Submitted by: @pavel-zotov

Consider script:

recreate exception foo 'Invalid value detected: @1';
set list on;
set echo on;

set term ^;
execute block as
declare x smallint; -- 'x' is NOT initialized
begin
x = 33000; -- make 'x' too big to fit in smallint via direct assignment
when any do
begin
exception foo using(:x);
end
end
^

execute block as
declare x smallint = 33; -- 'x' is initialized
begin
x = 33000; -- make 'x' too big to fit in smallint via direct assignment
when any do
begin
exception foo using(:x);
end
end
^

execute block as
declare x smallint = 33; -- 'x' is initialized
begin
x = x * 1000; -- make 'x' too big to fit in smallint via arithmetic expression
when any do
begin
exception foo using(:x);
end
end
^
set term ;^

select 33000 - 65536 from rdb$database;

It's STDERR will be:

Statement failed, SQLSTATE = HY000
exception 46
-FOO
-Invalid value detected: *** null ***
-At block line: 7, col: 13

Statement failed, SQLSTATE = HY000
exception 46
-FOO
-Invalid value detected: -32536
-At block line: 7, col: 13

Statement failed, SQLSTATE = HY000
exception 46
-FOO
-Invalid value detected: -32536
-At block line: 7, col: 13

First exception text (with "-Invalid value detected: *** null ***") can be easy explained: statement did not change variable content which was null at the beginning.
Second and third messages produces NOT invalid value but its complementation to 65536 with minus sign.