I've noticed this too and had already done a patch for it. The stack info is not set for the exception:s thrown (in the Octave sense, not C++ sense) by the bytecode interpreter itself, rather than by a call to C++ error().

However the patch is "stuck" between some wip-patches that addresses some design problems with how multiple return values are handled on the bytecode interpreter stack, so I think it could wait a bit to avoid some rebasing/rework and not put too many pending patches on the tracker.

I've attached the patch anyways to show the concept, however it shouldn't be apply-able straight off and is not tested that much.



(file #55932)

In certain cases the byte-code interpreter does not provide all information from a thrown error. Especially the line number of the root cause error may be missing (making debugging quite difficult).

Steps to reproduce with 2 function files

function ret = test1(in);
  in_1 = in * 2;
  if in_1 < 100
    ret = cos(in);
  else
    ret = cos(in(2)); # causes error if "in" is a scalar
  endif
endfunction

function ret = test2()
  ret = arrayfun(@(x) test1(x),1:200,'uni',false);
endfunction

The thrown error message(s) do not show where the root cause error occurs and in which line:

>> test2()
error: in(2): out of bound 1 (dimensions are 1x1)

When the byte-code interpreter is deactivated (do not forget to clear test1 + test2), one sees the root cause error + line:

__vm_enable__(0);
clear test1 test2;
test2()
error: in(2): out of bound 1 (dimensions are 1x1)
error: called from
    test1 at line 6 column 5
    test2>@<anonymous> at line 2 column 23
    test2 at line 2 column 3