from collections import Counter
import math

import cv2
import numpy as np
import torch

from pallet.mask_annotation_full_data import context,supervision_valid
from pallet.mask_annotation_full_train import epoch_indices,rf_target
from pallet.mask_annotation_update_data import START_SHA,raw_sha
from pallet import mask_annotation_update_train as training
from pallet.pose_review_geometry import canonical


def test_new_warm_update_is_fixed_budget_old_EMA_not_generic_restart():
    assert training.CONFIGS['m1']['epochs']==training.CONFIGS['m2']['epochs']==10
    assert training.CONFIGS['m2']['lr']<1e-4 and training.CONFIGS['m2']['microbatch']==2
    assert START_SHA['m2']=='92ce35476d4b0887a0e791dc20972c9640aab558ec876df84296b713ebe8ed4c'
    assert math.ceil(2*1896/8)*10==4740


def test_all_new_human_images_base_exposure_then_balanced_component_extra():
    records=[dict(component=i%240) for i in range(1896)];indices=epoch_indices(records,0)
    assert len(indices)==3792 and set(indices)==set(range(1896))
    assert min(Counter(indices).values())>=1
    assert indices==epoch_indices(records,0) and indices!=epoch_indices(records,1)


def test_actual_human_positive_pixels_survive_neighbor_ignore():
    mask=np.zeros((30,100),np.uint8);mask[3:25,5:95]=1
    roi=[0,0,100,30];other=dict(target_key='neighbor',bbox=dict(x=30,y=0,width=30,height=30))
    valid=supervision_valid(roi,[other],'current',mask)
    assert valid[mask>0].all() and not valid[0,40]


def test_context_has_no_target_polygon_or_pose_or_class_features():
    box=dict(x=100,y=80,width=100,height=30)
    assert context(box,[352,264])==[52,62,248,128]


def test_new_input_loader_uses_pixels_masks_and_validity_only(tmp_path,monkeypatch):
    monkeypatch.setattr(training,'ROOT',tmp_path)
    grey=np.arange(40*100,dtype=np.uint8).reshape(40,100);mask=np.zeros((40,100),np.uint8);mask[5:35,5:95]=255
    valid=np.ones((40,100),np.uint8)*255
    for name,image in [('im.png',grey),('mask.png',mask),('valid.png',valid)]:cv2.imwrite(str(tmp_path/name),image)
    record=dict(image='im.png',native_mask='mask.png',valid_mask='valid.png',class_id=9,component=5,
        polygon=[[0,0]]*16,pose='must_not_enter_network',ruler='must_not_enter_network',selected_faces='must_not_enter_network')
    inputs=training.Inputs([record],[576,576]);image,target,weight,canvas=inputs.batch([0],'cpu')
    assert image.shape==(1,3,576,576) and target.shape==weight.shape==(1,576,576)
    assert canvas.shape==(1,1,576,576) and torch.isfinite(image).all()
    changed=dict(record,class_id=1,component=239,polygon=[[10,10]]*4,pose='other',ruler='other')
    other=training.Inputs([changed],[576,576])
    assert torch.equal(inputs.items[0][0],other.items[0][0])


def test_raw_event_hash_does_not_canonicalize_original_UTF8():
    raw='{"note": "人工增点", "revision": 2}'
    assert raw_sha(raw)!=__import__('hashlib').sha256(canonical(__import__('json').loads(raw))).hexdigest()


def test_detector_targets_derive_from_mask_only_not_class_ID():
    mask=torch.zeros((20,50));mask[5:15,10:40]=1;valid=torch.ones_like(mask)
    target=rf_target(mask,valid)
    assert target['labels'].tolist()==[0] and target['boxes'].shape==(1,4)
    assert target['pixel_valid'].all() and target['masks'].shape==(1,20,50)
