Skip to content

Latest commit

 

History

History
114 lines (74 loc) · 3.78 KB

_converting_from_old_tests.md

File metadata and controls

114 lines (74 loc) · 3.78 KB

Converting from the old Selenium tests

Changes to the framework

Original tests Converted tests
Language Selenese Ruby
Framework Selenium Selenium 2 / WebDriver
Driver Selenium IDE, under Firefox RSpec

Problems with the original tests

Non-portable

  • When files were to be uploaded, the entire path was hardcoded. So the test file might contain the literal path ``

Not reliable

  • Some test steps rely on Javascripts to create the page. For example, those that use TinyMCE.
    • With no way to tell when the Javascript was complete, pause statements were inserted liberally.
  • Some test steps rely on the VIVO search indexer to produce the desired results.
    • With no way to tell when the indexing was complete, pause statements were inserted liberally.

Required manual setup

  • In some test suites, need to:
    • Run the initial tests.
    • Stop Tomcat, add rdf files, restart Tomcat.
    • Run the remaining tests.

Conversion process

  • Write this Ruby app to accomplish most of the conversion.
  • Curate the tests by hand to complete the conversion.

Quirks in the original tests

Omnipresent calls to expectTitle

  • The tests were recorded in the Selenium IDE and only minimally modified afterwards.
  • The Selenium IDE recorded an expectTitle assertion each time a page loaded.
  • Frequently, these assertions are repetitive, or irrelevant to the test.

Ambiguous element selectors

  • The Selenium IDE allowed one to specify an element as "LastName" for example.
    • Selenium would search for the element as if the specifier were "id=LastName" or "name=LastName", accepting either result.
    • I don't know which one would override the other.
  • In the new tests, the selection method must be explicitly stated as either :id or :name (or several other possiblilities).

Instructions for running the original tests

The staff at Cornell maintained a procedural checklist for running the original tests. Notations might include:

  • Instructions for setting up the test suite.
  • Manual operations to be performed between tests in the suite.
  • Results to be observed to determine whether the test succeeded.

Here is the most recent checklist available, the checklist for Release 1.8, release candidate 4.

Things that should be done to clean up the converted tests.

Improve element locators.

  • An element locator like this:

     :css, "li.nonSelectedGroupTab.clickable"

    does not show the user's intent, and also relies on the order of the tabs on the page. Change it to this:

    :xpath, ".//li[@groupname='publications']"

    or even this:

    :xpath, ".//ul[@class='propertyTabsList']/li[@groupname='publications']"

    to make it clearer and more reliable.

  • Here's another nice locator:

    ```

    :xpath, ".//a[text()='Atlantis']/img" ```

Remove unnecessary calls to vivo_wait_for_indexing

Many of these calls were added by the automated conversion process, and are not necessary. A call to vivo_wait_for_indexing is only required when the next operation relies on the contents of the search index.

Consider whether "tear-down" steps are useful

The old tests were written to clean up after themselves. Each item that was added for a test was also removed during that test. This attempted to insure that each subsequent test worked against empty content.

With the new test framework, this isn't necessary, since the triple-store is wiped after each test.

So the question resolves to this: does the "tear-down" serve to test functionality? If so, then keep it in. But if not, skip it.