[patch] Error on setenv(..., NULL, ...)

[Roland McGrath]

I like Eggert's approach.  Off hand it seems it would be better to simply
move the definition of VALLEN into the scope containing its uses, which is
the else branch of 'if (combined != NULL)'.  The only reason I can see not
to do that is to force the fault to be before the lock is taken.  If that
is the explicit intent of the code, then its comments should say so.