Isn't there a sane-ish way to somehow present both interfaces?

It seems pretty arbitrary to pick the first one.

I couldn't come up with a better proposal. Can you? Listing all interfaces can be pretty long very quickly, especially when considering namespaces. In practice, the first is often the most useful, so it's not that arbitrary to me.

I do agree it will get long fast. Not sure I have a better suggestion at this point besides somehow mashing the interfaces together with a delimiter.

In practice, the first is often the most useful

I don't understand why this would be the case. Unless I am missing your point the order of interfaces carries no weight one way or another.

I don't understand why this would be the case. Unless I am missing your point the order of interfaces carries no weight one way or another.

I think it's not so much that the first is the most useful, but that in the majority of the cases there will be only a single parent or a single interface, so this heuristic works well in practice.

We could of course include both the parent and all interfaces, and for that matter the file and line where it is declared, but ... that would make the name very long and unusable for error messages. We have to strike a reasonable compromise between providing relevant information and being concise here, and I think the approach proposed here is that compromise.

Here's an example of a project where I added tests where anonymous classes within tests are used: https://github.com/spatie/data-transfer-object/pull/81/files#diff-1e8fc8e188715be5ac191bc06e73ade9R30

I believe this one will break after this change (the anonymous class "extends" an actual class).

However this specific test already uses expectExceptionMessageRegExp so I guess I just need to adapt it and it's fine 👍

but that in the majority of the cases there will be only a single parent or a single interface, so this heuristic works well in practice

Which is totally fair as long as it is clear to people the arbitrary choice is made so that it actually is more clear instead of people getting confused (which I suspect would be the case when they first encounter it).

Listing all interfaces can be pretty long very quickly, especially when considering namespaces

About that, could you please add a test with a namespaced base, just to make it clear whether e.g. \get_class(new class() extends \My\Ns\Foo {}) will be 'My\Ns\Foo@anonymous' . "\0" . <unique_stuff> or 'Foo@anonymous' . "\0" . <unique_stuff>?

@guilliamxavier I've added a test for that. The namespace is (of course) included.

We could of course include both the parent and all interfaces, and for that matter the file and line

I would say that file and line would be most helpful for debugging purposes.

What happens if multiple anonymous classes extending the same parent class are declared one after the other?

@Naitsirch FWIW, the file and line are included in the name: The full name (even after this change) is something like ParentClass@anonymous\0/some/long/path/file.php:42$123. Everything after the \0 is cut off in most contexts though, because the resulting name would be too unwieldy.