Alex Peshkov escreveu:

> On Thursday 14 August 2008 17:58, Adriano dos Santos Fernandes wrote:
>
>> Dmitry Yemanov escreveu:
>>
>>> Imagine some user selecting two fields (debit and credit) from a
>>> procedure and someone else alters this procedure and changes the output
>>> parameters order to the opposite one. The message format is still the
>>> same, but the user application will start getting completely crazy
>>> financial information. Should we care about such dumb situations?
>>>
>> Yes, I think so. We can do it in DSQL verifying SQLDA members (type
>> information, relation and field name) instead of doing it in engine
>> format block.
>>
>
> Adriano, I suppose you see yourself that in general case this is not possible.
> Imagine that in the end of procedure someone changed:
>
> ret1 = UDF_debet(bla-la-la1);
> ret2 = UDF_kredit(bla-la-la2);
>
> to:
>
> ret2 = UDF_debet(bla-la-la1);
> ret1 = UDF_kredit(bla-la-la2);
>
> Type information is OK, no relation and field name at all. It's even possible
> to redeclare UDFs!
>
> I do not think we must take care about such cases - if format of messages is
> OK, it's absolutely useless to try to care about something else. Or we must
> always throw an error if some object is changed!

I think your example is not valid. On Dmitry example, it would be caused
only in "select *" or "insert into table values (...)" (i.e., without
field lists) and also with procedure input parameters.

The re-prepared statement should have consistent input and output with
the original one and it's simple to verify it. This is from client POV.

If you're arguing about procedure body, this is not related to
re-preparation verification.


Adriano