Overload decorator lines classified as Unanalyzed

Bug Report

@overload decorator lines are classified as Unanalyzed in --lineprecision-report. The lineprecision ratio (Precise + Imprecise) / (Lines - Empty) therefore drops with every overload added, even when all signatures are fully annotated.

To Reproduce

# mre.py
from typing import Literal, overload


class C:
    @overload
    def f(self, x: Literal["a"]) -> str: ...
    @overload
    def f(self, x: Literal["b"]) -> int: ...
    def f(self, x: Literal["a", "b"]) -> str | int:
        return "hello" if x == "a" else 42

$ mypy mre.py --lineprecision-report report/
Success: no issues found in 1 source file
$ cat report/lineprecision.txt
Name  Lines  Precise  Imprecise  Any  Empty  Unanalyzed
-------------------------------------------------------
mre      10        6          0    0      2           2

Expected Behavior

Every line is fully annotated and mypy reports no errors, so the ratio should be 8/8 = 100%. The 2 Unanalyzed lines are the two @overload decorator lines. overload is a well-typed symbol from typing; those lines should be Precise (or at minimum Empty), treating @overload as a pure annotation marker with no runtime value.

Actual Behavior

StatisticsVisitor.process_node looks up each expression in self.typemap. The NameExpr for @overload is not populated in the typemap (mypy treats it as a special form during semantic analysis), so typemap.get(node) returns None → type(None) → record_line(TYPE_UNANALYZED).

Each overload variant therefore contributes 1 Precise line (the def) and 1 Unanalyzed line (@overload). Adding more fully-annotated overloads strictly lowers the ratio, making it impossible to hold a stable lineprecision gate on any file with a growing overload set.

There is a related # TODO in the same file: pass # TODO: Handle overloaded functions, etc. (in record_call_target_precision).

Your Environment

• Mypy version used: 2.1.0
• Mypy command-line flags: --lineprecision-report
• Python version used: 3.13

Inizia da StatisticsVisitor.process_node, dove l’issue segnala che le espressioni dei decorator vengono classificate tramite self.typemap, e verifica il TODO correlato record_call_target_precision nello stesso file. Esegui l’esempio mre.py fornito con --lineprecision-report, quindi verifica che i decorator di overload non compaiano più come Unanalyzed e che il report raggiunga 8/8 righe precise.