I think what he's pushing is slightly different from dependency injection.
For one, one of the hallmarks of DI is that you should have a single, central location where all your dependencies are assembled. Whereas one of the slides in the talk is, "Avoid central configuration at all cost." His concrete example is with CSS files, but I could definitely a similar principal applying to your "composition.json" file in a large DI application.
It seems more to me like he's talking about the opposite of DI. It's a possible solution to how you decouple modules from their dependencies. But in DI, you accomplish it by having a central orchestrator inject the dependencies into your modules. In this "enhance" mechanism, by contrast, you reject that central orchestration, and instead have dependencies independently injecting themselves into the modules.
> one of the hallmarks of DI is that you should have a single, central location where all your dependencies are assembled.
This is very interesting to me mainly because it's the first time I've heard it. In Martin Fowler's bliki article [0] he discusses having an assembler, but for me that's just an abstract factory pattern (perhaps I'm wrong). If you don't have abstract dependencies, then you don't need it. Do you have some sources which discuss this? I don't really see the advantage of using a kind of repository for dependencies.
As for how this fits with abstract factories, in DI they exist to handle a specific problem: sometimes (but not always), actual instance construction has to be delayed until the last minute. But you still want things to be loosely coupled, and do the wire-up at the same place where you handle all the other object composition. So you do that by declaring an abstract factory, and then generating its concrete implementation at wire-up time. That way the details of the composition are still being injected, even if they're being executed at the site where they'll be consumed.
Enhance kind of looks like google hype. Essentially it’s a way of dynamic dependency discovery. Rather than explicitly adding imports, you just crawl through a directory convention and discover things to bundle and connect.
I liked the talk, dependency management and code splitting are big problems in large apps, but there was a lot of hand wavy “we at google invented this, so it must be great” reasoning.
>For one, one of the hallmarks of DI is that you should have a single, central location where all your dependencies are assembled
While this is true, if you're doing your composition root right in a large application, then you're assembling most of your dependency graph using conventions, and assembling only the most unique things by hand.
While it may be a hallmark of DI it’s not necessary. Spring is perhaps the most famous DI framework of them all and is frequently used by “classpath scanning” relevant components to wire in
For one, one of the hallmarks of DI is that you should have a single, central location where all your dependencies are assembled. Whereas one of the slides in the talk is, "Avoid central configuration at all cost." His concrete example is with CSS files, but I could definitely a similar principal applying to your "composition.json" file in a large DI application.
It seems more to me like he's talking about the opposite of DI. It's a possible solution to how you decouple modules from their dependencies. But in DI, you accomplish it by having a central orchestrator inject the dependencies into your modules. In this "enhance" mechanism, by contrast, you reject that central orchestration, and instead have dependencies independently injecting themselves into the modules.