[bfs2021]

Problem

The wheels ship .pyi stubs for the whole package, but no py.typed marker (PEP 561), so mypy ignores them and every ioh import degrades to Any (Skipping analyzing "ioh": module is installed, but missing library stubs or py.typed marker). Fixes #237.

Root cause

An empty ioh/py.typed is missing from the package. With the marker in place, mypy starts honouring the stubs — and then reports 22 errors in the stubs themselves, so the marker alone would trade one problem for another.

Fix

• Add the empty ioh/py.typed; wheel.packages = ["ioh"] and the sdist's /ioh/ include pick it up with no packaging changes.
• Fix the 22 stub errors the marker exposes:
  • wrap_problem, Experiment.__init__ and add_custom_problem: None defaults on parameters typed str / Callable / list → explicit | None;
  • Experiment.apply / add_custom_problem: any / callable used as types → typing.Any / typing.Callable[..., typing.Any];
  • iohcpp/logger/__init__.pyi: the property submodule was re-exported under the name property, shadowing the builtin decorator for every class-level @property in the file (9 × "Module not callable"); it is now imported by full path under a local alias;
  • iohcpp/problem/__init__.pyi: the 1-D and 2-D __call__ overloads were indistinguishable to mypy (the Annotated shape metadata is invisible to it), so the 2-D signature was reported as unreachable; the shape is now expressed in the ndarray type parameter;
  • MetaData.__eq__ widened to the object supertype signature, and the unhashable marker kept via an explicit ignore (mypy 2.x no longer accepts a bare __hash__ = None).

Testing

• mypy -p ioh: 22 errors → Success: no issues found in 12 source files.
• Simulated the installed-package experience with a stub-only copy (.pyi + py.typed, no compiled module): reveal_type(ioh.get_problem) now reports the full signature from the issue's expected output instead of Any.