Re: Portability concerns over pq_sendbyte?

Hi,

On 2018-06-13 22:02:13 +0100, Andrew Gierth wrote:
> >>>>> "Andres" == Andres Freund <andres(at)anarazel(dot)de> writes:
>
> >> unsigned char x = 128;
> >> pq_sendbyte(&buf, x);
> >>
> >> which I believe is not well-defined since pq_sendbyte takes an int8,
> >> and conversions of unrepresentable values to _signed_ integer types
> >> are (iirc) implementation-dependent.
>
> Andres> It's not implementation defined in postgres' dialect of C - we
> Andres> rely on accurate signed->unsigned conversions in a number of
> Andres> places.
>
> Converting signed integer to unsigned is ok as I understand it - what's
> happening here is the reverse, converting an unrepresentable unsigned
> value to a signed type.

Err, yes, I was thinking about that conversion. Sorry for the confusion.

> >> There are also some cases where pq_sendint16 is being called for an
> >> unsigned value or a value that might exceed 32767.
>
> Andres> Hm, which case were you thinking of here? The calls usually are
> Andres> exactly the types that the wire protocol expects, no?
>
> There are cases where it's not actually clear what the wire protocol
> expects - I'm thinking in particular of the number of entries in a list
> of parameter types/formats.

But that didn't change around the pq_send* changes? So I'm not sure I
understand how this is related? I mean I'm all for documenting the wire
protocol more extensively, but we can't just change the width?

Greetings,

Andres Freund