Skip to content

Latest commit

 

History

History
347 lines (236 loc) · 17.7 KB

DEV.md

File metadata and controls

347 lines (236 loc) · 17.7 KB

Developer Guide

Prerequisites

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

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:

npm run build

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/tests/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 diff -rq its/ruling/src/test/expected/jsts packages/ruling/tests/actual/jsts. For CSS, run diff -rq its/ruling/src/test/expected/css

If you have modified rules or their configuration, you will need to update the rule data used for these tests with: npm run update-ruling-data.

CSS (and old way for JS/TS)

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

To review the Ruling difference in SonarQube UI, put the breakpoint on assertThat(...) in {JsTs/CSS}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/resources/expected/.

From its/ruling/:

  • for JS/TS cp -R target/actual/jsts/ src/test/expected/jsts
  • for CSS cp -R target/actual/css/ src/test/expected/css

You can review the Ruling difference by running diff -rq src/test/expected/jsts target/actual/jsts 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
  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 rule metadata (JSON and HTML files) from RSPEC, by running this command from the project's root:
java -jar <location of rule-api jar> generate -rule S1234 [-branch <RSPEC branch>]
  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:

  • generates a Java check class for the rule S1234.java
  • generates a Java check test class for the rule S1234Test.java
  • generates a rules/S1234 folder
  • generates a rules/S1234/index.ts rule index file
  • generates a rules/S1234/rule.ts file for the rule implementation
  • generates a rules/S1234/cb.fixture.js comment-based test file (empty)
  • generates a rules/S1234/cb.test.js test launcher

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

  • 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.
    • In the generated metadata JSON file javascript-checks/src/main/resources/org/sonar/l10n/javascript/rules/javascript/S1234.json, add (one or both):
       "compatibleLanguages": [
         "JAVASCRIPT",
         "TYPESCRIPT"
       ]
  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.
    • Note that there are some helper functions in src/rules/helpers/, also searchable online
    • If writing a regex rule, use createRegExpRule
  3. If possible, implement quick fixes for the rule:

    • Add its rule key in src/linter/quickfixes/rules.ts
    • If the ESLint fix is at the root of the report (and not in a suggestion), add the message for the quick fix in src/linter/quickfixes/messages.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. Prefer comment-based tests as they are more readable!

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 N [[qf1,qf2,...]] {{Optional message to assert}}
//   ^^^^^^
// fix@qf1 {{Optional 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: the issue primary location (// ^^^^), issue messages ({{...}}), secondary location(s) (// ^^^<), issues count (N) and quick fixes are optional.

N is an integer defining the amount of issues will be reported in the line.

Only one of the methods ({{messageN}}+ OR N) to define the expected issues can be used in a Noncompliant line. If you set both N and messages, the framework will throw an error.

If no N nor messages are provided, the engine will expect one issue to be raised. Meaning, //Noncompliant is equivalent to //Noncompliant 1.

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

// Noncompliant@2 N [[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/jsts/custom/S1234.js for code
  • its/sources/jsts/custom/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

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.