June felt like a little bit of a down month for me in OSS. For various reasons, I struggled to find decent blocks of time to make headway into all the loops I opened back in May. But in the end, I found a way to nudge some other important things forward!
Rather than push against the grain at a low ebb, I used what time I had in June to do some more nourishing ”fun” work, to get back into dry-system again and prepare the way for Hanami to have a flatter Ruby source file directory structure.
What we want is instead of
Main::HelloWorld being defined in this file:
We want too remove that second redundant
main/ and have the file like so:
The idea here is to make the deeper hierarchies imposed by each namespaced Hanami slice a little less ”in your face,” and in doing so, make this whole framework just that much more approachable to a wider range of users.
loader = Zeitwerk::Loader.new loader.push_dir "slices/main/lib", namespace: Main
However, for Hanami to use both Zeitwerk and dry-system together, more work is required. This is also made more challenging by the fact that dry-system needs to map names in two directions, from identifier to source file/constant (when lazy loading a single registration) and from source file to constant/identifier (when finalizing a container and crawling its managed directories for all source files), whereas Zeitwerk only needs to go in one direction, from constant to source file.
Until now, the only support dry-system has had for namespacing was its
default_namespace setting on its component dirs. This worked very simplistically. Given a setting like the following:
config.component_dirs.add "lib" do |dir| dir.default_namespace = "main" end
If dry-system encountered a file like
main/hello_world.rb inside that component dir, then it would strip off the leading
"main" when forming its identifier, leaving it as
"hello_world". And when the container is lazy loading, it would first try prepending the
default_namespace to find a file for a given identifier, before falling back to searching without the namespace.
This has worked fine for most use cases, but it won’t cut it for our Hanami plans.
So to address this, I’m adding namespaces as a first-class concept to dry-system component dirs. A replacement for the
default_namespace config above would look like this:
config.component_dirs.add "lib" do |dir| dir.namespaces.add "main" end
To add a namespace, we specify the path within the component dir, along with an
const to use within that path.
The example above is actually shorthand for the following:
config.component_dirs.add "lib" do |dir| dir.namespaces.add "main", identifier: nil, const: "main" end
Playing this out, this means: ”within the
main/ subdirectory of this component dir, don’t add anything to the beginning of the component’s identifiers, but expect them all to be within the
Main object namespace.”
Following on from this, we should be able to configure some wildly divergent component loading rules. For example:
dir.namespaces.add "main", identifier: "bits_and_bobs", const: "stuff_and_things"
With this config, we’d expect files in
main/ to define classes inside the
StuffAndThings namespace, and have their identifiers all prefixed by
As you might’ve noticed, the
namespaces.add method means we can configure multiple namespaces:
dir.namespaces.add "main" dir.namespaces.root # special method for component dir root dir.namespaces.add "admin"
Namespaces will be referenced in order when resolving components or loading files into the container, too, which improves what was otherwise a poorly defined part of dry-system’s behaviour. So with the configuration above, if you had a
main/component.rb and an
admin/component.rb, the file inside
main/ would be favoured in all cases and would always appear in the container as the
"component" registration, because its namespace was added first.
So, after all of this, we now have the necessary tools to support our elided directory structure from the top, which would look something like this:
config.component_dirs.add "lib" do |dir| dir.namespaces.root const: "main" end
I have a messy work in progress PR underway to support all of this, so feel free to check it out if you’d like to see the code equivalent of the under construction gif. This will end up being yet another fairly big overhaul of dry-system internals, so I think it’ll probably take me the better part of July to get it finished. Iterating over the namespaces as part of component loading will also have some performance implications, so I’ll want to do some benchmarking of that process too.
That said, I’m really excited to find another opportunity to make dry-system more powerful and better structured than before! And of course, to then employ that to create a nicer experience for future Hanami users :)
A huge thank you to Bohdan V. for joining my GitHub sponsors during June!
See you all next month, folks!