import json
from types import SimpleNamespace

import pytest

from pallet.annotation import next_teacher_subset


def test_next_teacher_requires_resolved_source_frames_and_excludes_audit(tmp_path, monkeypatch):
    protocol = tmp_path / 'protocol'
    protocol.mkdir()
    (protocol / 'protocol.json').write_text('{}')
    rows = [SimpleNamespace(frame_id=fid, group_id=group, domain=domain,
                            source_sha256=f'raw-{fid}')
            for fid, group, domain in [('a', 'source', 'config-a'),
                                        ('old-audit', 'audit', 'config-a'),
                                        ('new', 'new-group', 'config-b'),
                                        ('v', 'validation', 'config-a'),
                                        ('t', 'test', 'config-b')]]
    assignments = {'source': 'train', 'audit': 'train', 'new-group': 'train',
                   'validation': 'val', 'test': 'test'}
    labels = SimpleNamespace(frames={row.frame_id: SimpleNamespace(
        status='pos', group_id=row.group_id,
        source_anchor_hash=row.source_sha256) for row in rows})
    monkeypatch.setattr(next_teacher_subset, 'load_protocol',
                        lambda _: (rows, assignments, {}))
    monkeypatch.setattr(next_teacher_subset, 'load_jsonl', lambda _: labels)
    monkeypatch.setattr(next_teacher_subset, 'is_resolved',
                        lambda frame, require_blind=False: frame.status in {'pos', 'negative', 'ignore'})
    prior = tmp_path / 'previous.json'
    prior.write_text(json.dumps({'train': ['a', 'old-audit'], 'val': ['v'], 'test': ['t']}))
    selection = tmp_path / 'round.json'
    selection.write_text(json.dumps({'train': ['new']}))
    reviewed = tmp_path / 'labels.jsonl'
    reviewed.write_text('reviewed')
    audit = tmp_path / 'audit.json'
    audit.write_text(json.dumps(['audit']))

    output = tmp_path / 'next.json'
    result = next_teacher_subset.build_subset(protocol, prior, selection,
                                              reviewed, audit, output)
    assert result['train_frames'] == 2
    assert json.loads(output.read_text()) == {
        'train': ['a', 'new'], 'val': ['v'], 'test': ['t']}

    held = tmp_path / 'held.json'
    next_teacher_subset.build_subset(protocol, prior, selection, reviewed,
                                     audit, held, target_domain='config-b')
    assert json.loads(held.read_text())['train'] == ['a']

    labels.frames['new'].status = 'unreviewed'
    with pytest.raises(ValueError, match='unresolved'):
        next_teacher_subset.build_subset(protocol, prior, selection,
                                         reviewed, audit, output)


def test_supplemental_teacher_frames_are_blind_source_only_and_audit_excluded(tmp_path, monkeypatch):
    protocol = tmp_path/'protocol'
    protocol.mkdir()
    (protocol/'protocol.json').write_text('{}')
    rows = [SimpleNamespace(frame_id=fid, group_id=group, domain='source',
                            source_sha256=f'raw-{fid}')
            for fid, group in [('old', 'source'), ('new', 'source'),
                               ('extra', 'source'), ('audit', 'audit'),
                               ('val', 'val'), ('test', 'test')]]
    assignments = {'source': 'train', 'audit': 'train', 'val': 'val', 'test': 'test'}
    labels = SimpleNamespace(frames={row.frame_id: SimpleNamespace(
        status='negative', split=assignments[row.group_id], group_id=row.group_id,
        source_anchor_hash=row.source_sha256, provenance=[]) for row in rows})
    monkeypatch.setattr(next_teacher_subset, 'load_protocol',
                        lambda _: (rows, assignments, {}))
    monkeypatch.setattr(next_teacher_subset, 'load_jsonl', lambda _: labels)
    monkeypatch.setattr(next_teacher_subset, 'is_resolved',
                        lambda frame, require_blind=False: frame.status == 'negative')
    prior = tmp_path/'prior.json'
    prior.write_text(json.dumps({'train': ['old'], 'val': ['val'], 'test': ['test']}))
    selection = tmp_path/'selection.json'
    selection.write_text(json.dumps({'train': ['new']}))
    supplement = tmp_path/'supplement.json'
    supplement.write_text(json.dumps({'train': ['extra', 'audit']}))
    reviewed = tmp_path/'labels.jsonl'
    reviewed.write_text('frozen')
    audit = tmp_path/'audit.json'
    audit.write_text(json.dumps(['audit']))
    output = tmp_path/'next.json'
    result = next_teacher_subset.build_subset(
        protocol, prior, selection, reviewed, audit, output,
        supplemental_selection=supplement)
    assert json.loads(output.read_text())['train'] == ['old', 'new', 'extra']
    assert result['supplemental_source_frames'] == 2
    assert result['supplemental_train_frames'] == 1
    exclusion=tmp_path/'exclude.json'
    exclusion.write_text(json.dumps({'train':['extra'],'val':[],'test':[]}))
    without=tmp_path/'without_conflicts.json'
    excluded=next_teacher_subset.build_subset(
        protocol,prior,selection,reviewed,audit,without,
        supplemental_selection=supplement,exclude_frame_selection=exclusion)
    assert json.loads(without.read_text())['train']==['old','new']
    assert excluded['excluded_source_frames']==1
    assert excluded['exclude_frame_selection_sha256']
    with pytest.raises(ValueError,match='subset differs'):
        next_teacher_subset.build_subset(protocol,prior,selection,reviewed,audit,
                                         output,supplemental_selection=supplement,
                                         exclude_frame_selection=exclusion)
    supplement.write_text(json.dumps({'train': ['test']}))
    with pytest.raises(ValueError, match='independently reviewed supplemental'):
        next_teacher_subset.build_subset(protocol, prior, selection, reviewed,
                                         audit, output, supplemental_selection=supplement)
    supplement.write_text(json.dumps({'train': ['extra']}))
    labels.frames['extra'].provenance = [SimpleNamespace(source='detector')]
    assisted_history = tmp_path/'assisted_history.json'
    accepted = next_teacher_subset.build_subset(
        protocol, prior, selection, reviewed, audit, assisted_history,
        supplemental_selection=supplement)
    assert accepted['supplemental_train_frames'] == 1
    assert json.loads(assisted_history.read_text())['train'] == ['old', 'new', 'extra']

    labels.frames['extra'].status = 'unreviewed'
    with pytest.raises(ValueError, match='independently reviewed supplemental'):
        next_teacher_subset.build_subset(protocol, prior, selection, reviewed,
                                         audit, tmp_path/'unresolved.json',
                                         supplemental_selection=supplement)
