fix(unchecked_base_model): handle bare dict in construct_type

[devteamaegis]

What's broken

Calling construct_type(type_=dict, object_={...}) raises ValueError: not enough values to unpack (expected 2, got 0). Any Pydantic model field annotated as bare dict (rather than Dict[str, Any]) triggers this crash during deserialization.

Why it happens

get_args(dict) returns an empty tuple, so the two-target unpacking key_type, items_type = get_args(type_) fails when no type parameters are present.

Fix

Replace the bare unpacking with args = get_args(type_); key_type, items_type = args if args else (typing.Any, typing.Any). This makes a bare dict behave identically to Dict[Any, Any].

Test

Added tests/test_unchecked_base_model.py::test_construct_type_bare_dict_no_unpack_error which calls construct_type with type_=dict and asserts the result equals the input dict without raising.

Fixes #774

---

Note

Low Risk
Small defensive change in deserialization coercion plus a unit test; no auth, security, or API contract changes.

Overview
Fixes a crash in construct_type when the declared type is an unparameterized dict. get_args(dict) is empty, so unpacking into key/value element types raised ValueError during model construction/deserialization for fields typed as bare dict.

The change reads type parameters from get_args and, when none are present, treats the mapping as Dict[Any, Any] (keys and values passed through construct_type without failing). A regression test asserts construct_type(type_=dict, object_={...}) returns the input mapping unchanged.

^{Reviewed by Cursor Bugbot for commit e6db0dd. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Reviewed by Cursor Bugbot for commit e6db0dd. Configure here.}

[cursor]

Unnecessary sys.path hack inconsistent with other tests

Low Severity

The sys.path.insert hack is inconsistent with every other test file in this project, which all import cohere directly without path manipulation. The project uses pyproject.toml with packages = [{ include = "cohere", from = "src"}] and expects the package to be installed in the environment (via poetry install or equivalent). This workaround bypasses that and could cause subtle import-ordering issues if the installed package and source tree ever diverge.

 

^{Reviewed by Cursor Bugbot for commit e6db0dd. Configure here.}