I believe there’s another way to keep this working without implementing the interface:https://github.com/symfony/symfony-docs/pull/20932/files#diff-738612fa300cdcb8d044a240dcec7bc9db05a1671f30e3d1b987bbf200037b8c but I might be missing something, could you please confirm?

@yceruto it should work indeed, but the proposed approach makes sense to me also.

@yceruto I remember we discussed this at the time and you had some arguments againstSignalableCommandInterface, namely that the listener approach is better design-wise because less intrusive. I agreed with that and still do but I think either we decide that we don't need this interface at all and then it should be deprecated, or invokable commands should support it

Yes, I'd prefer a single way to do this for invokable command, less variants to document and maintain.

@yceruto Just a heads up,symfony/event-dispatcher is an optional dependency of the console component. If not installed, users won't be able to handle signals. I'm not saying that's good or bad, just something to be aware of.

I understand the intention of supporting all features used in normal commands, but in standalone usage, the invokable command approach doesn’t work naturally, people need to call explicitly setCode(), name, description, etc. In those cases, you’ll probably extend theCommand class directly to avoid the extra burden.

My argument was more about why not to deprecate or remove theSignalableCommandInterface and since it has its use case (when the component is used standalone), it feels counter-intuitive for it not to work with invokable commands.

In those standalone cases without anevent-dispatcher, you can still use signalable invokable commands by extendingCommand and implementingSignalableCommandInterface. The additional requirement is the manual extension, which is fine in my opinion since these are edge cases.

If the goal is to simplify support for invokable commands, I suggest keeping this PR as-is with theCommand implements SignalableCommandInterface approach part only. This way, if someone wants to use this feature with an invokable command, they only need to extend theCommand class, just as we currently recommend forinteract() and other advanced use cases.

@yceruto If you mean we should not supportSomeInvokableCommand implements SignalableCommandInterface then I disagree, reducing differences/limitations using invokable commands is worthwhile IMHO. The fact that extendingCommand is easier with console standalone shouldn't be taken as given I think, might improves later. Addinginteract() orinitialize() equivalents at some point is not excluded to me, if there's some traction.

I’m fine with invokable commands not supporting all CLI features out of the box. If extending the Command class is the way to enable them, that’s totally acceptable to me, just like we already do with invokable controllers and AbstractController. I don’t mind extending a class if that’s all it takes to unlock the functionality I need.

Sure, but why make it more complex than expected? I don't see your point sorry.

Shortcut methods and unsupported features are different stories. Stop me if I'm wrong but there is no AbstractController feature that cannot be implemented the same way by controllers not extending it

My goal is to make it simpler for developers, but I’m probably wrong in this case, sorry for the noise and confusion@nicolas-grekas.

No worries, it's always useful to have a discussion. Here I don't see much added complexity (I see this as simpler actually, on the Command side) so I'm fine with the change.

Thank you@HypeMC.