[Octave-bug-tracker] [bug #51448] cell2mat doesn't preserve dimensions of empty input

[Hugo Milan]

Follow-up Comment #2, bug #51448 (project octave):

I recently upgraded my parallel package to the new version that contains this
bug fix. After that, the package stopped working.

Commenting out line 452 of the parcellfun.m fixes the code for me.

To give some context, my algorithm running in parallel outputs matrices (n x
10), where "n" depends on the inputs of the algorithm. Then, this algorithm is
run in parallel with m process using the pararrayfun.m. At the end, I expect
to obtain an output with the dimension (n x 10m), where the outputs from the
different parallel runs are concatenated into the column. So, I expect
pararayfun.m to return me an (n x 10m) matrix. This was working as expected
before I upgraded the parallel package (I believe it was version 3.1.1).

What I observed on my debugging was that, when the code gets in line 452 of
parcellfun.m, varargout{i} already has the output dimension (n x 10m), which
was obtained when running "varargout{i} = cell2mat (res(i, :));" in line 448.
So, it breaks when it runs "varargout{i} = reshape (varargout{i}, shape);" in
line 452 because "size(shape) = [1 m]" and "size(varargout{i}) = [n 10m]",
which means it cannot reshape varargout{i} to [1 m].

    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?51448>

_______________________________________________
  Message sent via Savannah
  https://savannah.gnu.org/