Codecov Report

Merging #987 (7da0125) into master (a7e94d6) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #987      +/-   ##
==========================================
- Coverage   91.85%   91.84%   -0.01%     
==========================================
  Files         101      101              
  Lines       12550    12550              
  Branches     2209     2209              
==========================================
- Hits        11528    11527       -1     
  Misses        683      683              
- Partials      339      340       +1     

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a7e94d6...7da0125. Read the comment docs.

The problem here is that get_data_dtype() could be int16 while the correct interpreted dtype could be float, if the scale factors are non-trivial.

This is just to say that I have seen this and will try to look through it soon, but I haven't had time to think about it carefully.

Apologies for the extremely slow response... 2021 kept getting away from me in a lot of ways, but hopefully I can be a little more organized this year.

Anyway, I've looked at this again, and the fix is going to be quite easy due to some other in-progress work (#1082). If you cherry-pick d4a6fed, you can use get_obj_dtype(img.dataobj) to determine the logical dtype of each image. I would probably then use numpy.result_type on the logical dtype of all images. So something like:

out_dtype = np.result_type(*(get_obj_dtype(img.dataobj) for img in images))
out_data = np.empty(out_shape, dtype=out_dtype)

The reason I'm not going purely on the single input is that you could concat a uint8 image and a float32 image and the result would depend on which one happened to be first.

Thanks @effigies. Sounds like it'd be best to wait for that commit and #1082 to be merged before updated this PR to incorporate the suggestions. Does that sync with your thoughts?