Mojolicious::Guides::Testing - Web Application Testing Made Easy
This document is an introduction to testing web applications with Test::Mojo. Test::Mojo can be thought of as a module that provides all of the tools and testing assertions needed to test web applications in a Perl-ish way.
While Test::Mojo can be used to test any web application, it has shortcuts designed to make testing Mojolicious web applications easy and pain-free.
Please refer to the Test::Mojo documentation for a complete reference to many of the ideas and syntax introduced in this document.
A test file for a simple web application might look like:
use Mojo::Base -strict;
use Test::Mojo;
use Test::More;
# Start a Mojolicious app named "Celestial"
my $t = Test::Mojo->new('Celestial');
# Post a JSON document
$t->post_ok('/notifications' => json => {event => 'full moon'})
->status_is(201)
->json_is('/message' => 'notification created');
# Perform GET requests and look at the responses
$t->get_ok('/sunrise')
->status_is(200)
->content_like(qr/ am$/);
$t->get_ok('/sunset')
->status_is(200)
->content_like(qr/ pm$/);
# Post a URL-encoded form
$t->post_ok('/insurance' => form => {name => 'Jimmy', amount => '€3.000.000'})
->status_is(200);
# Use Test::More's like() to check the response
like $t->tx->res->dom->at('div#thanks')->text, qr/thank you/, 'thanks';
done_testing();
In the rest of this document we'll explore these concepts and others related to Test::Mojo.
Essentials every Mojolicious developer should know.
The Test::More module bundled with Perl includes several primitive test assertions, such as ok
, is
, isnt
, like
, unlike
, cmp_ok
, etc. An assertion "passes" if its expression returns a true value. The assertion method prints "ok" or "not ok" if an assertion passes or fails (respectively).
Test::Mojo supplies additional test assertions organized around the web application request/response transaction (transport, response headers, response bodies, etc.), and WebSocket communications.
One interesting thing of note: the return value of Test::Mojo object assertions is always the test object itself, allowing us to "chain" test assertion methods. So rather than grouping related test statements like this:
$t->get_ok('/frogs');
$t->status_is(200);
$t->content_like(qr/bullfrog/);
$t->content_like(qr/hypnotoad/);
Method chaining allows us to connect test assertions that belong together:
$t->get_ok('/frogs')
->status_is(200)
->content_like(qr/bullfrog/)
->content_like(qr/hypnotoad/);
This makes for a much more concise and coherent testing experience: concise because we are not repeating the invocant for each test, and coherent because assertions that belong to the same request are syntactically bound in the same method chain.
Occasionally it makes sense to break up a test to perform more complex assertions on a response. Test::Mojo exposes the entire transaction object so you can get all the data you need from a response:
$t->put_ok('/bees' => json => {type => 'worker', name => 'Karl'})
->status_is(202)
->json_has('/id');
# Pull out the id from the response
my $newbee = $t->tx->res->json('/id');
# Make a new request with data from the previous response
$t->get_ok("/bees/$newbee")
->status_is(200)
->json_is('/name' => 'Karl');
The Test::Mojo object is stateful. As long as we haven't started a new transaction by invoking one of the *_ok
methods, the request and response objects from the previous transaction are available in the Test::Mojo object:
# First transaction
$t->get_ok('/frogs?q=bullfrog' => {'Content-Type' => 'application/json'})
->status_is(200)
->json_like('/0/species' => qr/catesbeianus/i);
# Still first transaction
$t->content_type_is('application/json');
# Second transaction
$t->get_ok('/frogs?q=banjo' => {'Content-Type' => 'text/html'})
->status_is(200)
->content_like(qr/interioris/i);
# Still second transaction
$t->content_type_is('text/html');
This statefulness also enables Test::Mojo to handle sessions, follow redirects, and inspect past responses during a redirect.
The Test::Mojo object manages the Mojolicious application lifecycle (if a Mojolicious application class is supplied) as well as exposes the built-in Mojo::UserAgent object. To create a bare Test::Mojo object:
my $t = Test::Mojo->new;
This object initializes a Mojo::UserAgent object and provides a variety of test assertion methods for accessing a web application. For example, with this object, we could test any running web application:
$t->get_ok('https://www.google.com/')
->status_is(200)
->content_like(qr/search/i);
You can access the user agent directly if you want to make web requests without triggering test assertions:
my $tx = $t->ua->post('https://duckduckgo.com/html' => form => {q => 'hypnotoad'});
$tx->result->dom->find('a.result__a')->each(sub { say $_->text });
See Mojo::UserAgent for the complete API and return values.
If you pass the name of a Mojolicious application class (e.g., 'MyApp') to the Test::Mojo constructor, Test::Mojo will instantiate the class and start it, and cause it to listen on a random (unused) port number. Testing a Mojolicious application using Test::Mojo will never conflict with running applications, including the application you're testing.
The Mojo::UserAgent object in Test::Mojo will know where the application is running and make requests to it. Once the tests have completed, the Mojolicious application will be torn down.
# Listens on localhost:32114 (some unused TCP port)
my $t = Test::Mojo->new('Frogs');
To test a Mojolicious::Lite application, pass the file path to the application script to the constructor.
# Load application script relative to the "t" directory
use Mojo::File qw(curfile);
my $t = Test::Mojo->new(curfile->dirname->sibling('myapp.pl'));
The object initializes a Mojo::UserAgent object, loads the Mojolicious application, binds and listens on a free TCP port (e.g., 32114), and starts the application event loop. When the Test::Mojo object ($t
) goes out of scope, the application is stopped.
Relative URLs in the test object method assertions (get_ok
, post_ok
, etc.) will be sent to the Mojolicious application started by Test::Mojo:
# Rewritten to "http://localhost:32114/frogs"
$t->get_ok('/frogs');
Test::Mojo has a lot of handy shortcuts built into it to make testing Mojolicious or Mojolicious::Lite applications enjoyable.
Let's spin up a Mojolicious application using mojo generate app MyApp
. The mojo
utility will create a working application and a t
directory with a working test file:
$ mojo generate app MyApp
[mkdir] /my_app/script
[write] /my_app/script/my_app
[chmod] /my_app/script/my_app 744
...
[mkdir] /my_app/t
[write] /my_app/t/basic.t
...
Let's run the tests (we'll create the log
directory to quiet the application output):
$ cd my_app
$ mkdir log
$ prove -lv t
t/basic.t ..
ok 1 - GET /
ok 2 - 200 OK
ok 3 - content is similar
1..3
ok
All tests successful.
Files=1, Tests=3, 0 wallclock secs ( 0.03 usr 0.01 sys + 0.33 cusr 0.07 csys = 0.44 CPU)
Result: PASS
The boilerplate test file looks like this:
use Mojo::Base -strict;
use Test::More;
use Test::Mojo;
my $t = Test::Mojo->new('MyApp');
$t->get_ok('/')->status_is(200)->content_like(qr/Mojolicious/i);
done_testing();
Here we can see our application class name MyApp
is passed to the Test::Mojo constructor. Under the hood, Test::Mojo creates a new Mojo::Server instance, loads MyApp
(which we just created), and runs the application. We write our tests with relative URLs because Test::Mojo takes care of getting the request to the running test application (since its port may change between runs).
We can alter the behavior of our application using environment variables (such as MOJO_MODE
) and through configuration values. One nice feature of Test::Mojo is its ability to pass configuration values directly from its constructor.
Let's modify our application and add a "feature flag" to enable a new feature when the enable_weather
configuration value is set:
# Load configuration from hash returned by "my_app.conf"
my $config = $self->plugin('Config');
# Normal route to controller
$r->get('/')->to('example#welcome');
# NEW: this route only exists if "enable_weather" is set in the configuration
if ($config->{enable_weather}) {
$r->get('/weather' => sub ($c) {
$c->render(text => "It's hot! 🔥");
});
}
To test this new feature, we don't even need to create a configuration file—we can simply pass the configuration data to the application directly via Test::Mojo's constructor:
my $t = Test::Mojo->new(MyApp => {enable_weather => 1});
$t->get_ok('/')->status_is(200)->content_like(qr/Mojolicious/i);
$t->get_ok('/weather')->status_is(200)->content_like(qr/🔥/);
When we run these tests, Test::Mojo will pass this configuration data to the application, which will cause it to create a special /weather
route that we can access in our tests. Unless enable_weather
is set in a configuration file, this route will not exist when the application runs. Feature flags like this allow us to do soft rollouts of features, targeting a small audience for a period of time. Once the feature has been proven, we can refactor the conditional and make it a full release.
This example shows how easy it is to start testing a Mojolicious application and how to set specific application configuration directives from a test file.
Let's say we register a helper in our application to generate an HTTP Basic Authorization header:
use Mojo::Util qw(b64_encode);
app->helper(basic_auth => sub ($c, @values) {
return {Authorization => 'Basic ' . b64_encode join(':' => @values), ''};
});
How do we test application helpers like this? Test::Mojo has access to the application object, which allows us to invoke helpers from our test file:
my $t = Test::Mojo->new('MyApp');
is_deeply $t->app->basic_auth(bif => "Bif's Passwerdd"), {Authorization => 'Basic YmlmOkJpZidzIFBhc3N3ZXJkZA=='},
'correct header value';
Any aspect of the application (helpers, plugins, routes, etc.) can be introspected from Test::Mojo through the application object. This enables us to get deep test coverage of Mojolicious-based applications.
This section describes the basic test assertions supplied by Test::Mojo. There are four broad categories of assertions for HTTP requests:
HTTP requests
HTTP response status
HTTP response headers
HTTP response content/body
WebSocket test assertions are covered in "Testing WebSocket web services".
Test::Mojo has a Mojo::UserAgent object that allows it to make HTTP requests and check for HTTP transport errors. HTTP request assertions include get_ok
, post_ok
, etc. These assertions do not test whether the request was handled successfully, only that the web application handled the request in an HTTP compliant way.
You may also make HTTP requests using custom verbs (beyond GET
, POST
, PUT
, etc.) by building your own transaction object. See "Custom transactions" below.
To post a URL-encoded form to the /calls
endpoint of an application, we simply use the form
content type shortcut:
$t->post_ok('/calls' => form => {to => '+43.55.555.5555'});
Which will create the following HTTP request:
POST /calls HTTP/1.1
Content-Length: 20
Content-Type: application/x-www-form-urlencoded
to=%2B43.55.555.5555
The *_ok
HTTP request assertion methods accept the same arguments as their corresponding Mojo::UserAgent methods (except for the callback argument). This allows us to set headers and build query strings for authentic test situations:
$t->get_ok('/internal/personnel' => {Authorization => 'Token secret-password'} => form => {q => 'Professor Plum'});
which generates the following request:
GET /internal/personnel?q=Professor+Plum HTTP/1.1
Content-Length: 0
Authorization: Token secret-password
The form
content generator (see Mojo::UserAgent::Transactor) will generate a query string for GET
requests and application/x-www-form-urlencoded
or multipart/form-data
for POST requests.
While these *_ok
assertions make the HTTP requests we expect, they tell us little about how well the application handled the request. The application we're testing might have returned any content-type, body, or HTTP status code (200, 302, 400, 404, 500, etc.) and we wouldn't know it.
Test::Mojo provides assertions to test almost every aspect of the HTTP response, including the HTTP response status code, the value of the Content-Type
header, and other arbitrary HTTP header information.
While not technically an HTTP header, the status line is the first line in an HTTP response and is followed by the response headers. Testing the response status code is common in REST-based and other web applications that use the HTTP status codes to broadly indicate the type of response the server is returning.
Testing the status code is as simple as adding the status_is
assertion:
$t->post_ok('/doorbell' => form => {action => 'ring once'})
->status_is(200);
Along with status_isnt
, this will cover most needs. For more elaborate status code testing, you can access the response internals directly:
$t->post_ok('/doorbell' => form => {action => 'ring once'});
is $t->tx->res->message, 'Moved Permanently', 'try next door';
Test::Mojo allows us to inspect and make assertions about HTTP response headers. The Content-Type
header is commonly tested and has its own assertion:
$t->get_ok('/map-of-the-world.pdf')
->content_type_is('application/pdf');
This is equivalent to the more verbose:
$t->get_ok('/map-of-the-world.pdf')
->header_is('Content-Type' => 'application/pdf');
We can test for multiple headers in a single response using method chains:
$t->get_ok('/map-of-the-world.pdf')
->content_type_is('application/pdf')
->header_isnt('Compression' => 'gzip')
->header_unlike('Server' => qr/IIS/i);
Test::Mojo also exposes a rich set of assertions for testing the body of a response, whether that body be HTML, plain-text, or JSON. The content_*
methods look at the body of the response as plain text (as defined by the response's character set):
$t->get_ok('/scary-things/spiders.json')
->content_is('{"arachnid":"brown recluse"}');
Although this is a JSON document, content_is
treats it as if it were a text document. This may be useful for situations where we're looking for a particular string and not concerned with the structure of the document. For example, we can do the same thing with an HTML document:
$t->get_ok('/scary-things/spiders.html')
->content_like(qr{<title>All The Spiders</title>});
But because Test::Mojo has access to everything that Mojo::UserAgent does, we can introspect JSON documents as well as DOM-based documents (HTML, XML) with assertions that allow us to check for the existence of elements as well as inspect the content of text nodes.
Test::Mojo's Mojo::UserAgent has access to a JSON parser, which allows us to test to see if a JSON response contains a value at a location in the document using JSON pointer syntax:
$t->get_ok('/animals/friendly.json')
->json_has('/beings/jeremiah/age');
This assertion tells us that the friendly.json
document contains a value at the /beings/jeremiah/age
JSON pointer location. We can also inspect the value at JSON pointer locations:
$t->get_ok('/animals/friendly.json')
->json_has('/beings/jeremiah/age')
->json_is('/beings/jeremiah/age' => 42)
->json_like('/beings/jeremiah/species' => qr/bullfrog/i);
JSON pointer syntax makes testing JSON responses simple and readable.
We can also inspect HTML and XML responses using the Mojo::DOM parser in the user agent. Here are a few examples from the Test::Mojo documentation:
$t->text_is('div.foo[x=y]' => 'Hello!');
$t->text_is('html head title' => 'Hello!', 'right title');
The Mojo::DOM parser uses the CSS selector syntax described in Mojo::DOM::CSS, allowing us to test for values in HTML and XML documents without resorting to typically verbose and inflexible DOM traversal methods.
This section describes some complex (but common) testing situations that Test::Mojo excels in making simple.
The Mojo::UserAgent object in Test::Mojo can handle HTTP redirections internally to whatever level you need. Let's say we have a web service that redirects /1
to /2
, /2
redirects to /3
, /3
redirects to /4
, and /4
redirects to /5
:
GET /1
returns:
302 Found
Location: /2
and:
GET /2
returns:
302 Found
Location: /3
and so forth, up to /5
:
GET /5
which returns the data we wanted:
200 OK
{"message":"this is five"}
We can tell the user agent in Test::Mojo how to deal with redirects. Each test is making a request to GET /1
, but we vary the number of redirects the user agent should follow with each test:
my $t = Test::Mojo->new;
$t->get_ok('/1')
->header_is(location => '/2');
$t->ua->max_redirects(1);
$t->get_ok('/1')
->header_is(location => '/3');
$t->ua->max_redirects(2);
$t->get_ok('/1')
->header_is(location => '/4');
# Look at the previous hop
is $t->tx->previous->res->headers->location, '/3', 'previous redirect';
$t->ua->max_redirects(3);
$t->get_ok('/1')
->header_is(location => '/5');
$t->ua->max_redirects(4);
$t->get_ok('/1')
->json_is('/message' => 'this is five');
When we set max_redirects
, it stays set for the life of the test object until we change it.
Test::Mojo's handling of HTTP redirects eliminates the need for making many, sometimes an unknown number, of redirections to keep testing precise and easy to follow (ahem).
We can use Test::Mojo to test applications that keep session state in cookies. By default, the Mojo::UserAgent object in Test::Mojo will manage session for us by saving and sending cookies automatically, just like common web browsers:
use Mojo::Base -strict;
use Test::More;
use Test::Mojo;
my $t = Test::Mojo->new('MyApp');
# No authorization cookie
$t->get_ok('/')
->status_is(401)
->content_is('Please log in');
# Application sets an authorization cookie
$t->post_ok('/login' => form => {password => 'let me in'})
->status_is(200)
->content_is('You are logged in');
# Sends the cookie from the previous transaction
$t->get_ok('/')
->status_is(200)
->content_like(qr/You logged in at \d+/);
# Clear the cookies
$t->reset_session;
# No authorization cookie again
$t->get_ok('/')
->status_is(401)
->content_is('Please log in');
We can also inspect cookies in responses for special values through the transaction's response (Mojo::Message::Response) object:
$t->get_ok('/');
like $t->tx->res->cookie('smarty'), qr/smarty=pants/, 'cookie found';
Let's say we have an application that responds to a new HTTP verb RING
and to use it we must also pass in a secret cookie value. This is not a problem. We can test the application by creating a Mojo::Transaction object, setting the cookie (see Mojo::Message::Request), then passing the transaction object to request_ok
:
# Use custom "RING" verb
my $tx = $t->ua->build_tx(RING => '/doorbell');
# Set a special cookie
$tx->req->cookies({name => 'Secret', value => "don't tell anybody"});
# Make the request
$t->request_ok($tx)
->status_is(200)
->json_is('/status' => 'ding dong');
While the message flow on WebSocket connections can be rather dynamic, it more often than not is quite predictable, which allows this rather pleasant Test::Mojo WebSocket API to be used:
use Mojo::Base -strict;
use Test::More;
use Test::Mojo;
# Test echo web service
my $t = Test::Mojo->new('EchoService');
$t->websocket_ok('/echo')
->send_ok('Hello Mojo!')
->message_ok
->message_is('echo: Hello Mojo!')
->finish_ok;
# Test JSON web service
$t->websocket_ok('/echo.json')
->send_ok({json => {test => [1, 2, 3]}})
->message_ok
->json_message_is('/test' => [1, 2, 3])
->finish_ok;
done_testing();
Because of their inherent asynchronous nature, testing WebSocket communications can be tricky. The Test::Mojo WebSocket assertions serialize messages via event loop primitives. This enables us to treat WebSocket messages as if they were using the same request-response communication pattern we're accustomed to with HTTP.
To illustrate, let's walk through these tests. In the first test, we use the websocket_ok
assertion to ensure that we can connect to our application's WebSocket route at /echo
and that it's "speaking" WebSocket protocol to us. The next send_ok
assertion tests the connection again (in case it closed, for example) and attempts to send the message Hello Mojo!
. The next assertion, message_ok
, blocks (using the Mojo::IOLoop singleton in the application) and waits for a response from the server. The response is then compared with 'echo: Hello Mojo!'
in the message_is
assertion, and finally we close and test our connection status again with finish_ok
.
The second test is like the first, but now we're sending and expecting JSON documents at /echo.json
. In the send_ok
assertion we take advantage of Mojo::UserAgent's JSON content generator (see Mojo::UserAgent::Transactor) to marshal hash and array references into JSON documents, and then send them as a WebSocket message. We wait (block) for a response from the server with message_ok
. Then because we're expecting a JSON document back, we can leverage json_message_ok
which parses the WebSocket response body and returns an object we can access through Mojo::JSON::Pointer syntax. Then we close (and test) our WebSocket connection.
Testing WebSocket servers does not get any simpler than with Test::Mojo.
If you see that you're writing a lot of test assertions that aren't chainable, you may benefit from writing your own test assertions. Let's say we want to test the Location
header after a redirect. We'll create a new class with Role::Tiny that implements a test assertion named location_is
:
package Test::Mojo::Role::Location;
use Mojo::Base -role, -signatures;
sub location_is ($self, $value, $desc = "Location: $value") {
return $self->test('is', $self->tx->res->headers->location, $value, $desc);
}
1;
When we make new test assertions using roles, we want to use method signatures that match other *_is
methods in Test::Mojo, so here we accept the test object, the value to compare, and an optional description.
We assign a default description value ($desc
), then we use "test" in Test::Mojo to compare the location header with the expected header value, and finally propagates the Test::Mojo object for method chaining.
With this new package, we're ready to compose a new test object that uses the role:
my $t = Test::Mojo->with_roles('+Location')->new('MyApp');
$t->post_ok('/redirect/mojo' => json => {message => 'Mojo, here I come!'})
->status_is(302)
->location_is('http://mojolicious.org')
->or(sub { diag 'I miss tempire.' });
In this section we've covered how to add custom test assertions to Test::Mojo with roles and how to use those roles to simplify testing.
You can continue with Mojolicious::Guides now or take a look at the Mojolicious wiki, which contains a lot more documentation and examples by many different authors.
If you have any questions the documentation might not yet answer, don't hesitate to ask in the Forum, or on IRC.