Only use DidYouMean-integrated Error for Component loading failure

[cllns]

Without this, any NameError (including NoMethodError) will go through the DidYouMean integration and
it will fail since that code expects only a Constant, not any name (including "undefined local variable or method")

Instead, this code will ensure we only try to do that when we get a NameError for specifically trying to load the constant we're looking for. Everything else will be re-raised and go through the normal flow (including any more DidYouMean default behavior).

Fixes hanami/hanami#1252. (Specifically see this comment)

Bug introduced in #217

Thoughts on whether we should test this? We could create a fixture file to load a component that has a NameError in it and ensure that error passes through instead of getting created as a ComponentNotLoadableError... but that would error out anyway. Creating a NameError in the spec just raises that before we can even get to the example.

I manually confirmed that it works with the repo @katafrakt created to illustrate the bug.

[timriley]

@cllns This sounds and looks good!

Do you reckon there's any chance you could add a test demonstrating one of these unrelated errors not getting caught up in our DidYouMean handling?

[timriley]

Oh, haha, I just read your description more closely.

We could create a fixture file to load a component that has a NameError in it and ensure that error passes through instead of getting created as a ComponentNotLoadableError

This sounds about right to me. I'd want this to be an integration-style test more than anything lower-level, since it's the end-to-end user experience we care about particularly here. Couldn't the test wrap the loading of that file in a before/rescure, or in an expect { }.to raise_error construct?

[cllns]

@timriley Can you take a look at my newest commit? I haven't been able to reproduce the issue from hanami/hanami#1252. Note that I've added || true to the main conditional here so it's effectively the existing library code. I want to reproduce the existing behavior so I can be sure this fix works as desired.

I've included comments in the spec/integration/did_you_mean_integration_spec.rb file. The first one, about finalize, isn't exactly necessary, just something that surprised me based on my (now rusty) memory of dry-system's terminology. The second one, after system["undefined_variable"] explains where the issue is. I guess the question is: what is Hanami doing that I'm not doing here... A lot! I guess, but what's relevant to the Loader not seemingly being used here?

[cllns]

Thanks for finishing the spec @solnic!