Most tests in Node.js core are JavaScript programs that exercise a functionality
provided by Node.js and check that it behaves as expected. Tests should exit
with code 0
on success. A test will fail if:
- It exits by setting
process.exitCode
to a non-zero number.- This is usually done by having an assertion throw an uncaught Error.
- Occasionally, using
process.exit(code)
may be appropriate.
- It never exits. In this case, the test runner will terminate the test because it sets a maximum time limit.
Add tests when:
- Adding new functionality.
- Fixing regressions and bugs.
- Expanding test coverage.
Let's analyze this basic test from the Node.js test suite:
'use strict'; // 1
const common = require('../common'); // 2
// This test ensures that the http-parser can handle UTF-8 characters // 4
// in the http header. // 5
const assert = require('assert'); // 7
const http = require('http'); // 8
const server = http.createServer(common.mustCall((req, res) => { // 10
res.end('ok'); // 11
})); // 12
server.listen(0, () => { // 13
http.get({ // 14
port: server.address().port, // 15
headers: { 'Test': 'Düsseldorf' } // 16
}, common.mustCall((res) => { // 17
assert.strictEqual(res.statusCode, 200); // 18
server.close(); // 19
})); // 20
}); // 21
'use strict';
const common = require('../common');
The first line enables strict mode. All tests should be in strict mode unless the nature of the test requires that the test run without it.
The second line loads the common
module. The common
module is a helper
module that provides useful tools for the tests.
Even if a test uses no functions or other properties exported by common
,
the test should still include the common
module before any other modules. This
is because the common
module includes code that will cause a test to fail if
the test leaks variables into the global space. In situations where a test uses
no functions or other properties exported by common
, include it without
assigning it to an identifier:
require('../common');
// This test ensures that the http-parser can handle UTF-8 characters
// in the http header.
A test should start with a comment containing a brief description of what it is designed to test.
const assert = require('assert');
const http = require('http');
The test checks functionality in the http
module.
Most tests use the assert
module to confirm expectations of the test.
The require statements are sorted in
ASCII order (digits, upper
case, _
, lower case).
This is the body of the test. This test is simple, it just tests that an
HTTP server accepts non-ASCII
characters in the headers of an incoming
request. Interesting things to notice:
- If the test doesn't depend on a specific port number, then always use 0 instead of an arbitrary value, as it allows tests to run in parallel safely, as the operating system will assign a random port. If the test requires a specific port, for example if the test checks that assigning a specific port works as expected, then it is ok to assign a specific port number.
- The use of
common.mustCall
to check that some callbacks/listeners are called. - The HTTP server closes once all the checks have run. This way, the test can exit gracefully. Remember that for a test to succeed, it must exit with a status code of 0.
Avoid timers unless the test is specifically testing timers. There are multiple reasons for this. Mainly, they are a source of flakiness. For a thorough explanation go here.
In the event a test needs a timer, consider using the
common.platformTimeout()
method. It allows setting specific timeouts
depending on the platform. For example:
const timer = setTimeout(fail, common.platformTimeout(4000));
will create a 4-second timeout on most platforms but a longer timeout on slower platforms.
Make use of the helpers from the common
module as much as possible.
One interesting case is common.mustCall
. The use of common.mustCall
may
avoid the use of extra variables and the corresponding assertions. Let's explain
this with a real test from the test suite.
'use strict';
require('../common');
const assert = require('assert');
const http = require('http');
let request = 0;
let response = 0;
process.on('exit', function() {
assert.equal(request, 1, 'http server "request" callback was not called');
assert.equal(response, 1, 'http request "response" callback was not called');
});
const server = http.createServer(function(req, res) {
request++;
res.end();
}).listen(0, function() {
const options = {
agent: null,
port: this.address().port
};
http.get(options, function(res) {
response++;
res.resume();
server.close();
});
});
This test could be greatly simplified by using common.mustCall
like this:
'use strict';
const common = require('../common');
const http = require('http');
const server = http.createServer(common.mustCall(function(req, res) {
res.end();
})).listen(0, function() {
const options = {
agent: null,
port: this.address().port
};
http.get(options, common.mustCall(function(res) {
res.resume();
server.close();
}));
});
Some tests will require running Node.js with specific command line flags set. To
accomplish this, add a // Flags:
comment in the preamble of the
test followed by the flags. For example, to allow a test to require some of the
internal/*
modules, add the --expose-internals
flag.
A test that would require internal/freelist
could start like this:
'use strict';
// Flags: --expose-internals
require('../common');
const assert = require('assert');
const freelist = require('internal/freelist');
When writing assertions, prefer the strict versions:
assert.strictEqual()
overassert.equal()
assert.deepStrictEqual()
overassert.deepEqual()
When using assert.throws()
, if possible, provide the full error message:
assert.throws(
() => {
throw new Error('Wrong value');
},
/^Error: Wrong value$/ // Instead of something like /Wrong value/
);
For performance considerations, we only use a selected subset of ES.Next
features in JavaScript code in the lib
directory. However, when writing
tests, for the ease of backporting, it is encouraged to use those ES.Next
features that can be used directly without a flag in
all maintained branches. node.green lists available features
in each release.
For example:
let
andconst
overvar
- Template literals over string concatenation
- Arrow functions when appropriate
Test files are named using kebab casing. The first component of the name is
test
. The second is the module or subsystem being tested. The third is usually
the method or event name being tested. Subsequent components of the name add
more information about what is being tested.
For example, a test for the beforeExit
event on the process
object might be
named test-process-before-exit.js
. If the test specifically checked that arrow
functions worked correctly with the beforeExit
event, then it might be named
test-process-before-exit-arrow-functions.js
.
Some of the tests for the WHATWG URL implementation (named
test-whatwg-url-*.js
) are imported from the Web Platform Tests Project.
These imported tests will be wrapped like this:
/* The following tests are copied from WPT. Modifications to them should be
upstreamed first. Refs:
https://github.com/w3c/web-platform-tests/blob/8791bed/url/urlsearchparams-stringifier.html
License: http://www.w3.org/Consortium/Legal/2008/04-testsuite-copyright.html
*/
/* eslint-disable */
// Test code
/* eslint-enable */
To improve tests that have been imported this way, please send
a PR to the upstream project first. When the proposed change is merged in
the upstream project, send another PR here to update Node.js accordingly.
Be sure to update the hash in the URL following WPT Refs:
.
C++ code can be tested using Google Test. Most features in Node.js can be tested using the methods described previously in this document. But there are cases where these might not be enough, for example writing code for Node.js that will only be called when Node.js is embedded.
The unit test should be placed in test/cctest
and be named with the prefix
test
followed by the name of unit being tested. For example, the code below
would be placed in test/cctest/test_env.cc
:
#include "gtest/gtest.h"
#include "node_test_fixture.h"
#include "env.h"
#include "node.h"
#include "v8.h"
static bool called_cb = false;
static void at_exit_callback(void* arg);
class EnvTest : public NodeTestFixture { };
TEST_F(EnvTest, RunAtExit) {
v8::HandleScope handle_scope(isolate_);
v8::Local<v8::Context> context = v8::Context::New(isolate_);
node::IsolateData* isolateData = node::CreateIsolateData(isolate_, uv_default_loop());
Argv argv{"node", "-e", ";"};
auto env = Environment:CreateEnvironment(isolateData, context, 1, *argv, 2, *argv);
node::AtExit(at_exit_callback);
node::RunAtExit(env);
EXPECT_TRUE(called_cb);
}
static void at_exit_callback(void* arg) {
called_cb = true;
}
Next add the test to the sources
in the cctest
target in node.gyp:
'sources': [
'test/cctest/test_env.cc',
...
],
The test can be executed by running the cctest
target:
$ make cctest
There is a test fixture named node_test_fixture.h
which can be included by
unit tests. The fixture takes care of setting up the Node.js environment
and tearing it down after the tests have finished.
It also contains a helper to create arguments to be passed into Node.js. It will depend on what is being tested if this is required or not.