In 8.2.1:

octave:13> s.a = 1; s.b = 2;
octave:14> ["", s]
error: octave_base_value::convert_to_str_internal (): wrong type argument 'scalar struct'
octave:15> ['', s]
error: octave_base_value::convert_to_str_internal (): wrong type argument 'scalar struct'
octave:16> [[],s]
ans =

  scalar structure containing the fields:

    a = 1
    b = 2

octave:17> version
ans = 8.2.1

Dmitri.
--

Just updating that this is still around in 6.4.0. 

I suspect the fact that strings fail in Matlab (still fails in v2021a) is just part of Matlab being sloppy about consistency with string class implementation, as we've seen pop up elsewhere. Wouldn't surprise me to see ["",s] suddenly start working in a future version. but since we don't currently have a string class, it's probably not a huge point of difference for Octave.

still exists in 4.2.0. Octave errors on 4 empty string examples in original submission, while Matlab 2016b produces:

>> [[],s]
ans =
  struct with fields:

    a: 1
    b: 2

I changed the category.  I wouldn't close this report though, it's sitll a valid Matlab incompatibility.  I guess the rule is just to check whether something is empty and if so skip it in the concatenation.  I don't know for sure though.

The issue is still present (3/16/14).  Concatenation of an empty string value, '' or "", with a struct results in an error.

s.a = 1; s.b = 2;
['', s]
["", s]
[s, '']
[s, ""]

[s, '']
error: octave_base_value::convert_to_str_internal (): wrong type argument 'scalar struct'

The following should all work

s.a = 1; s.b = 2;
[[], s]
['', s]
["", s]
[s, []]
[s, '']
[s, ""]

but the character string versions fail on invalid conversions.

Should we be treating literal null matrices specially in tree_matrix::rvalue1 in pt-mat.cc?  Because of the way that function handles initial empty values, it seems that functions like
oct_catop_matrix_struct in op-struct.cc are not called in all cases.  Or maybe they are never called?  In what cases are they be called?  Should they be called for all empty value cases?  Just for non-literal empty matrices?  At this point, I'm obviously not sure what the right fix is...