Expand description
Python’s semantic diagnostics: proof-gated unresolved-name reporting.
Every candidate this pass reaches leaves one typed outcome in the
SemanticDiagnosticReport: a resolution, a complete absence proof, or a
typed reason why absence could not be proven. A candidate is never dropped
in silence, and an error is only ever published behind a
SemanticAbsenceProof over a surface that was complete.
The analyzer facts this needs are the PythonSource the import resolver
already takes, a BoundedDefinitionLookup for “is this fqn indexed”, and
a PythonEnvironmentSurface for “what do the activated environment packs
prove about this imported module”. analyzer/python/diagnostics.rs in
brokk-bifrost-analysis keeps the downcast that produces all three: this
crate cannot name the semantic-model overlay those answers come from.
What this pass judges:
- every bare-name reference, against the file’s lexical surface and the workspace index, which are both complete and workspace-local;
- every import declaration, against the retained environment surface;
- an attribute read through a module binder (
os.path.join), which is the one receiver whose owner Python’s syntax proves without type inference.
An attribute on any other receiver is not a candidate here: proving that
value.method is absent needs a proven receiver type, which this pass does
not have.
Enums§
- Python
Environment Boundary - What the analyzer’s retained environment evidence proves about one name a Python file reaches across an import boundary.
Constants§
Traits§
- Python
Environment Surface - The retained Python environment surface a diagnostic request may read.
Functions§
- collect_
python_ semantic_ diagnostics - Collect Python semantic diagnostics and the proof behind each one.
- is_
python_ builtin_ or_ constant