Codecov Report

Merging #448 (43c95cf) into master (d43160c) will increase coverage by 0.03%.
The diff coverage is 89.83%.

@@            Coverage Diff             @@
##           master     #448      +/-   ##
==========================================
+ Coverage   85.92%   85.96%   +0.03%     
==========================================
  Files         126      126              
  Lines        5216     5236      +20     
==========================================
+ Hits         4482     4501      +19     
- Misses        734      735       +1     

Help us with your feedback. Take ten seconds to tell us how you rate us.

@lawrence-mbf is this ready to merge?

@bendichter I don't think so. There was still some instability in the last hackathon that caused full MATLAB crashes and the "solution" in the last commit only really resolved it at the high level. Unfortunately, I narrowed this down to a MATLAB bug present in versions older than R2021a where H5F.close() would close all file IDs, even if you've cloned it properly.

To reproduce:

filename = 'test.nwb';
datasetPath = '/file_create_date';

fid = H5F.open(filename, 'H5F_ACC_RDONLY', 'H5P_DEFAULT');
sourceDid = H5D.open(fid, datasetPath);
sourceGid = H5G.open(fid, '/general');

% this can occur with both H5F.open and H5F.reopen
fidReopened = H5F.open(filename, 'H5F_ACC_RDONLY', 'H5P_DEFAULT');
% fidReopened = H5F.reopen(fid);
didReopened = H5D.open(fidReopened, datasetPath);

% The error only occurs when ALL HDF5 IDs are closed including the source file id.
% not closing any of the below ids keeps it "alive" and does not cause a crash.
H5D.close(sourceDid);
H5G.close(sourceGid);
H5F.close(fid);

data = H5D.read(didReopened); % MATLAB crashes here.

I see. Is there any way we could file this with Mathworks?

I suppose you could but it is technically fixed in R2021a and later. You could also bring it up in the working group I suppose.

I think they still maintain patches for older versions