import detector as D


def _blocks(*paras):
    return D.classify([D.Block(D._norm(p)) for p in paras], D.PASTE_MIN_WORDS)


def test_split_keeps_every_character():
    for s in ['He said "stop." Then left. The U.S. economy grew.',
              "The model (trained in 2020.) Was good. It won [3]. [4] shows more. (See Fig. 2.) Next one.",
              "It rose by 3.5 %. Then it fell (see Fig. 2). Smith et al. reported this."]:
        assert " ".join(D.split_sentences(s)) == s


def test_windows_cover_every_sentence():
    lengths = [12, 30, 8, 25, 40, 5, 18, 22, 350, 9]
    w = D.make_windows(lengths)
    covered = {k for a, b in w for k in range(a, b)}
    assert covered == set(range(len(lengths)))
    assert all(a < b for a, b in w)


def test_analyse_and_summarise_with_a_fake_scorer():
    paras = ["This sentence is written by a machine for sure. Another machine sentence follows it here.",
             "A human wrote this one on a slow afternoon. And this one too, with some care."]
    blocks = _blocks(*paras)
    D.analyse(blocks, scorer=lambda texts: [1.0 if "machine" in t and "human" not in t else 0.0 for t in texts])
    stats = D.summarise(blocks, 0.8, 0.5)
    assert stats["analysed"] == sum(len(p.split()) for p in paras)
    assert 0 <= stats["pct"] <= 100
    assert all(len(b.scores) == len(b.sents) for b in blocks if b.kind == "para")


def test_score_cache_scores_each_text_once(monkeypatch):
    calls = []

    def fake(texts):
        calls.append(list(texts))
        return [len(t) / 100 for t in texts]
    monkeypatch.setattr(D, "_score_uncached", fake)
    monkeypatch.setattr(D, "_score_cache", {})
    out = D.score_texts(["bbb", "a", "bbb", "cc"])
    assert out == [0.03, 0.01, 0.03, 0.02]
    assert calls == [["a", "cc", "bbb"]]                    # unique, shortest first
    assert D.score_texts(["a", "dddd"]) == [0.01, 0.04]
    assert calls[-1] == ["dddd"]                             # only the new text reaches the model
    assert D.score_texts([]) == []


def test_names_and_case_citations_do_not_end_a_sentence():
    import textparse as TP
    cases = {
        "The Court in Justice K.S. Puttaswamy v. Union of India (2017) recognised privacy. It shapes the law.": 2,
        "It grew in the U.S. The economy then slowed. In the U.K. it did not.": 3,
        "The dose for group A. Results were clear. Written by J. Smith and A. Kumar in 2020. Then done.": 4,
        "The fluid moves at velocity v. The pressure drops. See Roe v. Wade for more.": 3,
    }
    for text, n in cases.items():
        assert len(TP.split_sentences(text)) == n, text
        assert D.split_sentences(text) == TP.split_sentences(text)