This is the documentation for Object Oriented approach used in automated GUI testing for (Open)SUSE products.
- Context
- Overview
SUSE Products changes and evolves across versions, and we are expected to write tests for various versions of the same product. However, we may still want to re-use the same business logic and avoid to write different code for each specific case.
Throughout the course of the project history, we attempted to solve this issue with many approaches; in the following order:
-
The "naive" way: leads to dealing with a lot of if-else conditions "Spaghetti Code" with things like
if (is_food AND (is_a_banana OR ! is_a_fruit))
while it's intuitive to write, and the inner logic of test differentiation is directly modeled in the code, it can be exhausting to read and maintain as the conditions sum-up over time, and therefore should be avoided. -
Passing variables from outside (external test data): tests are still written in a generic way but with a Data Driven design. This solution is simpler than the previous, but still complex, as the number of variable combinations increases exponentially, and leads to increase difficulty on debugging and understanding the control flow of the program. Consider also that OpenQA variables can be defined at many levels: medium, job group, schedule, command line...etc. And by looking at a job result it's not always clear where a variable comes from.
-
Very simple modules doing only one specific task, with schedules adapted for each case. Variables (test data) are passed from the test module to some libraries only if needed. The schedule is used to handle the various scenarios rather than conditions in the code, and the name of the module accurately describes what it does. As the number of special situations grows, use inheritance and interchangeable test data to simplify the code while maintaining backward compatibility. With the combined usage of declarative scheduling we obtain certain advantages: code is straightforward to write, understand, and debug, while the complexity is partially shifted to the YAML schedule. We accept the trade-off of an increased number of (simple) modules to build and maintain, also because the development work can be evenly distributed across the team.
In the following guide let's explain in detail the last solution.
The framework proposed here is based on Page Object Desing Pattern, implemented using "Old school" object-oriented perl with a certain adaptation related to the environment-specific demands.
It is broken on several Layers. The interactions between the layers could be represented with the following diagram.
main.pm is an entry point for all the tests in openQA, the distribution is set here with DistributionProvider.
use testapi;
...
testapi::set_distribution(DistributionProvider->provide());
DistributionProvider (lib/DistrubutionProvider.pm) is a factory that returns the required "distribution" depending on openQA environment variables ('VERSION', 'DISTRI'). Currently, Tumbleweed is returned as the default one if none specified, following "Factory First" rule.
package DistributionProvider;
...
sub provide {
return Distribution::Sle::15->new() if version_utils::is_sle('15+');
return Distribution::Sle::12->new() if version_utils::is_sle('12+');
return Distribution::Opensuse::Leap::15->new() if version_utils::is_leap('15.0+');
return Distribution::Opensuse::Leap::42->new() if version_utils::is_leap('42.0+');
return Distribution::Opensuse::Tumbleweed->new();
}
Each product has its class under lib/Distribution. The parent class is Tumbleweed.pm, as all other products are derived from it, so what works on TW usually works everywhere else. When it is not the case, a specific method can be created in the appropriate Distribution class in lib/Distribution:
.
├── Opensuse
│ ├── Leap
│ │ ├── 15.pm
│ │ └── 42.pm
│ └── Tumbleweed.pm
└── Sle
├── 12.pm
├── 15_current.pm
├── 15sp0.pm
└── 15sp2.pm
main.pm calls a scheduled Test Module using declarative scheduling (please don't schedule modules in main*.pm directly !), the module uses the factory in DistributionProvider to determine what product version we are using, then the Distribution file (eg 12.pm) determines what business logic applies for this distribution, so in practice, what controller has to be used:
Important: Test Module must be inherited from opensusebasetest or one of its children to have an access to the Distribution.
# Test module
use parent 'opensusebasetest';
sub run {
my $partitioner = $testapi::distri->get_partitioner();
$partitioner->some_method_defined_in_controller();
}
The method get_partitioner calls the right controller according to the product version, eg for Tumbleweed:
package Distribution::Opensuse::Tumbleweed;
use Installation::Partitioner::LibstorageNG::v4_3::ExpertPartitionerController;
sub get_expert_partitioner {
return Installation::Partitioner::LibstorageNG::v4_3::ExpertPartitionerController->new();
}
In this example, the path to the proper partitioner's controller for Tumbleweed is
libsorageNG/v4.3/ExpertPartitionerController.pm
. The class "ExpertPartitionerController"
re-uses common parts from older versions of the product.
.
├── ExpertPartitionerPage.pm # Base page class used by ExpertPartitionerController
├── FormattingOptionsPage.pm # Another page that can also be used by the controller
├── Libstorage
│ ├── ExpertPartitionerController.pm # libstorage (and base) controller class. Calls methods from the page(s)
├── LibstorageNG
│ ├── ExpertPartitionerPage.pm # Page class common to v3 and v4, derived from the base page class.
│ ├── v3
│ │ └── ExpertPartitionerController.pm # Controller class derived from libstorage
│ ├── v4
│ │ └── ExpertPartitionerController.pm # Controller class derived from v3
│ └── v4_3
│ ├── ExpertPartitionerController.pm # Controller class derived from v4, used here for Tumbleweed.
│ ├── ExpertPartitionerPage.pm # Page class derived from LibstorageNG v3/4
- Note: The example is taken from a real case, libstorage was the "expert partitioner" for SLE12 / Leap 42, it was re-written and called libstorage-ng to become the default for the next product versions, and then it kept evolving. This model proved to be adaptive and backward-compatible in such context, we can still schedule a same test module on both SLE12 and the latest Tumbleweed, without any conditions in the code: we just need to adjust the test data and maybe some needles.
Abstract: All direct interactions with the GUI are defined in a page, those methods are grouped in higher-level methods within a controller to form "business logic" actions, and the methods from the controller are called from a test module.
- Note: The page-object model only applies to interactions with GUI, so any interaction with the command-line can take place directly in the test module and use testapi directly. But the use of declarative scheduling and external test data should always be considered in order to avoid "spaghetti code".
Test Module is a layer containing test case steps that need to be executed on the system under test (SUT).
The variables to be used in the test are defined here before being passed to the other layers.
In our test module, we should not care about the specifics of each SUT. If we want to format some disks,
We may create a method called format_disks
, defined in a controller, that can be re-used
in all variants of products. It should describe a business
logic workflow, using a verb and a business-logic object (examples:
select_dynamic_address_for_ethernet, create_encrypted_partition).
Instead of adding conditions in the module, we should use different test data
for each specific case and create some sub-classes (page and/or controller) for each specific workflow where needed.
All the test data should be provided at this level, preferably not hard-coded. Do not provide any test data in Controller or Page layers.
Example with hard-coded test data:
# Should partitioner enable separate home partition is set in Test Module.
sub run {
my $partitioner = $testapi::distri->get_partitioner();
$partitioner->edit_proposal(has_separate_home => 1);
}
Or if we are using external test data (recommended)
sub run {
my $test_data = get_test_suite_data()
my $partitioner = $testapi::distri->get_partitioner();
foreach my $partition (@{test_data->{partitions}) { $partitioner->partition_disk($partition) }
}
Test Module is not allowed to use os-autoinst testapi functions directly. It should use methods, provided by Controller layer instead. This allows to hide the details of the UI structure and operate with the business logic the system provides.
This being said, for logging purposes, "diag", "record_info" or "save_screenshot", "record_soft_failure" can be used in the test module when it cannot be done at the page level.
Test Module is able to interact only with the Controller layer.
The controller is a layer that provides methods to interact with the system under test in business terms. Those methods combine together lower-level methods from the page layer, and are used by the test modules.
Example:
For instance, there might be a test, that should create an encrypted partition. In this example, methods such as "enter_password" are defined in the page.
sub create_encrypted_partition {
my ($self) = @_;
$self->get_partitioning_scheme_page()->select_enable_disk_encryption_checkbox();
$self->get_partitioning_scheme_page()->enter_password();
$self->get_partitioning_scheme_page()->enter_password_confirmation();
$self->get_partitioning_scheme_page()->press_next();
}
Then it is called in all the Test Modules, where the encrypted partition needs to be created.
sub run {
my $partitioner = $testapi::distri->get_partitioner();
$partitioner->create_encrypted_partition();
}
Do not define any test data in Controller layer as it could make test maintenance more complicated. Use the test data passed from the Test Module layer instead.
As everywhere else, avoid conditions in general as much as possible. Try instead to create another appropriate business-logic method and/or think how to organize the test data according to what is expected.
Example:
To make an action depending on a variable, we could do something like this:
sub create_encrypted_partition {
my ($self, $args) = @_;
if ($args->{is_lvm}) {
$self->get_partitioning_scheme_page()->select_lvm_checkbox();
}
[...]
$self->get_expert_partitioner_page()->press_next()
}
But a better way is probably to create an appropriate method:
sub create_encrypted_lvm_partition {
my ($self, $args) = @_;
$self->get_partitioning_scheme_page()->select_lvm_checkbox();
[...]
$self->get_expert_partitioner_page()->press_next()
}
If you need to use a variable, you can define it in the test data and pass it as follows from the test module:
sub create_filesystem {
my ($self, $args) = @_;
$self->get_filesystem_options_page()->select_filesystem($args->{filesystem});
$self->get_expert_partitioner_page()->press_next()
}
-
Do not use testapi methods that communicates with the SUT (e.g.
send_keys
,assert_screen
). Wrap them into Page methods with the meaningful names instead. -
Using
get_var
to change the flow of a test or get a data for the test should be avoided as much as possible (e.g. to decide whether check or uncheck checkbox, use method parameters instead and pass the data from Test Module). -
Local libs from os-autoins-distri-opensuse should not be used here either, but rather in the test module.
-
"diag", "record_info" or "save_screenshot", "record_soft_failure" could be used here, as this does not change the test flow, since it is just used for logging. But consider to put it in the module or page instead.
It knows only about Pages and is called by the Test Module.
The page layer is where the direct interactions with the UI (like pressing a button) take place, so here we can use testapi. The layer introduces accessing methods to elements of the page, or section of the page.
All the page classes (but not the page element or section classes) can inherit from a base page (e.g. in case of pages for installation wizard, it is Installation::WizardPage). If some elements or sections are common for several pages (like OK/Cancel buttons in a wizard) the corresponding accessing methods may be written in this base page or in a separate class, to be used by all other pages.
Unlike the classic POM approach, methods of Page layer in the Framework are not returning Objects. This compromise was introduced because the behavior of SUT may vary depending on the steps, that were done in the previous test Modules and also due to a large set of versions, which behavior also may differs.
Example:
package Installation::Partitioner::Libstorage::PasswordDialog;
sub press_ok {
assert_screen(ENTER_PASSWORD_DIALOG);
send_key('alt-o');
}
Do not provide any test data in Page layer. Use the test data passed from the Test Module layer instead.
This is the only layer having full access to testapi.
NOTE: Using
get_var
or similar methods to change the flow of a test should be avoided (e.g. to decide whether select checkbox or not by checking openQA variable. Please, use method parameters instead).
It should not use methods of another layers. It just provides page accessing methods for Controller layer.
-
Package and Class names should be nouns, using mixed case with the first letter of each word capitalized.
Example:
package Installation::Partitioner::Libstorage::EditProposalSettingsController;
-
Method names should be verbs, using lowercase with the underscores between the words.
Example:
sub get_password_dialog; sub edit_proposal;
-
Variable names should be lowercase with the underscores between the words.
Example:
my $is_lvm; my $filesystem;
-
Constant names should be uppercase with the underscores between the words.
Example:
use constant { SUGGESTED_PARTITIONING_PAGE => 'inst-suggested-partitioning-step', LVM_ENCRYPTED_PARTITION_IN_LIST => 'partitioning-encrypt-activated' };
-
Methods in the controllers should describe a business-logic action using a verb and business-logic object (examples: create_encrypted_partition or_select_dynamic_address_for_ethernet_). Those should not be things like "select_this_button" as those functions should be in pages.
-
Methods returning true/false or variables that store them, should be named beginning with is_ or has_.
Example:
sub is_lvm; sub has_separate_home; my $is_checkbox_checked; my $has_license_agreement;
-
Note: such methods should be defined only in pages
Use named arguments in hash reference if Method has more than one argument.
Example:
sub edit_proposal {
($self, $args_ref) = @;
my $is_lvm = $args_ref->{is_lvm};
my $has_separate_home = $args_ref->{has_separate_home};
...
}
# Then usage:
edit_proposal({is_lvm => 1, has_separate_home => 1});
To avoid confusion identifiers that refer to UI widgets should be prefixed with a three-letter abbreviation for the widget followed by an underscore '_'.
Building rules for prefixes
- prefixes have the length of 3 letters, all in lower case
- all letters are consonants, except you have not enough, then take vowels as well (e.g. the Tab widget transforms to 'tab')
- if a widget name consists of 2 words in CamelCase type style then the first two characters come from the first word, the 3rd character from the second word
- avoid repetition of the same letter, therefore it is "btn" for Button and not "btt"
Applying the rules above gives us the following prefixes:
Prefix | Widget | Example |
---|---|---|
btn | Button | btn_ok |
chb | CheckBox | chb_autologin |
cmb | ComboBox | cmb_filesystem |
its | ItemSelector | its_keyboard_layout |
lbl | Label | lbl_warning |
mnc | MenuCollection | mnc_operation |
prb | ProgressBar | prb_total_packages |
rdb | RadioButton | rdb_skip_registration |
rct | RichText | rct_welcome |
slb | SelectionBox | slb_addons |
tbl | Table | tbl_devices |
txb | TextBox | txb_maximum_channel |
tre | Tree | tre_system_view |
tab | Tab | tab_boot_loader_settings |
So, basically a new test requires to have at least one package/class per each layer to be created (or updated if the required class already exists).
Let's assume there might be a new test to create an account in the system during installation.
{project_root}/tests/installation/create_account.pm
use strict;
use warnings;
use parent "installbasetest";
sub run {
my $test_data = get_test_suite_data();
my $user_settings_widget = $testapi::distri->get_user_settings_widget();
$user_settings_widget->create_user({
username => $test_data->{username},
user_full_name => $test_data->{user_full_name}
});
}
1;
In the module we do not know how to "create an account", as it may vary across products. The most important here is to set clear expectations and create a data structure accordingly. In this case we could create a very simple test data file containing:
username: frankie
user_full_name: "Frank Einstein"
- tip: Obviously this is a very simple data structure. if something more complex is needed, it can be helpful to use Data::Dumper to verify that the structure gets interpreted as per our expectations.
{project_root}/lib/Installation/UserSettingsController.pm
package Installation::UserSettingsController
use strict;
use warnings;
use parent 'Installation::WizardPage';
use Installation::UserSettingsPage;
sub new {
my ($class, $args) = @_;
my $self = bless {
user_settings_page => Installation::UserSettingsPage
}, $class;
}
sub get_user_settings_page {
my ($self) = @_;
return $self->{user_settings_page};
sub create_user {
my ($self, $args_ref) = @_;
my $username = $args_ref->{username};
my $user_full_name = $args_ref->{user_full_name};
get_user_settings_page()->fill_in_username($username);
get_user_settings_page()->fill_in_user_full_name($user_full_name);
get_user_settings_page()->fill_in_password();
get_user_settings_page()->fill_in_password_confirmation();
get_user_settings_page()->press_next();
}
1;
{project_root}/lib/Installation/UserSettingsPage.pm
package Installation::UserSettingsPage;
use strict;
use warnings;
use testapi;
use parent 'Installation::WizardPage';
use constant {
# The needle to represent the page (e.g. Title in Installation Wizard). It is used to make sure that
# action is performed on the right Page.
USER_SETTINGS_PAGE => 'user-settings-page'
};
sub fill_in_username {
my ($self, $username) = @_;
assert_screen(USER_SETTINGS_PAGE); # ensure the correct Page is shown before performing an action
send_key('alt-u'); # make the field to be in focus
type_string($username); # type the username
}
sub fill_in_user_full_name {
my ($self, $user_full_name) = @_;
assert_screen(USER_SETTINGS_PAGE); # ensure the correct Page is shown before performing an action
send_key('alt-f'); # make the field to be in focus
type_string($user_full_name); # type the User's Full Name
}
sub fill_in_password {
assert_screen(USER_SETTINGS_PAGE); # ensure the correct Page is shown before performing an action
send_key('alt-p'); # make the field to be in focus
type_password(); # testapi method to enter the default secret password
}
sub fill_in_password_confirmation {
assert_screen(USER_SETTINGS_PAGE); # ensure the correct Page is shown before performing an action
send_key('alt-o'); # make the field to be in focus
type_password(); # testapi method to enter the default secret password
}
# overrides parent 'Installation::WizardPage' method.
sub press_next {
my ($self) = @_;
$self->SUPER::press_next(USER_SETTINGS_PAGE);
}
1;
-
Let's assume all the distributions have the same implementation of the User Settings. Then add the controller to Tumbleweed distribution, as all other distributions are inherited from it to follow 'factory first' rule.
{project_root}/lib/Distribution/Opensuse/Tumbleweed.pm
package Distribution::Opensuse::Tumbleweed; use strict; use warnings FATAL => 'all'; use parent 'susedistribution'; use Installation::UserSettingsController; sub get_user_settings { return Installation::UserSettingsController->new(); } 1;
-
If some of the Distributions has different implementation of User Settings for the same feature. For example, it still allows to create new user, but with different steps.
In this case, just override the
get_user_settings
method in the required Distribution.package Distribution::Opensuse::Leap::42; use strict; use warnings FATAL => 'all'; use parent 'Distribution::Sle::12'; sub get_user_settings { return Installation::SomeAnotherImplementationOfUserSettingsController->new(); } 1;
In order to run the Test Module, it should be added to a scheduling file. Please, do not add spaghetti code in main*.pm !
---
name: Incredible test suite
description: >
Do incredible things
schedule:
[...]
- installation/create_account
[...]