crates/codegen/src/compile.rs (3)

4585-4592: ⚠️ Potential issue | 🟡 Minor

Use the same format varname as the module/class annotation symbol table.

symboltable.rs registers module/class __annotate__ scopes with "format", but Line 4589 appends ".format". Since enter_scope already copies symbol-table varnames, this creates an extra unused local and breaks co_varnames parity.

Proposed fix

         self.current_code_info()
             .metadata
             .varnames
-            .insert(".format".to_owned());
+            .insert("format".to_owned());

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@crates/codegen/src/compile.rs` around lines 4585 - 4592, The code is
inserting ".format" into the current code object's metadata.varnames, creating
an extra local that breaks co_varnames parity with symboltable.rs which
registers "format" for __annotate__ scopes; change the insertion in the block
that calls self.current_code_info().metadata.varnames.insert(...) to insert
"format" (no leading dot) so it matches the symbol-table/enter_scope behavior
and keeps co_varnames aligned with emit_format_validation() and the __annotate__
scopes managed in symboltable.rs.

---

1892-1911: ⚠️ Potential issue | 🟠 Major

Skip synthetic parents repeatedly when building __qualname__.

This only skips one parent. For a nested scope inside an annotation scope inside a type-params scope, the grandparent can still be TypeParams, so the generated qualname can include <generic parameters of ...> instead of the real owner.

Proposed fix

-        let parent_scope = self
-            .symbol_table_stack
-            .get(parent_idx)
-            .map(|table| table.typ);
-
-        // CPython skips both generic-parameter scopes and annotation scopes
-        // when building qualnames for the contained function/class code object.
-        if matches!(
-            parent_scope,
-            Some(CompilerScope::TypeParams | CompilerScope::Annotation)
-        ) || parent.metadata.name.starts_with("<generic parameters of ")
-        {
-            if stack_size == 2 {
-                // If we're immediately within the module, qualname is just the name
-                return current_obj_name;
-            }
-            // Use grandparent
-            parent_idx = stack_size - 3;
-            parent = &self.code_stack[parent_idx];
-        }
+        // CPython skips generic-parameter and annotation scopes when building
+        // qualnames for contained function/class code objects. These synthetic
+        // scopes can be stacked, so keep walking until a real parent is found.
+        loop {
+            let parent_scope = self
+                .symbol_table_stack
+                .get(parent_idx)
+                .map(|table| table.typ);
+            let skip_parent = matches!(
+                parent_scope,
+                Some(CompilerScope::TypeParams | CompilerScope::Annotation)
+            ) || parent.metadata.name.starts_with("<generic parameters of ");
+            if !skip_parent {
+                break;
+            }
+            if parent_idx == 0 {
+                return current_obj_name;
+            }
+            parent_idx -= 1;
+            parent = &self.code_stack[parent_idx];
+        }

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@crates/codegen/src/compile.rs` around lines 1892 - 1911, The current logic
only skips one synthetic parent when building __qualname__ (using parent_scope,
symbol_table_stack and checks for CompilerScope::TypeParams |
CompilerScope::Annotation and parent.metadata.name starting with "<generic
parameters of "), which can leave further synthetic ancestors in place; change
this to repeatedly skip synthetic parents by looping (while) to move parent_idx
up until the parent is not a TypeParams/Annotation and its metadata.name is not
the generic-parameters sentinel, updating parent_scope and parent from
symbol_table_stack and code_stack each iteration, and preserve the existing
early-return when the adjusted stack_size indicates the module-level parent
(using stack_size to decide when to return current_obj_name).

---

3011-3016: ⚠️ Potential issue | 🔴 Critical

Fix type-parameter evaluator parameter spelling to match type-alias evaluator and symbol table registration.

Change "format" to ".format" at line 3012. The type-parameter bound/default evaluator scope should use the same parameter name as the type-alias evaluator (compile_typealias_value_closure line 3068) and the symbol table registration (symboltable.rs lines 1803, 2323), which all use ".format". Using different spellings will create inconsistent co_varnames indices.

Current code

        self.current_code_info()
            .metadata
            .varnames
            .insert("format".to_owned());

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@crates/codegen/src/compile.rs` around lines 3011 - 3016, The varname inserted
into the current code info is misspelled — replace the string "format" with
".format" in the call to self.current_code_info().metadata.varnames.insert(...)
so the type-parameter evaluator parameter matches the type-alias evaluator (see
compile_typealias_value_closure) and the symbol table registration entries; this
ensures co_varnames indices remain consistent with emit_format_validation() and
the symbol table registrations.