@title Differential User Guide: Test Plans
@group userguide

This document describes things you should think about when developing a test
plan.

= Overview =

When you send a revision for review in Differential you must include a test
plan. A test plan is a repeatable list of steps which document what you have
done to verify the behavior of a change. A good test plan convinces a reviewer
that you have been thorough in making sure your change works as intended and
has enough detail to allow someone unfamiliar with your change to verify its
behavior.

This document has some common things to think about when developing or reviewing
a test plan. Some of these suggestions might not be applicable to the software
you are writing; they are adapted from Facebook's internal documentation.

= All Changes =

  - **Error Handling:** Are errors detected and handled properly? How does your
    change deal with error cases? Did you test them and make sure you got the
    right error messages and the right behavior? It's important that you test
    what happens when things go wrong, not just that the change works if
    everything else also works.
  - **Service Impact:** How does your change affect services like memcache,
    thrift, or databases? Are you adding new cachekeys or queries? Will
    this change add a lot of load to services?
  - **Performance:** How does your change affect performance? **NOTE**: If
    your change is a performance-motivated change, you should include profiles
    in your test plan proving that you have improved performance.
  - **Unit Tests:** Is your change adequately covered by unit tests? Could you
    improve test coverage? If you're fixing a bug, did you add a test to prevent
    it from happening again? Are the unit tests testing just the code in
    question, or would a failure of a database or network service cause your
    test to fail?
  - **Concurrent Change Robustness:** If you're making a refactoring change, is
    it robust against people introducing new calls between the time you started
    the change and when you commit it? For example, if you change the parameter
    order of some function from ##f(a, b)## to ##f(b, a)## and a new callsite is
    introduced in the meantime, could it go unnoticed? How bad would that be?
    (Because of this risk, you should almost never make parameter order
    changes.)
  - **Revert Plan:** If your change needs to be reverted and you aren't around,
    are any special steps or considerations that the reverter needs to know
    about? If there are, make sure they're adequately described in the "Revert
    Plan" field so someone without any knowledge of your patch (but with a
    general knowledge of the system) can successfully revert your change.
  - **Security:** Is your change robust against XSS, CSRF, and injection
    attacks? Are you verifying the user has the right capabilities or
    permissions? Are you consistently treating user data as untrustworthy? Are
    you escaping data properly, and using dangerous functions only
    when they are strictly necessary?
  - **Architecture:** Is this the right change? Could there be a better way to
    solve the problem? Have you talked to (or added as reviewers) domain experts
    if you aren't one yourself? What are the limitations of this solution? What
    tradeoffs did you make, and why?

= Frontend / User-Facing Changes =

  - **Static Resources:** Will your change cause the application to serve more
    JS or CSS? Can you use less JS/CSS, or reuse more?
  - **Browsers:** Have you tested your change in multiple browsers?