Making testing of UIs fantastic
Project description
Widgetastic - Making testing of UIs fantastic.
Written originally by Milan Falesnik (mfalesni@redhat.com, http://www.falesnik.net/) and other contributors since 2016.
Licensed under Apache license, Version 2.0
WARNING: Until this library reaches v1.0, the interfaces may change!
Introduction
Widgetastic is a Python library designed to abstract out web UI widgets into a nice object-oriented layer. This library includes the core classes and some basic widgets that are universal enough to exist in this core repository.
Features
Individual interactive and non-interactive elements on the web pages are represented as widgets; that is, classes with defined behaviour. A good candidate for a widget might be something a like custom HTML button.
Widgets are grouped on Views. A View descends from the Widget class but it is specifically designed to hold other widgets.
All Widgets (including Views because they descend from them) have a read/fill interface useful for filling in forms etc. This interface works recursively.
Views can be nested.
Widgets defined on Views are read/filled in exact order that they were defined. The only exception to this default behaviour is for nested Views as there is limitation in the language. However, this can be worked around by using View.nested decorator on the nested View.
Includes a wrapper around selenium functionality that tries to make the experience as hassle-free as possible including customizable hooks and built-in “JavaScript wait” code.
Views can define their root locators and those are automatically honoured in the element lookup in the child Widgets.
Supports Parametrized views.
Supports Version picking.
Supports automatic Constructor object collapsing for objects passed into the widget constructors.
Supports Fillable objects that can coerce themselves into an appropriate filling value.
Projects using widgetastic
ManageIQ integration_tests: https://github.com/ManageIQ/integration_tests
Installation
pip install -U widgetastic.core
Contributing
Fork
Clone
Create a branch in your repository for your feature or fix
Write the code, make sure you add unit tests.
Run tox to ensure your change does not break other things
Push
Create a pull request
Basic usage
from selenium import webdriver
from widgetastic.browser import Browser
from widgetastic.widget import View, Text, TextInput
# Subclass the default browser, add product_version property, plug in the hooks ...
class CustomBrowser(Browser):
pass
# Create a view that represents a page
class MyView(View):
a_text = Text('.//h3[@id="title"]')
an_input = TextInput(name='my_input')
# Or a portion of it
@View.nested # not necessary but you need it if you need to keep things ordered
class my_subview(View):
# You can specify a root locator, then this view responds to is_displayed and can be
# used as a parent for widget lookup
ROOT = 'div#somediv'
another_text = Text('#h2') # Whatever takes a locator can automatically detect simple CSS locators
selenium = webdriver.Firefox() # For example
browser = CustomBrowser(selenium)
# Now we have the widgetastic browser ready for work
# Let's instantiate a view.
a_view = MyView(browser)
# ^^ you would typically come up with some way of integrating this in your framework.
# The defined widgets now work as you would expect
a_view.read() # returns a recursive dictionary of values that all widgets provide via read()
a_view.a_text.text # Accesses the text
# but the .text is widget-specific, so you might like to use just .read()
a_view.fill({'an_input': 'foo'}) # Fills an_input with foo and returns boolean whether anything changed
# Basically equivalent to:
a_view.an_input.fill('foo') # Since views just dispatch fill to the widgets based on the order
a_view.an_input.is_displayed
Version picking
By version picking you can tackle the challenge of widgets changing between versions.
In order to use this feature, you have to provide product_version property in the Browser which should return the current version (ideally utils.Version, otherwise you would need to redefine the VERSION_CLASS on utils.VersionPick to point at you version handling class of choice) of the product tested.
Then you can version pick widgets on a view for example:
from widgetastic.utils import Version, VersionPick
from widgetastic.widget import View, TextInput
class MyVerpickedView(View):
hostname = VersionPick({
# Version.lowest will match anything lower than 2.0.0 here.
Version.lowest(): TextInput(name='hostname'),
'2.0.0': TextInput(name='host_name'),
})
When you instantiate the MyVerpickedView and then subsequently access hostname it will automatically pick the right widget under the hood.
VersionPick is not limited to resolving widgets and can be used for anything.
You can also pass the VersionPick instance as a constructor parameter into widget instantiation on the view class. Because it utilizes Constructor object collapsing, it will resolve itself automatically.
Parametrized views
If there is a repeated pattern on a page that differs only by eg. a title or an id, widgetastic has a solution for that. You can use a ParametrizedView that takes an arbitrary number of parameters and then you can use the parameters eg. in locators.
from widgetastic.utils import ParametrizedLocator, ParametrizedString
from widgetastic.widget import ParametrizedView, TextInput
class MyParametrizedView(ParametrizedView):
# Defining one parameter
PARAMETERS = ('thing_id', )
# ParametrizedLocator coerces to a string upon access
# It follows similar formatting syntax as .format
# You can use the xpath quote filter as shown
ROOT = ParametrizedLocator('.//thing[@id={thing_id|quote}]')
# Widget definition *args and values of **kwargs (only the first level) are processed as well
widget = TextInput(name=ParametrizedString('#asdf_{thing_id}'))
# Then for invoking this:
view = MyParametrizedView(browser, additional_context={'thing_id': 'foo'})
It is also possible to nest the parametrized view inside another view, parametrized or otherwise. In this case the invocation of a nested view looks like a method call, instead of looking like a property. The invocation supports passing the arguments both ways, positional and keyword based.
from widgetastic.utils import ParametrizedLocator, ParametrizedString
from widgetastic.widget import ParametrizedView, TextInput, View
class MyView(View):
class this_is_parametrized(ParametrizedView):
# Defining one parameter
PARAMETERS = ('thing_id', )
# ParametrizedLocator coerces to a string upon access
# It follows similar formatting syntax as .format
# You can use the xpath quote filter as shown
ROOT = ParametrizedLocator('.//thing[@id={thing_id|quote}]')
# Widget definition *args and values of **kwargs (only the first level) are processed as well
the_widget = TextInput(name=ParametrizedString('#asdf_{thing_id}'))
# We create the root view
view = MyView(browser)
# Now if it was an ordinary nested view, view.this_is_parametrized.the_widget would give us the
# nested view instance directly and then the the_widget widget. But this is a parametrized view
# and it will give us an intermediate object whose task is to collect the parameters upon
# calling and then pass them through into the real view object.
# This example will be invoking the parametrized view with the exactly same param like the
# previous example:
view.this_is_parametrized('foo')
# So, when we have that view, you can use it as you are used to
view.this_is_parametrized('foo').the_widget.do_something()
# Or with keyword params
view.this_is_parametrized(thing_id='foo').the_widget.do_something()
The parametrized views also support list-like access using square braces. For that to work, you need the all classmethod defined on the view so Widgetastic would be aware of all the items. You can access the parametrized views by member index [i] and slice [i:j].
It is also possible to iterate through all the occurences of the parametrized view. Let’s assume the previous code sample is still loaded and the this_is_parametrized class has the all() defined. In that case, the code would like like this:
for p_view in view.this_is_parametrized:
print(p_view.the_widget.read())
This sample code would go through all the occurences of the parametrization. Remember that the all classmethod IS REQUIRED in this case.
You can also pass the ParametrizedString instance as a constructor parameter into widget instantiation on the view class. Because it utilizes Constructor object collapsing, it will resolve itself automatically.
Constructor object collapsing
By using widgetastic.utils.ConstructorResolvable you can create an object that can lazily resolve itself into a different object upon widget instantiation. This is used eg. for the Version picking where VersionPick descends from this class or for the parametrized strings. Just subclass this class and implement .resolve(self, parent_object) where parent_object is the to-be parent of the widget.
Fillable objects
I bet that if you have ever used modelling approach to the entities represented in the product, you have come across filling values in the UI and if you wanted to select the item representing given object in the UI, you had to pick a correct attribute and know it. So you had to do something like this (simplified example)
some_form.item.fill(o.description)
If you let the class of o implement widgetastic.utils.Fillable, you can implement the method .as_fill_value which should return such value that is used in the UI. In that case, the simplification is as follows.
some_form.item.fill(o)
You no longer have to care, the object itself know how it will be displayed in the UI. Unfortunately this does not work the other way (automatic instantiation of objects based on values read) as that would involve knowledge of metadata etc. That is a possible future feature.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.