Now that imports can be limited to smaller scopes than the entire module,
unused import warnings need to work like unused def warnings.
This commit moves unused import warnings discovery and reporting from load
to canonicalization where we can track their usage per scope.
This also fixes a longstanding bug where unused exposed names from an import
were not reported if they were only used in a qualified manner.
Previously, all imports were available in the header, so we could start
processing dependencies as soon as we parsed it. However, the new imports
are treated as defs, so we have to parse the whole module to find them.
This commit essentially moves the dependency resolution from the `LoadHeader`
phase to the `Parse` phase, and it updates canonicalization to introduce
module symbols into scope when a `ValueDef::ModuleImport` is encountered.
NOTE:
- The `imports` header still parses, but it's no longer wired up. I will remove
it in an upcoming commit.
- Ingested files and imports that appear in nested expressions are not
yet supported by load
I think for now it is not worth considering adding hashflooding mitigation to the Roc Dict.
1. Wyhash is a secure enough has to pass SMHasher.
2. Languages like Go use wyhash in production and have not seen hashflooding issues.
3. We do have a random seed that Dicts use that changes per application linking. (Could monomorphize on Dict type for more randomness)
4. Any sort of fallback checks will lead to worse performance in the normal case.
5. A reasonable http server or app can limit the size of JSON or number of HTTP headers to accept.
ankerl::dense_unordered is a very fast hash map that is built to be an index map.
This enables extra optimizations compared to just wrapping a regular hash map.
As such, I think this map is very well suited for our index map impl in Roc.
I also think this dictionary implementation is simpler overall.
On top of that, this removes the need for SIMD instructions for peak performance.
Benchmarks of the C++ version and other C++ hash maps are here: https://martin.ankerl.com/2022/08/27/hashmap-bench-01/
Though this has clear bias of being written by the author of ankerl::dense_unordered,
the results all look correct and the benchmarks thorough.
