Commented by: Sean Leyne (seanleyne)

Karol,

Although the ALTER is the "operation", the fact that the body of the trigger is evaluated/verified before the existence of the trigger is (IMO) not significant. In fact, it can be thought of as logical, since the command syntax must be valid before executed.

The first example is syntactically valid/correct, so the command is attempted, which is why the "not found" is reported.

Commented by: @livius2

Sean,

i understand your point. Yes, whole script must be valid.
But how it can be validated if there is no context? You know, alter trigger does not have name of the table.
Then to validate something, you must refer to table first to see if OLD.FIELD exists in that table.

Trigger existence must be first step here.

Commented by: Sean Leyne (seanleyne)

Karol,

I also take your point, but trigger existence does not need to be the first step.

The fact that the error refers to "column unknown" can be considered as pointer to the user that their base syntax is missing the "... FOR {relation name}". But that fails to consider the DB triggers syntax for which there is no "... FOR {relation name}".

Commented by: @livius2

Sean,
another good point, but i this particular example,
error message refer exactly that FB know that it work with trigger. It know exactly with which one

unsuccessful metadata update
ALTER TRIGGER XYZXX failed
Dynamic SQL Error

FB know also that this is "ALTER" operation. For alter operation first should be check existence of the object.