|
||
|
fix: fixer => [fixer.replaceTextRange(argument.range, quoteStringValue(argument).replace(/^([`'"]).+? /u, '$1'))]
|
||
|
});
|
||
|
}
|
||
|
|
||
|
const [jestFunctionName] = nodeName.split('.');
|
||
|
const mustNotMatchPattern = mustNotMatchPatterns[jestFunctionName];
|
||
|
|
||
|
if (mustNotMatchPattern) {
|
||
|
if (mustNotMatchPattern.test(title)) {
|
||
|
context.report({
|
||
|
messageId: 'mustNotMatch',
|
||
|
node: argument,
|
||
|
data: {
|
||
|
jestFunctionName,
|
||
|
pattern: mustNotMatchPattern
|
||
|
}
|
||
|
});
|
||
|
return;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
const mustMatchPattern = mustMatchPatterns[jestFunctionName];
|
||
|
|
||
|
if (mustMatchPattern) {
|
||
|
if (!mustMatchPattern.test(title)) {
|
||
|
context.report({
|
||
|
messageId: 'mustMatch',
|
||
|
node: argument,
|
||
|
data: {
|
||
|
jestFunctionName,
|
||
|
pattern: mustMatchPattern
|
||
|
}
|
||
|
});
|
||
|
return;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
};
|
||
|
}
|
||
|
|
||
|
});
|
||
|
|
||
|
exports.default = _default;
|