Fix static hooks

[jordanbrauer]

Q	A
Bug fix?	yes
New feature?	no
Fixed tickets	#351

Issue

The issue is essentially, exactly, as described by @fnzr in the related ticket (great job and good find!).

Basically, the problem happens because each test, when it's being parsed, registers the global hook on a shared variable (Testable::$beforeAll) as if it were a new hook. The solution would be...well, not do that. [...]

More specifically it is a result of the static nature of these hooks, which is why we do not see this behaviour on the beforeEach and afterEach variants. This would have been a huge oversight on my part when initially implementing this feature in #282 , and could pose a potential performance issue for some folks, should they be doing any heavy work or network activity within.

Solution

Funnily enough, it's a one-liner ... 😑 ... 🤣 . Nearly all changes are test related.

Source

• ensure that we re-initialize static ?Closure $beforeAll and static ?Closure $afterAll at constructor-time for each test case, within the Testable concern trait;

Tests

Due to how these hooks work (e.g., can be defined in multiple files, variable amount of tests using them, etc.), it's a little tricky to ensure they are only run once per file. The solution to this is outlined below.

• add a calls property to the global $globalHook object;
  • within here we keep a running total of hook calls, starting at 0;
• within each hook (we are testing), assert & increment it's call count;
  • call count is always incremented by 1, except in the tests themselves;
  • when asserting the call count, depending on the test we need to calculate and apply a delta/offset to our expected value;
    • see BeforeAllTest.php and AfterAllTest.php header comments for full details

[jordanbrauer]

CC @nunomaduro @owenvoke @lukeraymonddowning

Requesting a review for this bug fix when you find some time 🙂

[lukeraymonddowning]

I know this is ripe coming from the guy who promotes higher order expectations inside higher order tests but should we split this over two lines for clarity?

[jordanbrauer]

Probably, yeah. It is kind of easy to miss, if not careful. Two lines it is!

[jordanbrauer]

Updated!

[lukeraymonddowning]

@jordanbrauer just for my own understanding, if we add another test to the AfterEachHook test down the line, we'll need to update the BeforeEachHook test offset, yes?

[jordanbrauer]

if we add another test to the AfterEachHook test down the line, we'll need to update the BeforeEachHook test offset, yes?

@lukeraymonddowning , not quite, and thank goodness haha. The offset would only need to be updated if another test file was added to the tests/Hooks/ directory.

For a better visual of why offset is 2,

tests/Hooks
├── AfterAllTest.php           <-- no tests run before this one, so no offset
├── AfterEachTest.php          <-- 2 is our offset for the hook calls count
├── BeforeAllTest.php          <-- two tests run before this one
└── BeforeEachTest.php

This kind of sucks though still, because if we ever have a feature to randomize the test order (or if we currently do), it would fail for false positive. However, that being said, shouldn't be a problem since there are snapshot tests that indirectly care about the order.

Here I added some dummy tests that just expect true, but no changes to offset.