Re[6]: [Firebird-devel] Computation problem in 1.5RC3

[Gabriel J. <gj...@co...>]

Hi Calin,

Monday, June 30, 2003, 2:49:40 PM, you wrote:

CIP> ----- Original Message ----- 

CIP> :)) You're absolutely right. I didn't think it through enough.

CIP> Of course there's always the posibility to write a UDF multiply lib with
CIP> extended
CIP> as intermediate type. ;-)

Of course :) One can write an udf to do almost all than Firebird
can't, but question is if some of these should'n go into
Firebird.

CIP> As for firebird this is the intended behaviour, and too many rely on it
CIP> already.

That's what I was afraid of...
As of "rely on", I don't understand how can the "corrected" multiply
broke existing applications. Are you saying that someone relies on the
overflow exception to check that the internal result does not exceed
18 digits?! Because what it gets in the target variable will surely be
less or equal to 18 digits anyway! The results that don't throw
exception now will remain unchanged even if the internal
multiplication is made on 128 bits!
As I see it, it will have much less impact on existing applications than,
let's say, short-circuit boolean evaluation that appeared in 1.5b3.

I still think that current behaviour should be changed, but as I see I
am the only one with that oppinion, so I'd better learn to live with
it ;-)

Thank you all for your patience,
Gabriel