On 17.08.2012 05:07, Bruce Momjian wrote:
> On Mon, Dec 5, 2011 at 02:53:22PM +0300, Dmitriy Igrishin wrote:
>> Hey,
>>
>> The section 46.4 describes the base data types used in messages.
>> http://www.postgresql.org/docs/9.1/static/protocol-message-types.html
>>
>> According to section 46.5
>> http://www.postgresql.org/docs/9.1/static/protocol-message-formats.html
>> there are cases when Int32 can be negative (e.g. see DataRow(B) message
>> description.)
>>
>> Thus, I would like to suggest to change the description of Int(i)
>> from
>> "An n-bit integer in network byte order ..."
>> to
>> "An n-bit signed integer in network byte order ..."
>
> OK, documentation updated.
Actually, in some cases the integers are signed, and in others unsigned.
For example, in a Bind('F') message, the number of parameters is an
Int16 according to the docs, but it is treated as unsigned. The maximum
number of parameters is 65535.
The sentence used to be factually correct, when it didn't mention
whether they're signed or unsigned. If we want to do better than that,
we'd need to go through all the mentions of IntN in the docs and
explicitly say which ones are signed and which ones unsigned. Perhaps
use Uint16 or Uint32 for the unsigned ones.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com