Regression in opencv 4.11 when reading double precision matrix data from yaml files

[lostoliv]

Reading this yaml file in opencv 4.10 works while failing with opencv 4.11 (all int values like 0 get corrupted in 4.11 - while 0.0 is fine):

%YAML:1.0
matrix1:
  rows: 1
  cols: 4
  dt: d
  data: [ 0, 0.0, 1, 1.0 ]

Note that changing dt: d to dt: f works fine.

You can run the following code to repro (assert failing only with 4.11):

# Good: python -m pip install --upgrade opencv-python==4.10.0.84
# Bad:  python -m pip install --upgrade opencv-python==4.11.0.86
import cv2, numpy as np
mat_str = '[ 0, 0.0, 1, 1.0 ]'
yaml_file =\
'%YAML:1.0\n'\
'matrix1:\n'\
'  rows: 1\n'\
'  cols: 4\n'\
'  dt: d\n'\
f'  data: {mat_str}\n'
node_name = 'matrix1'
fs = cv2.FileStorage(yaml_file, cv2.FILE_STORAGE_MEMORY)
mat = fs.getNode(node_name).mat().flatten()
expected = np.array(eval(mat_str))
print(f'mat from yaml: {mat}')
print(f'mat expected:  {expected}')
assert np.allclose(mat, expected)

I see some changes related to FileStorage in 4.11 (like opencv/opencv#26434) that might have introduced this regression.

[asmorkalov]

@dkurt Could you take a look? May it be related to int64 support?

[dkurt]

opencv/opencv#26846 resolves this issue too