Published: Monday, May 11, 2009 Last modified: Sunday, May 19, 2024

Intro: As part of the W3C Mobile Web Test Suites Working Group I’m co-ordinating in tests of the Widget runtime, in which W3C compliant Widget run times implement Widgets 1.0: Packaging and Configuration.

WGTQA

To begin with individual tests were written by hand in a WGTQA git repository. My employer Aplix supported the work (~1 working day a week) as it would be useful to partake in compliance efforts for both BONDI Widget runtime efforts and our own commercial widget runtime offerings.

I made a couple of mistakes. They were mostly interactive tests and didn’t make use of Javascript to automate them in several cases. For example:

5.7-startfile-pass1/
index.html:
	<h1 style="background: green;">PASSED</h1>

Firstly I tried to pin the name of the test on the section of the specification it was testing. Unfortunately the numbering changed, so 5.7 had worse than no meaning. It became confusing and a huge pain to track to keep updated.

index.html just has a simple PASSED rendering. Great for interactive tests, though useless for automation. The “start file” should have been a simple Javascript call to indicate the test case passed.

Testing Error conditions

The next mistake was assuming a malformed config would halt processing. In reality a Widget runtime will ignore bogus values. I wrote several negative tests for config.xml, which wrongly assumed that if the index.html was seen, it would be a FAIL. Assuming the UA would halt and show the widget has been rejected instead, as a PASS.

A better test would be check the default values were set when a bad one was given. Unfortunately Widget APIs are required to read back and test such values. This API is not available on any current widget runtime like BONDI or Opera AFAIK.

So you can’t really do negative tests, instead you need to check default values instead.

WTF !

The “Widget Testing Framework” (WTF) is a PHP Zend powered contribution by Opera, which basically deals with the earlier automation problem. WTF comes without any substantial tests, though Opera has stated they will share some of their existing Widget tests.

Robert Græsdal from Opera is working on WTF also just one day a week.

For example, lets look at a test which tests the order in which index files are run. So the start files are modified from the index template from data/std_includes/index.html which references testpass.js:testcase.result(true); on the first case and testfail.js:testcase.result(false); on all the others:

  1. index.htm – PASS
  2. index.html – FAIL
  3. index.svg – FAIL
  4. index.xhtml – FAIL
  5. index.xht – FAIL

When the widget is downloaded a testlib.js & style.css is also included in the widget. testpass.js executes the function implemented by testlib.js which reports the test result with the magic of AJAX to a configurable reporturi. Unfortunately the backend service for this isn’t provided yet by Opera, though it shouldn’t be hard to recreate.

So the framework builds the widget on the fly. Sooner or later, you’ll realise this is good testing framework for mainly checking to see if the Javascript APIs e.g. BONDI & W3C for different features works correctly within a widget.

Though WTF is not good for testing things like broken zips and odd filenames and screwy encodings and such things, before a widget is actually running. So WTF is not a “silver bullet”, a solution for all problems related to ensuring W3C Widget runtime conformance.

Non-interactive interactive tests

In terms of processing a test, there are 3 types:

  1. Automated
  2. Visual
  3. Interactive

Automated tests are best. You could run all you automated tests via a test harness (which you need to write) after every change to your Widget runtime for example. Hence this would become also known as a regression test suite.

Visual tests are generally written as interactive tests at first. The tester is basically looking out for some green blob and then to mark it off with a button control.

You do not want expensive, labour intensive and mind numbing interactive tests. :) Imagine for every test run after a source check in, having to mark whether you saw a green blob or not?

Opera has an übercool internal system called SPARTAN which hashes these values (those green blobs), so that these interactive (bad) tests can become like automated (good) visual tests. Since SPARTAN isn’t public, ‘visual tests’ will remain ‘interactive tests’ for the time being.

Therefore it’s a good idea to concentrate on writing automated tests for the time being.

How a widget is built up by WTF

The widget.txt is not just a description of the test. It also allows some easy feature switches in an INI format.

For example setting fileio=1 should enable opera.io by building an appropriate config.xml to enable that feature.

The idea behind that, is that it’s easier for tests to be maintained that way, than relying on the feature element.

Most of a widget is built up. For example using standard includes:

hendry@x61 tmp$ unzip true.wgt
Archive:  true.wgt
  inflating: index.html
  inflating: style.css
  inflating: config.xml
  inflating: testlib.js
  inflating: widget.txt
  inflating: test.js

So for a basic true test: index.html, style.css, testlib.js was included and config.xml was generated.

Open issues

config.xml validation

We know what makes an invalid config.xml with Widget RelaxNG schema. I’ve written a widget validator which uses rnv.

This is what a UA should do before running a widget, though how do you test a UA (widget runtime) is doing the same as:

rnv -q widget.rnc config.xml

Please contact me or mail the public-mwts if you have some ideas.

Pinning tests

Anchor names seem to be more stable than section names, but still it seems a little hard to reference the part of the specification, the test is actually testing.

Alternatively, HTML5 uses an annotation system for linking to tests from the specification.

Making the WTF properly opensourced with open SCM hosting etc. etc.

The Widget Test Framework is Opera specific atm and needs several changes.

My debug copy is publicly available in Mercurial.

Test harness

Something to run the tests and collect the tests results is needed.

Visual tests

A system like SPARTAN is needed to make visual tests, less interactive.

Notes on different Widget runtimes

Opera

Does not allow alert(). Workaround:

var alert = function (msg) { widget.showNotification(msg); }

Serving from *.opera.com will avoid security prompt. i.e. add 127.0.0.1 foo.opera.com to /etc/hosts