Masked samples inconsistency when there is nodata in hdr file

[nmoreaud]

When I add nodata in a hdr file, this code behavior changes: dataset.sample(... masked=True) .

Expected behavior and actual behavior.

I need to distinguish pixels that have valid value and pixels that are out of bounds/nodata.

Steps to reproduce the problem.

Please look at joined example .
test_raster.zip

Output on my computer:

python test_raster.py 
path: bordeaux.bil
detected nodata constant: None
[70] False
[97] False
[0] [False]
[0] [False]
----------------------------
path: bordeaux_nodata.bil
detected nodata constant: 97.0
[70] [False]      *** how should I interpret this?
[--] [ True]
[--] [ True]
[--] [ True]

Environment Information

rio --show-versions
rasterio info:
  rasterio: 1.3.6
      GDAL: 3.5.3
      PROJ: 9.0.1
      GEOS: 3.11.1
 PROJ DATA: /home/dev/projects/test_raster/.venv/lib/python3.8/site-packages/rasterio/proj_data
 GDAL DATA: /home/dev/projects/test_raster/.venv/lib/python3.8/site-packages/rasterio/gdal_data

System:
    python: 3.8.10 (default, Nov 14 2022, 12:59:47)  [GCC 9.4.0]
executable: /home/dev/projects/test_raster/.venv/bin/python
   machine: Linux-5.15.90.1-microsoft-standard-WSL2-x86_64-with-glibc2.29

Python deps:
    affine: 2.3.1
     attrs: 21.4.0
   certifi: 2022.05.18.1
     click: 8.1.3
     cligj: 0.7.2
    cython: None
     numpy: 1.24.2
    snuggs: 1.4.7
click-plugins: None
setuptools: 44.0.0

Installation Method

Ubuntu 20.04
sudo add-apt-repository ppa:ubuntugis/ppa
sudo apt install libgdal-dev
gdal and rasterio Installed with pip.

[sgillies]

@nmoreaud for the first file [70] False means the sample is [70] and all values for that array are valid, the source raster has no masked values. For the second file [70] [False] means that the source raster may have masked values, but that the particular value of 70 is not masked and is a valid value.

[nmoreaud]

Ok, then 0 [False] means that 0 is a valid value, but it matches a point that is out of the raster. Is that normal?

[sgillies]

sample's default, for datasets that are all valid (no mask, no nodata value), is to return 0 outside the extent of the dataset.

I've just made a small change to rasterio such that your program produces this:

python /app/test_raster.py
path: bordeaux.bil
detected nodata constant: None
[70] [False]
[97] [False]
[--] [ True]
[--] [ True]
----------------------------
path: bordeaux_nodata.bil
detected nodata constant: 97.0
[70] [False]
[--] [ True]
[--] [ True]
[--] [ True]
----------------------------

For bordeaux.bil. the first [--] [ True] means that the value is outside of the dataset bounds and thus masked. This is what sample intends and advertises and there was a bug.