Skip to content

Latest commit

 

History

History
640 lines (458 loc) · 27.9 KB

DEV.md

File metadata and controls

640 lines (458 loc) · 27.9 KB

Developer Guide

Prerequisites

To work on this project, it is required to have the following tools installed:

RSPEC Rule Metadata

Rule metadata is automatically fetched from the SonarSource/rspec repository as part of the build. On first run, the rspec repo is cloned (sparse checkout) into resources/rspec/ via SSH. On subsequent runs, the latest changes are fetched automatically.

Use --rspec-path to point to an existing local clone instead:

npm run sync-rspec -- --rspec-path ../rspec --language javascript

Pinning a specific rspec version

To use a specific rspec commit instead of the latest dogfood-automerge, create a rspec.sha file at the repo root:

echo "<commit-sha>" > rspec.sha

When this file is present, sync-rspec fetches that exact SHA and the skip check becomes local (no network call needed). The file is not tracked by default but can be committed to make CI builds reproducible against a specific rspec version.

You can also use Docker container defined in ./.cirrus/nodejs.Dockerfile which bundles all required dependencies and is used for our CI pipeline.

Build and run unit tests

To build the plugin and run its unit tests, execute this command from the project's root directory:

mvn clean install

To skip tests:

mvn clean install -DskipTests

Integration Tests

First make sure the submodules are checked out:

 git submodule init
 git submodule update

Plugin Tests

The "Plugin Test" is an integration test which verifies plugin features such as metric calculation, coverage etc.

cd its/plugin
mvn clean install

Ruling Tests

The "Ruling Test" is an integration test which launches the analysis of a large code base of third-party projects (stored as submodules), saves the issues created by the plugin in report files, and then compares those results to the set of expected issues (stored as JSON files). This test gives you the opportunity to examine the issues created by each rule and make sure that they are what you expect.

JS/TS

npm run ruling

You can copy the files with the actual issues located at packages/ruling/actual/ into the directory with the expected issues its/ruling/src/test/resources/expected/.

From the project root, run: npm run ruling-sync

You can review the Ruling difference by running sh tools/ruling-debug-script.sh.

Java ruling (old way)

cd its/ruling
mvn verify -Dtest=RulingTest -Dmaven.test.redirectTestOutputToFile=false

To review the Ruling difference in SonarQube UI, put the breakpoint on assertThat(...) in RulingTest.java and open in the browser the orchestrated local SonarQube. Note that you can fix the port in orchestrator.properties files, e.g. orchestrator.container.port=9100.

If everything looks good to you, you can copy the file with the actual issues located at its/ruling/target/actual/ into the directory with the expected issues its/ruling/src/test/expected/.

You can review the Ruling difference by running diff -rq src/test/expected target/actual from its/ruling.

⚠️ Please note that running ruling tests will remove node_modules from the root to avoid affecting the results. Run npm ci to put them back.

Debug node process during scan

You can run your own Node.js process manually and set the environment variable SONARJS_EXISTING_NODE_PROCESS_PORT with the value of the port where your process is listening to. When set, SonarJS will not start a new Node process and will send the analysis requests to the specified port instead.

When using this for the ruling tests, make sure that you run them in series (and not in parallel), by removing @Execution(ExecutionMode.CONCURRENT) from the ruling test.

Adding a rule

Rule Description

  1. Create a PR with a rule description in RSPEC repo like described here
  • Tag the RSPEC with type-dependent if the rule relies partially or fully on type information
  • Add a field dependencies if your rule should only be executed if it relies on a specific import (example: 'react' or 'jest')
  • Add a field compatibleLanguages with an array including, which languages you support (js and/or ts).
  1. Link this RSPEC PR to the implementation issue in this repo
  2. Make sure the implementation issue title contains the RSPEC number and name

Implementing a rule

  1. Generate other files required for a new rule. Just choose your options in the prompt of the new-rule script
npm run new-rule

This script will ask a few questions and:

  • generates a rules/SXXXX folder
  • generates a rules/SXXXX/index.ts rule index file
  • generates a rules/SXXXX/rule.ts file for the rule implementation
  • generates a rules/SXXXX/cb.fixture.js comment-based test file (empty)
  • generates a rules/SXXXX/cb.test.js test launcher

It will also update some files which are not tracked by Git as they are automatically generated:

  • generates a Java check class for the rule SXXXX.java
  • updates the rules/rules.ts file to include the new rule
  • updates the rules/plugin-rules.ts file to include the new rule
  • updates the AllRules.java to include the new rule

  1. Update generated files

    • Make sure annotations in the Java class specify languages to cover (@JavaScriptRule and/or @TypeScriptRule)
    • If your rule has configurations, or you are using some from an ESLint rule, override the configurations() method of the Java check class
      • You can use a MyRuleCheckTest.java test case to verify how the configurations will be serialized to JSON as shown here
    • If writing a rule for the test files, replace extends Check with extends TestFileCheck in the Java class. This will be done by the new-rule script, but make sure you are extending the right base class.

  2. Implement the rule logic in S1234/rule.ts

    • Prefer using meta.messages to specify messages through messageIds. Message can be part of the RSPEC description, like here.
    • If writing a regex rule, use createRegExpRule

  3. If possible, implement quick fixes for the rule:

    • If the ESLint fix is at the root of the report (and not in a suggestion), add the message for the quick fix in rules/SXXXX/meta.ts.
    • Add a code fixture that should provide a quickfix in tests/linter/fixtures/wrapper/quickfixes/<ESLint-style rulekey>.{js,ts}. The following test asserts that the quickfix is enabled.

Testing a rule

We support 2 kinds of rule unit-tests: ESLint's RuleTester or our comment-based tests.

Comment-based testing

These tests are located in the rule folder and they MUST be named *.fixture.* (where the extension could be one of js, ts, jsx, tsx, vue). If options are to be passed to the tested rule, add a JSON file to the same directory named cb.options.json. The file must contain the array of options.

The contents of the test code have the following structure:

some.clean.code();
some.faulty.code(); // Noncompliant [[qf1,qf2,...]] {{Message to assert}}
//   ^^^^^^
// fix@qf1 {{Suggestion description}}
// edit@qf1 [[sc=1;ec=5]] {{text to replace line from [sc] column to [ec] column}}
faulty.setFaultyParam(true);
//     ^^^^^^^^^^^^^^< {{Optional secondary message to assert}}

The contents of the options file must be a valid JSON array:

// brace-style.json
['1tbs', { allowSingleLine: true }];

If your rule depends on a dependency declared in the package.json file, you can add the following clause to your test:

process.chdir(__dirname); // change current working dir to avoid the package.json lookup to up in the tree

and define multiple subfolders for your different settings like:

  • fixtures/setup-1/cb.test.ts
  • fixtures/setup-1/cb.fixture.ts
  • fixtures/setup-2/cb.test.ts
  • fixtures/setup-2/cb.fixture.ts

You can find an example at the bottom of this document.

Tests syntax

Given the above test snippet, issue messages ({{...}}) and quick fixes (if the rule provides them) are mandatory. The issue primary location (// ^^^^) and secondary location(s) (// ^^^<) are optional.

Noncompliant lines will be associated by default to the line of code where they are written. The syntax @line_number allows for an issue to be associated to another line:

// Noncompliant@2 [[qf1,qf2,...]] {{Optional message to assert}}
some.faulty.code();

Another option is to use relative line increments (@+line_increment) or decrements (@-line_decrement):

// Noncompliant@+1
some.faulty.code();

another.faulty.code();
// another comment
// Noncompliant@-2

Secondary locations

Secondary locations are part of Sonar issues. They provide additional context to the raised issue. In order to use them, you must call the toEncodedMessage() function when reporting the issue message like this:

context.report({
   node,
   message: toEncodedMessage(...),
});

In order to indicate secondary locations, you must use either // ^^^^^< or // ^^^^>, the arrow indicating whether the matching main location is either before or after the secondary one.

As stated before, the message is optional.

**/!** If you have used a secondary location in your test file, you must always report error messages using toEncodedMessage() in your rule, as it will be expecting it.

Quick fixes

Quick fixes refer to both ESLint Suggestions and Fixes. In our comment-based framework both use the same syntax, with the difference that a quick fix ID followed by an exclamation mark (!) will be internally treated as a fix with ESLint instead of as a suggestion. Please note that rules providing fixes MUST be tested always with fixes, otherwise the test will fail with the following error: The rule fixed the code. Please add 'output' property.. On the other side, it is optional to check against rule suggestions, meaning that even if a rule provides them, the tests can choose not to test their contents.

The fix@ comment referring to a quick fix provides the suggestion description and is optional. Eslint fixes do not support descriptions, meaning a quick fix ID declared with an exclamation mark (i.e. qf1!) must NOT have a fix@ matching comment (i.e. fix@qf1).

Each quick fix can have multiple editions associated to it. There are three different kind of operations to edit the code with quick fixes. Given a quick fix ID qf, these are the syntaxes used for each operation:

  • add@qf {{code to add}} Add the string between the double brackets to a new line in the code.
  • del@qf Remove the line
  • edit@qf1 [[sc=1;ec=5]] {{text to replace the range }} Edit the line from start column sc to end column ec (both 0-based) with the provided string between the double brackets. Alternatively, one can conveniently use only sc or ec. also optional, meaning this syntax can be used too:
    • edit@qf1 {{text to replace the whole line -do not include //Noncompliant comment- }}

The line affected in each of these operations will be the line of the issue to which the quick fix is linked to. It is possible to use the line modifier syntax (@[+|-]?line). When using line increments/decrements, keep in mind the base number is the issue line number, not the line of the quick fix edit comment. Example for rule brace-style:

//Noncompliant@+1 [[qf!]]
if (condition) {
  doSomething();
}
// edit@qf [[sc=16]] {{}}
// add@qf@+1 {{ doSomething()}}

The expected output is:

if (condition) {
  doSomething();
}

Let's go through the syntax used in this example:

  • The test provides a fix (note the ! after the ID qf).
  • The line //Noncompliant@+1 [[qf!]] means that in the following (@+1) line there is an issue for which we provide a quick fix.
  • The line // edit@qf [[sc=16]] {{}} is providing an edit to the same line of the issue, replacing the contents after column 16 (sc=16) by an empty string ({{}}). An alternative with the same effect would be // edit@qf {{if (condition) {}}, which would replace the whole line by if (condition) {.
  • Lastly, the line // add@qf@+1 {{ doSomething()}} will add a new line just after the issue line (@+1) with the contents &nbsp;doSomething()

Note that the length of the list of quick fixes cannot surpass the number of issues declared by N or the number of expected messages unless their matching issue is reassigned (see below).

Quick fixes IDs can be any string, they don't have to follow the qfN convention. The order of the list is important, as they will be assigned to the message in the matching position. If one provides 3 messages and 2 quick fixes which are not to be matched against first and second message, there are two options:

  • A dummy quick fix can be used as placeholder:
some.faulty.code(); // Noncompliant [[qf1,qf2,qf3]] {{message1}} {{message2}} {{message3}}
// edit@qf1 {{fix for message1}}
// edit@qf3 {{fix for message3}}
// qf2 is declared but never used --> ignored by the engine
  • Explicitly set the index (0-based) of the message to which the quick fix refers to with the syntax =index next to the quick fix ID:
some.faulty.code(); // Noncompliant [[qf1,qf3=2]] {{message1}} {{message2}} {{message3}}
// edit@qf1 {{fix for message1}}
// edit@qf3 {{fix for message3}}

This last syntax is also needed if multiple suggestions are to be provided for the same issue:

some.faulty.code(); // Noncompliant [[qf1,qf2=0]]
// fix@qf1 {{first alternative quickfix description}}
// edit@qf1 {{some.faulty?.code();}}
// fix@qf2 {{second alternative quickfix description}}
// edit@qf2 {{some.faulty && some.faulty.code();}}

Ruling

Make sure to run Ruling ITs for the new or updated rule (don't forget to rebuild the jar before that!).

If your rule does not raise any issue, you should write your own code that triggers your rule in:

  • its/sources/custom/jsts/S1234.js for code
  • its/sources/custom/jsts/tests/S1234.js for test code

You can simply copy and paste compliant and non-compliant examples from your RSPEC HTML description.

Examples

  • Security Hotspot implementation: PR
  • Quality rule implemented with quickfix: PR
  • Adding a rule already covered by ESLint or its plugins: PR
  • Adding a quickfix for rule covered by ESLint or its plugins: PR
  • Adding a rule covered by ESLint with an ESLint "fix" quick fix: PR
  • Decorate a rule covered by ESLint: PR
  • Merge 2 ESLint rules: PR
  • Use comment-based tests with package.json dependencies dependent rule: PR
  • Use ESLint's Rule tester with package.json dependencies dependent rule: PR

Rule Options Architecture

This section explains how rule options (configurations) work across the SonarJS stack.

Overview

There are two parallel workflows for requesting JS/TS analysis from Node.js:

1. SonarQube workflow (HTTP bridge via WebSocket):

SonarQube UI → Java Check Class → HTTP/WebSocket → analyzeProject() → ESLint Linter
                     ↓
              configurations() returns
              typed objects (int, boolean, etc.)

2. External workflow (gRPC - without SonarQube):

External Client → gRPC → transformers.ts → analyzeProject() → ESLint Linter
                              ↓
                   parseParamValue() converts
                   string params to typed values

The key difference is that SonarQube's Java side sends already-typed values via configurations(), while the gRPC endpoint receives string key-value pairs that need type parsing.

Each rule can have configurable options defined in several places that serve different purposes.

File Structure for a Rule

Each rule lives in packages/jsts/src/rules/SXXXX/ with these key files:

File Purpose
rule.ts Rule implementation (ESLint rule factory)
meta.ts Manual metadata: implementation, eslintId, schema, re-exports fields
config.ts Option definitions with fields array
generated-meta.ts Auto-generated: defaultOptions, sonarKey, scope, languages

Implementation Types

The implementation field in meta.ts determines how a rule is structured:

original

Rules written from scratch for SonarJS. If the rule accepts options, it defines its own JSON Schema in meta.ts. Rules without options don't need a schema or config.ts:

// S100/meta.ts
export const implementation = 'original';
export const eslintId = 'function-name';
export * from './config.js';
import type { JSONSchema4 } from '@typescript-eslint/utils/json-schema';
export const schema = {
  type: 'array',
  items: [{ type: 'object', properties: { format: { type: 'string' } } }],
} as const satisfies JSONSchema4;

decorated

Rules that wrap/extend an existing ESLint rule, adding SonarJS-specific behavior. They may optionally define a schema if needed:

// S109/meta.ts - no schema, uses external rule's schema at runtime
export const implementation = 'decorated';
export const eslintId = 'no-magic-numbers';
export const externalRules = [
  { externalPlugin: 'typescript-eslint', externalRule: 'no-magic-numbers' },
];
export * from './config.js';
// S107/meta.ts - explicit schema (when customization is needed)
export const implementation = 'decorated';
export const eslintId = 'max-params';
export const externalRules = [{ externalPlugin: 'eslint', externalRule: 'max-params' }];
export * from './config.js';
export const schema = {
  /* ... */
} as const satisfies JSONSchema4;

external

Rules that directly use an ESLint rule without modification. The schema is inherited from the external rule at runtime. Some external rules expose user-configurable options via config.ts (e.g., S103, S139, S1441):

// S106/meta.ts
export const implementation = 'external';
export const eslintId = 'no-console';
export const externalPlugin = 'eslint';
export * from './config.js';

The fields Array (config.ts)

The fields array is the source of truth for rule options. It defines:

  • ESLint field names
  • Default values (which determine types)
  • SonarQube UI descriptions
  • Key mappings when SQ and ESLint names differ
// S107/config.ts
export const fields = [
  [
    {
      field: 'max', // ESLint option name
      displayName: 'maximumFunctionParameters', // SonarQube UI name (optional)
      description: 'Maximum authorized...', // Shows in SQ UI
      default: 7, // Default value & type inference
    },
  ],
] as const satisfies ESLintConfiguration;

Field Properties

Property Required Purpose
field Yes ESLint/schema key name
default Yes Default value; also determines type (number, string, boolean, arrays)
description For SQ visibility Makes the option visible in SonarQube UI
displayName No SonarQube key if different from field
items For arrays { type: 'string' } or { type: 'integer' }
customDefault No Different default for SQ than JS/TS
fieldType No Override SQ field type (e.g., 'TEXT')

Making Options Visible in SonarQube

A field is only visible in SonarQube if it has a description.

The Java code generator (tools/generate-java-rule-classes.ts) checks:

function isSonarSQProperty(property): property is ESLintConfigurationSQProperty {
  return property.description !== undefined;
}

Fields without description are internal-only defaults that users cannot configure.

Example - S109 (Magic Numbers):

// S109/config.ts - NO descriptions, so not exposed in SQ
export const fields = [
  [
    { field: 'ignore', default: [0, 1, -1, 24, 60] }, // Internal only
    { field: 'ignoreDefaultValues', default: true }, // Internal only
  ],
] as const satisfies ESLintConfiguration;

Example - S2068 (Hardcoded Credentials):

// S2068/config.ts - HAS description, so visible in SQ
export const fields = [
  [
    {
      field: 'passwordWords',
      items: { type: 'string' },
      description: 'Comma separated list of words identifying potential passwords.',
      default: ['password', 'pwd', 'passwd', 'passphrase'],
    },
  ],
] as const satisfies ESLintConfiguration;

Key Mapping: SonarQube ↔ ESLint

When SonarQube and ESLint use different names for the same option:

SonarQube Key ESLint Key Mapping
maximumFunctionParameters max displayName: 'maximumFunctionParameters' in config.ts
format format No displayName needed (same name)

The transformation layer (packages/grpc/src/transformers.ts) handles this mapping at runtime.

JSON Schema vs fields

Aspect JSON Schema (meta.ts) fields (config.ts)
Purpose ESLint validation SQ UI + defaults + key mapping
Used by ESLint at runtime Java codegen, meta generation, linter
Required for original rules with options All rules with options
Defines Structure & constraints Defaults, descriptions, SQ keys

Important: The schema is for ESLint validation. The fields array provides default values and metadata for SonarQube integration. For original rules with options, both schema and fields must be kept in sync manually. For decorated/external rules, the schema is inherited from the external rule at runtime.

defaultOptions in generated-meta.ts

The npm run generate-meta script reads fields and generates defaultOptions:

// generated-meta.ts (auto-generated)
export const meta = {
  // ...
  defaultOptions: [
    { format: '^[_a-z][a-zA-Z0-9]*$' }, // From fields[0][0].default
  ],
};

This is extracted using the defaultOptions() helper from helpers/configs.ts.

How Options Flow at Runtime

SonarQube workflow (HTTP/WebSocket)

  1. Java Side: @RuleProperty fields are read, configurations() returns typed List<Object> (e.g., Map.of("max", 7))
  2. Transport: Gson serializes to JSON with proper types preserved
  3. Linter: linter.ts:createRulesRecord() merges defaults with user config: 
    rules[`sonarjs/${rule.key}`] = [
  'error',
  ...merge(defaultOptions(ruleMeta.fields), rule.configurations),
];

gRPC workflow (external clients)

  1. Client: Sends rule params as string key-value pairs via proto3
  2. Transformer: transformers.ts maps SQ keys → ESLint keys and parses string values to proper types
  3. Linter: Same merging as above

Type Parsing from Strings (gRPC only)

The gRPC workflow receives all param values as strings. The transformer parses them based on the default value type in fields:

Default Type Input String Parsed Result
number "5" 5
boolean "true" true
string "pattern" "pattern"
string[] "a,b,c" ["a", "b", "c"]
number[] "1,2,3" [1, 2, 3]

Adding Options to an Existing Rule

  1. Update config.ts with the new field in the fields array
  2. Add description if it should be visible in SonarQube
  3. Update meta.ts schema (for original/decorated rules) to match
  4. Run npm run generate-meta to update generated-meta.ts
  5. Run npm run generate-java-rule-classes to update Java check classes

Common Patterns

Object-style configuration (most common):

// config.ts
export const fields = [
  [
    { field: 'max', description: '...', default: 7 },
    { field: 'ignoreIIFE', description: '...', default: false },
  ],
] as const satisfies ESLintConfiguration;

// ESLint receives: [{ max: 7, ignoreIIFE: false }]

Primitive configuration:

// config.ts
export const fields = [
  { default: '^[a-z]+$' }, // Single non-array element
] as const satisfies ESLintConfiguration;

// ESLint receives: ['^[a-z]+$']

Array options (comma-separated in SQ):

// config.ts
export const fields = [
  [
    {
      field: 'passwordWords',
      items: { type: 'string' }, // Required for Java codegen
      description: 'Comma separated list...',
      default: ['password', 'pwd'],
    },
  ],
] as const satisfies ESLintConfiguration;

// SQ sends: "password,pwd,secret"
// ESLint receives: [{ passwordWords: ['password', 'pwd', 'secret'] }]

Misc

Issue tracking

Working on a rule

You don't need to make separate Jira tickets for RSPEC and rule implementation, a single one is good enough.

Add a link to the RSPEC PR from the SonarJS PR as shown in this example.