Codecov Report

All modified and coverable lines are covered by tests ✅

📢 Thoughts on this report? Let us know!

@schlessera wrote down the details now and would appreciate some extra eyes.

Some initial debugging suggests this is just an issue with the order packages are being autoloaded. the cli override example works because it happens to get loaded after the normal cli command, where as the eval override doesn't work because for whatever autoload magic reason the regular eval command is loaded again after the Custom_Eval_Command class is evaluated. I can reproduce this locally with tests, though not sure if the order is based on environment in anyway.

This suggests to me that it isn't possible to reliably actually override an existing command in the way we test, and it just happened to 'work' before because we used the same class names. If you add --debug to one of the failing examples you can see the eval command get added several times. The last one is from the original, not the override. Unfortunately I'm not more familiar with autoload to have a better idea.

I should use --debug more :-)

One of the early bootstrap steps is IncludePackageAutoloader, where commands from packages will be added. This includes the custom cli & eval commands.

Later on, IncludeFallbackAutoloader runs, which loads the fallback autoloader that is provided through the composer.json file (vendor/autoload.php). And I think here is the difference:

In tests, this will add the commands found in vendor/wp-cli, which includes the eval command. That's why the override happens. But when I perform the steps manually using the phar (loading phar://wp-cli.phar/vendor/autoload.php), there is no eval command being autoloaded, so it doesn't override the custom command from the package again.

I don't know whether that means it is working as intended.

However, conceptually, I would think that a reliable override would need to happen by actually waiting for the original command to be registered first, by hooking into after_add_command:eval. So something like this:

WP_CLI::add_hook(
  'after_add_command:eval',
  static function () {
    WP_CLI::add_command( 'eval', 'Custom_Eval_Command', array( 'when' => 'before_wp_load' ) );
  }
);

^ That seems to cause an infinite loop during tests though because the hook fires every time the command is added again. With a simple static variable it seems to work now...

It worked 🥳