Re: [Firebird-docs] Scale of number and precision on floating data types

[Dmitry Y. <fir...@ya...>]

10.08.2016 23:35, Helen Borrie wrote:
>
>> I hang on a text part in fblangref25-datatypes.xml -> fblangref25-datatypes-floattypes:
>> Floating-point data types are examples of data stored in the DBMS with the precision matching the scale of the number.
>
>> What is the scale of number? Please make this clear. Otherwise I don’t get the correct translation.
>
> I don't like this at all.  I think it needs some input from Paul and
> Dmitry and we should rewrite it.

I agree the wording is incorrect. The Russian original is slightly 
better but still far from being a good one.

> The way I understand the storage of floating point numbers:
>
> FLOAT is stored as INTEGER (precision 9, scale 0) with the scale
> (stored separately) depending on the position of the decimal point.
>
> DOUBLE PRECISION is stored as BIGINT (18,0) with scale
> depending on the position of the decimal point.

Nope, floating point numbers are stored completely different to the 
fixed point numbers. FLOAT is single-precision format which occupies 4 
bytes. DOUBLE is double-precision format which occupies 8 bytes. But 
they have no relationship to INT/BIGINT despite the byte size (4/8 
bytes), the binary layout is different (see IEEE 754) and it embeds 
{sign, exponent, mantissa}.

I suppose the intended meaning was something like: floating point data 
types represent data stored in the database with the dynamic precision 
corresponding its storage format (physical size of the value).


Dmitry