tests/test_voxel_eval_cli.py

import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "tools" / "voxel-model-eval"))
import eval as ev


def test_parse_models_arg_single():
    assert ev.parse_models_arg("Gemma4-12B=unsloth/gemma-4-12b-it-GGUF") == [
        ("Gemma4-12B", "unsloth/gemma-4-12b-it-GGUF")
    ]


def test_parse_models_arg_multiple():
    out = ev.parse_models_arg("A=org/a,B=org/b")
    assert out == [("A", "org/a"), ("B", "org/b")]


def test_parse_models_arg_empty_returns_empty():
    assert ev.parse_models_arg("") == []


def test_parse_models_arg_tolerates_whitespace():
    assert ev.parse_models_arg(" A = org/a , B = org/b ") == [("A", "org/a"), ("B", "org/b")]


def test_merge_rows_appends_new_model():
    existing = [{"model": "A", "prompt": "cat"}]
    new = [{"model": "B", "prompt": "cat"}]
    assert ev.merge_rows(existing, new) == existing + new


def test_merge_rows_overwrites_same_model():
    # re-testing model A drops ALL its old rows (incl. other themes) then appends
    # the fresh run; model B is untouched.
    existing = [
        {"model": "A", "prompt": "cat"},
        {"model": "A", "prompt": "snake"},
        {"model": "B", "prompt": "cat"},
    ]
    new = [{"model": "A", "prompt": "cat"}, {"model": "A", "prompt": "snake"}]
    out = ev.merge_rows(existing, new)
    assert out == [{"model": "B", "prompt": "cat"}] + new
    # no duplicate A rows
    assert sum(1 for r in out if r["model"] == "A") == 2


def test_merge_rows_into_empty():
    new = [{"model": "A", "prompt": "cat"}]
    assert ev.merge_rows([], new) == new