Fred, have you looked at the example I posted? The only 'sin' committed by the code that got a global by reference was to cause an implicit refresh of current relations by calling SET DATE, yet this caused a completely different, unrelated subsystem - symbolized by the RELA_() function - to break since the global was no longer visible (type([m.g_uTest]) == 'U').

In some cases the problem can be avoided by rewriting the interface from

function ModifyFoo (roFoo)
   * ...
   roFoo = m.oNewFoo
   return <whatever>

* elsewhere ...
if ModifyFoo(@m.g_oFoo) == <whatever>
   * ...
endif

to

function ModifyFoo (oFoo, ruStatus)
   * ...
   ruStatus = <whatever>
   return m.oNewFoo

* elsewhere ...
local uStatus
g_oFoo = ModifyFoo(m.g_oFoo, @m.uStatus)
if m.uStatus == <whatever>
   * ...
endif

But this means that the ModifyFoo() call can no longer participate in things like CASE statements and other conditionals unless a wrapper is written:

function ModifyGlobalFoo
   local uStatus
   g_oFoo = ModifyFoo(m.g_oFoo, @m.uStatus)
   return m.uStatus

if ModifyGlobalFoo() == <whatever>
   * ...
endif

Bottomline: the ModifyFoo() code has to be changed even though it does not deal with any globals directly, and the code that calls ModifyFoo() has to be changed one way or another. And the whole shebang just got a bit messier instead of leaner/meaner/cleaner during maintenance. ;-)