Fix/infinite loop on restub invoke #30
Conversation
| if ( isset( $value ) ) { | ||
| return self::filter_return_for( $value ); | ||
| } | ||
| return \Patchwork\relay(); |
There was a problem hiding this comment.
I am not entirely sure why we need to relay here.
I did not intend to change this part (this is not directly part of the fix) but without this, the call count of a stubbed method gets increased. For example:
1) ConsecutiveExecutionTest::test__undefined_function___can_be_restubbed
Failed asserting that 2 matches expected 1.
There was a problem hiding this comment.
I'm having trouble finding the docs for this version of Patchwork to remember what this did exactly, but I assume the reason we relay here is so that a global function could be spied upon without needing to also mock it. Something like:
function getName() {
return 'human';
}
//...
$spy = \Spies\get_spy_for( 'getName' );
$name = getName();
$this->assertSame( 'human', $name );
$this->assertTrue( $spy->was_called() );If this is removed, won't that prevent the original global function from being called?
Although, I see that you have a test test__global_function__can_be_restubbed which covers this case, and it passes. How does that work?
There was a problem hiding this comment.
Ok, I followed the code and it looks like when GlobalSpies::handle_call_for() is called, it eventually calls GlobalSpies::call_original_global_function() thanks to Spy->get_return_for(), which uses \Patchwork\restore() to call the function before recreating the mock with GlobalSpies::replace_global_function().
That explains why this works. Maybe that wasn't always the case which would explain the relay? Or maybe there's another edge case we're not considering here. 🤷
c01ea2d to
cff9765
Compare
cff9765 to
6a8bc03
Compare
| if ( isset( $value ) ) { | ||
| return self::filter_return_for( $value ); | ||
| } | ||
| return \Patchwork\relay(); |
There was a problem hiding this comment.
I'm having trouble finding the docs for this version of Patchwork to remember what this did exactly, but I assume the reason we relay here is so that a global function could be spied upon without needing to also mock it. Something like:
function getName() {
return 'human';
}
//...
$spy = \Spies\get_spy_for( 'getName' );
$name = getName();
$this->assertSame( 'human', $name );
$this->assertTrue( $spy->was_called() );If this is removed, won't that prevent the original global function from being called?
Although, I see that you have a test test__global_function__can_be_restubbed which covers this case, and it passes. How does that work?
| if ( isset( $value ) ) { | ||
| return self::filter_return_for( $value ); | ||
| } | ||
| return \Patchwork\relay(); |
There was a problem hiding this comment.
Ok, I followed the code and it looks like when GlobalSpies::handle_call_for() is called, it eventually calls GlobalSpies::call_original_global_function() thanks to Spy->get_return_for(), which uses \Patchwork\restore() to call the function before recreating the mock with GlobalSpies::replace_global_function().
That explains why this works. Maybe that wasn't always the case which would explain the relay? Or maybe there's another edge case we're not considering here. 🤷
Fixes #31
(I know this repo has not been developed in long time but your feedback is appreciated. 😄 )
The first commit adds tests for #28, to make sure there is no regression.
The second and third commit respectively tests and fixes the problem described in #31.
Without the change in the third commit, the tests would get stuck in the infinite loop or cause
Allowed Memory Exhaustederror.Testing instructions
Please run tests with a memory limit (i.e.
vendor/bin/phpunit -d memory_limit=24M) so that tests are not stuck in a loop for a long time.Apply this patch (to remove @runTestsInSeparateProcesses and add finish_spying)
master, comment outself::$global_functions = [];insrc/Spies/GlobalSpies.php(one liner change from Forget the name of the global functions on restore #28) run the tests; some tests fail, so there could be come problems withfinish_spying(indicates the problem that PR tries to fix)master, revert the one liner change and run tests again: tests get memory limit error, so the infinite loop is possibly introduced (indicates the problem this PR is trying to fix)