Bugfix: Modules with same original name wrongly being marked as unused

[seafoamteal]

Description

This resolves #5145.

Three changes were made:

1. When a module was aliased, the entity for the original name has to be placed in the Shadowed layer and not the Module layer. This prevents it from shadowing previous imports with the same module name.
2. When registering a module reference from a module access site, the module_name_to_node map should not be checked. This map contains full original names of modules (e.g. wibble/wobble and wobble) and hence isn't guaranteed to point us to the correct node when we have limited information (e.g. just wobble). We skip this check and directly look up the entity in the Module layer. The module_name_to_node map is only needed/used when registering references from unqualified imports and aliases, since we are guaranteed to have information only about the original name of the module in those cases.
3. This is not directly related to the bugfix, but the assert_no_warnings! macro was defined ambiguously. If src was any expr, then the compiler would not know whether a second module tuple was part of the repeated list of module tuples or src. Changing src to be a literal fixes this.

Example

Before

import wibble/wobble
import wobble as wob

pub fn main() {
  wobble.wimble()
  wob.womble()
}

Here, at wobble.wimble(), a lookup in the module_name_to_node map would return the top-level wobble module that is aliased to wob. Hence, no reference to wibble/wobble is registered and it is incorrectly marked unused.

A less visible problem is that wibble/wobble is added to the Module layer as entity with name wobble. Then, on the next line, wobble is once again added to the Module layer with the same name, shadowing the first import.

After

At wobble.wimble(), we look up the wobble entity in the Module layer instead. This gives us the correct wibble/wobble module.

The Module layer contains two entities: wobble and wob. The Shadowed layer contains one entity: wobble.

[lpil]

Thank you! I've left some questions inline.

[lpil]

I don't understand this new name. It seems to say something about the context in which it is used, but not name the function itself, and the meaning is unclear.

Could these two functions be given clearer names please, and documentation comments could be useful 🙏

[seafoamteal]

You're right, sorry about that. I've given them both better names now, relating to the sort of module name they take as arguments.

[lpil]

What does this new name mean? What distinguishes it from the new function with the previous name?

[seafoamteal]

I believe I was trying to work around making EntityLayer public and changing all the call sites of register_module, but it does seem cleaner to avoid the extra wrapper function. I've made the changes necessary to remove it.

[seafoamteal]

Hey @lpil, I'm just bumping this because I'm not sure if the inline responses show up as notifications.