Inconsistent error message within and outside of functions

[JanDitzen]

I came across very inconsistent error messages while writing a program which picks a maximum given a set of conditions. The bug (?) appeared when a subscript was invalid, mata returned the correct error code but pointed to a _editvalue() several lines above. Figuring the error out took me some time because of the inconsistent messages. When I write down the function in a sandbox, mata also returns inconsistent error messages.

A working example (just an example!) of the code is:

mata:

function checkBug(real matrix mat)
{
    N = rows(mat)    
    maxindex(mat,1,max_indic=.,tmp=.)
    _editvalue(mat,0,.)
    "msg after _editvalue()"
    if (mat[max_indic+1,1] == . ) check1 = 0
    return(max_indic)
}

Clearly, the code should fail if the maximum is the last element in mat. This is a coding error. I would expect the error message that a subscript in invalid.

If I run

: checkBug((1,2,3,4,5,6)')
  msg after _editvalue()
            _editvalue():  3301  subscript invalid
              checkBug():     -  function returned error
                 <istmt>:     -  function returned error
(1 line skipped)

The message is confusing in several ways. First of all the function which reports the error is_editvalue(), but mata tells me the right number of lines skipped. Secondly the code after (!) _editvalue runs. Obviously, the error is that the subscript in if (mat[max_indic+1,1] == . ) check1 = 0 is invalid. So part of the error message is correct, it just points to the wrong function.

If I replace _editvalue(mat,0,.) with rnormal(max_indic,1,0,1) then mata returns

 rnormal():  3301  subscript invalid

So pointing again to the wrong function.

I then checked the code outside of a function, so I coded:

mata:
    mat = (1,2,3,4,5,6)'
    N = rows(mat)    
    maxindex(mat,1,max_indic=.,tmp=.)
    _editvalue(mat,0,.)
    if (mat[max_indic+1,1] == . ) check1 = 0
end

Here the error message became:

unexpected end of line
(0 lines skipped)

I somewhat understand that mata has a problem because mat[max_indic+1,1], but why does it report unexpected end of line? Should it not realise that it is a invalid subscript?

If the code is instead:

mata:
    mat = (1,2,3,4,5,6)'
    N = rows(mat)    
    maxindex(mat,1,max_indic=.,tmp=.)
    _editvalue(mat,0,.)
    (mat[max_indic+1,1] == . )    
end

or

mata:
    mat = (1,2,3,4,5,6)'
    N = rows(mat)    
    maxindex(mat,1,maxi=.,tmp=.)
    _editvalue(mat,0,.)
    i = 0
    check = 0
    while (check==0) {
        i++
        max_indic = maxi[i,1]
        if (mat[max_indic+1,1] == . ) check1 = 0
    }    
end

Then the error message is "correct":

                 <istmt>:  3301  subscript invalid
(0 lines skipped)

My question is, what is the reason for the difference if the code is within a function or not? I do know that mata compiles functions first, but why does the error point to the wrong command? In the sandbox mode out of a function, the error leads to different messages as well.

Any help would be highly appreciated.

[Sebastian Kripfganz]

JanDitzen
I can replicate your problem. If I replace _editvalue(mat,0,.) with

mat = editvalue(mat,0,.)

I receive the expected error message:

: checkBug((1,2,3,4,5,6)')
  msg after _editvalue()
              checkBug():  3301  subscript invalid
                 <istmt>:     -  function returned error

It might be worth reporting this to Stata's technical support team.