There are only two instances in libinterp that I can find:

operators/op-bm-b.cc:82:  boolNDArray v2 = a2.bool_array_value (true);
operators/op-bm-bm.cc:123:  boolNDArray v2 = a2.bool_array_value (true);

If I look at the code in op-bm-b.cc I find

static octave_value
oct_assignop_conv_and_assign (octave_base_value& a1,
                              const octave_value_list& idx,
                              const octave_base_value& a2)
{
  octave_bool_matrix& v1 = dynamic_cast<octave_bool_matrix&> (a1);

  // FIXME: perhaps add a warning for this conversion
  //        if the values are not all 0 or 1?

  boolNDArray v2 = a2.bool_array_value (true);

  v1.assign (idx, v2);

  return octave_value ();
}

There is a FIXME note about warning if the conversion was from an array like you mentioned such as [1, 1, 2, 3].  However, these files are for operator functions that are only called when both objects are already logical values (either bool matrix or bool).  So, any testing on conversion from octave_value to a strongly typed bool class won't result in anything and is wasted effort.  Or, I should say I think this is the correct interpretation of these bits of code.

@jwe: Could you take a look at my comment #3?  I think this is the right line of attack for this bug, but I want to be sure before going further.

Everything turns into a project.  I thought this would be easy, and it mostly is, but then there are all sorts of questions that arise.  To have "logical" work on single-quoted strings is the easiest.  See the attached changeset.  The tricky thing is that the function I added, bool_array_value, is probably not correct.  The function I wrote is

+  boolNDArray bool_array_value (bool = false) const
+  { return boolNDArray (m_matrix); }

But, if I look at how it is implemented in ov-re-mat.cc, it seems like the input "warn" needs to be used.

boolNDArray
octave_matrix::bool_array_value (bool warn) const
{
  if (m_matrix.any_element_is_nan ())
    octave::err_nan_to_logical_conversion ();
  if (warn && m_matrix.any_element_not_one_or_zero ())
    warn_logical_conversion ();

  return boolNDArray (m_matrix);
}

char arrays do not hold NaN elements so that test isn't needed, but the function "any_element_not_one_or_zero" is.  Unfortunately, the class charNDArray does not implement this so it would require adding it to that class which is is a change in liboctave.  All of this is doable, but I'm starting to wonder if I'm on the right track.  Adding jwe to the CC list so he can comment.


(file #54088)

I checked with Matlab and the conversion from a character vector to a logical vector works as expected without throwing an error.

logical ('str')
 => [1 1 1]

Trying to perform logical on a string object DOES throw an error.

logical ("str")

I've marked this as a Matlab Compatibility issue.