Example of using $container for $dispatcher injection

[HLeithner]

No description provided.

[Fedik]

This can work indeed.
However looks hacky, kind of. Not sure about this approach.

But, booting the extension in to your own container can be a nice feature.

[HLeithner]

which part would you say is hacky? parts which are b/c things or providing the container for booting the component?

[Fedik]

This part, which is b/c thing:

joomla-cms/libraries/src/Plugin/PluginHelper.php

Lines 234 to 240 in 4f2f071

	$container = Factory::getContainer();
	if ($dispatcher) {
	$container = $container->createChild()->set(DispatcherInterface::class, $dispatcher);
	}
	$plugin = Factory::getApplication()->bootPlugin($plugin->name, $plugin->type, $container);

But can work.

[HLeithner]

That's actually not a b/c thing ;-)
We don't expect a $dispatcher, but inject it if provided, we always specify the container for the boot process.

There is no other possibility to provide the dispatcher to plugin. Because setDispatcher be called (here) and $dispatcher in the CMSPlugin::constructor has been removed.

[Fedik]

I somehow ignored that you changed the boot method signature, sure it is b/c 😄
I was thinking about code that help fixing b/c 😄

[Fedik]

I have another idea.
If we could reorganize bootPotato() form Application to dedicated ExtensdionManager class, we could do:

// Regular
$app->getExtensionManager()
  ->bootPlugin($name, $type);

// With custom container
$app->getExtensionManager()
  ->setContainer($container->createChild()->set(DispatcherInterface::class, $dispatcher))
  ->bootPlugin($name, $type);

This can be fully backward compatible, and provide much more flexibility.

What do you think?

upd. hmhm, but in this way we will have cashing issue, also with your suggestion in this PR

[HLeithner]

yes I already found out, or better allon pointed me to the static cache in the boot method, which is not really useful in my opinion because it prevent you booting the same component in 2 different settings...

maybe it can be extended on the container hash... not sure

anyway your suggestion looks a way better then mine if it works.