>Ok; what you are saying makes sense. But what puzzles me is why I don't get the error on my system?
>I can change the code to your suggestion but first I would like to duplicate the problem on my side.

Dmitry,
First line of defense is a polished error handler blasting to the max into text file giving you detailed info on current stack, mem values, list of DS each with currentalias, list of open workarea, alias, dbf, recno, fieldvalues of said recno, buffered changes of each workarea. Also list of suspicious objects and their properties: this, this.parent while there is one, thisform, activepage. currentcontrol and so on

This should answer your current pressing problem. Of remaining cases In 20-30% something on client machine results in the error - for those cases we would get total directory of app (including tables and tempdata) and backup of backend where applicable. Either as RAR or via Acronis.

Second line of defense in the program is logging: you instumentalize the code to have external switches, piping relevant runtime info/events into text file started with candidate filename and cleanup a week later.
This can answer "who closed my alias", when culprit is long gone from the stack when error occurs.

But I guess you should look for a full fledged error handler first.
If you had one you would have written:
my controlsource is mytable.myfield,
alias() is whatever
mytable is on recno ##, not eof or bof or deleted bla bla

build on existing errorhandler either from a framework, one or more of your tools or buy.
Takes some time until you navigate the error file to find error causes early on:
you will get faster!