test(schema): use explicit shapes in testMultipleNullShapesDataset

[ehennestad]

Motivation

Problem — The prepare release workflow runs the test suite across every MATLAB release crossed with every platform, 36 jobs in total. tests.unit.schema.MultipleShapesTest/testMultipleNullShapesDataset failed on exactly one of them, R2022a on windows-latest, and passed on the other 35. The failure reported a shape mismatch between the exported and the re-read dataset:

Actual size:
    50     1
Expected size:
     1    50

The test assigned 100 randomly shaped arrays to a MultiNullShapeDataset and exported whichever one the last iteration happened to produce. Occasionally that is a row vector, which cannot survive a round trip: MatNWB writes any vector as a rank-1 HDF5 dataset, and MATLAB reads a rank-1 dataset back as a column. MATLAB resets its random number generator only at session start, so the shape reached on iteration 100 depends on how many random numbers the tests before it drew. A release-and-platform combination that draws a different number of them lands on a different shape, which is why the failure was confined to a single matrix job and could not be reproduced from the others.

Solution — Assign an explicit list of shapes instead of random ones, and export a fixed matrix. Every matrix job now exercises the same shapes, so the test either passes everywhere or fails everywhere.

What changed

• testMultipleNullShapesDataset assigns four named shapes — a scalar, a column vector, a row vector and a matrix — in place of 100 random ones.
• The value handed to the round trip is a fixed 7x5 matrix rather than whatever the loop left behind.
• The row vector case, which produced the original failure, is now covered on every run instead of appearing at random.

Implementation notes

The orientation loss is intended and is documented under dimension ordering: one-dimensional data has to be rank-1 in the file to stay readable by other NWB implementations, and MATLAB has no rank-1 array type to read it back into.

fid = H5F.create(tmp, 'H5F_ACC_TRUNC', 'H5P_DEFAULT', 'H5P_DEFAULT');
io.writeDataset(fid, '/rowvec', rand(1, 50));
H5F.close(fid);

written in MATLAB as : [1 50]
dataspace in the file: 50  (rank 1)
read back in MATLAB  : [50 1]

Assigning a row vector is still valid for the type — the schema shape is [[null], [null, null]] — so the loop keeps covering it. Only the exported value had to become orientation-stable.

Examples

The shape handed to the round trip

Standing in for the random numbers drawn by earlier tests with a variable offset into the default stream shows what the exported shape depended on. Offsets 82 and 284 both end on a row vector.

Before

randiMax = intmax('int8');
for i = 1:100
    if rand() > 0.5
        data = rand(randi(randiMax), 1);
    else
        data = rand(randi(randiMax), randi(randiMax));
    end
end

  stream offset    0 -> [3 88]
  stream offset   82 -> [1 113]
  stream offset  284 -> [1 44]

After

validShapes = {[1, 1], [23, 1], [1, 23], [7, 5]};
for iShape = 1:numel(validShapes)
    data = rand(validShapes{iShape});
end
data = rand(7, 5);

  stream offset    0 -> [7 5]
  stream offset   82 -> [7 5]
  stream offset  284 -> [7 5]

Sweeping 2001 offsets, 8 of them end on a row vector. That is the roughly 1-in-250 chance the test carried on every job.

How to test

nwbtest('Name', 'tests.unit.schema.MultipleShapesTest')

Checklist

• Have you ensured the PR description clearly describes the problem and solutions?
• Have you checked to ensure that there aren't other open or previously closed Pull Requests for the same change?
• If this PR fixes an issue, is the first line of the PR description fix #XX where XX is the issue number?

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.27%. Comparing base (51e3e32) to head (8af3196).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #881   +/-   ##
=======================================
  Coverage   95.27%   95.27%           
=======================================
  Files         234      234           
  Lines        8311     8311           
=======================================
  Hits         7918     7918           
  Misses        393      393           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.