Send M errors back instead of disconnecting clients

This issue was logged in our internal bug tracker back in 2007 as artf3951. We applied a hotfix for the issue, which has since been essentially forgotten. The fix never made it into an official KIDS build, as far as I know. We need to review the fix to see if it's still applicable and include it in our build.

Here's the original bug report:

"When running VistA in GT.M, the error trap handling seems to simply disconnect clients, instead of sending M errors back across the network when they happen. This isn't as nice for clients/users, and makes it harder to do initial debugging on what's going on."

Here's the fix:

"I have the fixed the error in GT.M where if Vista encounter an error it would drop the client. The problem was in the routine %ZTER, the main Vista error trap handler.

Below is a section of code that illustrates the problem and fix:

;Get $ZB
 I ^%ZOSF("OS")["DSM"!(^%ZOSF("OS")["GT.M") D
 . Q:'$L($ZB) ; rhl/medsphere 20070518
 . F %ZTER11I=1:1:$L($ZB) S %ZTER11A=$E($ZB,%ZTER11I),%ZTER11B=$G(%ZTER11B)_$S(%ZTER11A?1C:$A(%ZTER11A),1:%ZTER11A)_","
 . S %ZTER11B=$E(%ZTER11B,1,$L(%ZTER11B)-1)
 . Q
 S:'$D(%ZTER11B) %ZTER11B=$ZB

the line I added is in blue above; it tests that $ZB has a length greater than 0. the problem was that when is $ZB="" then the for loop is not iterated, and therefore %ZTER11B is not defined. in the next line, S %ZTER11B=$E(%ZTER11B,1,$L(%ZTER11B)-1), $ZTER11B is referenced, but since it is not defined a new error is generated and the processed is halted.

I did a unit test with the new code, and a error message properly pops-up in the client."