Re: Portability concerns over pq_sendbyte?

On Wed, Jun 13, 2018 at 11:53:21AM -0700, Andres Freund wrote:
> On 2018-06-13 14:10:37 +0900, Michael Paquier wrote:
>> On Mon, Jun 11, 2018 at 02:25:44PM +0900, Michael Paquier wrote:
>>> On top of that it seems to me that we'd want to rename any new
>>> routines to include "uint" in their name instead of "int", and for
>>> compatibility with past code pq_sendint should not be touched.
>
> I'm very doubtful about this one, unless you mean that just the
> signature shouldn't be touched. Otherwise we'll just increase code
> duplication unnecessarily?

Yeah, actually that would be assuming that many modules use it, but that
does not seem to be much the case, at least from github's point of view.

>> And also pq_sendint64 needs to be kept around for compatibility.
>
> :(. Wonder if it's better to just break people's code.

Indeed. At least breaking compilation has the advantage of making
people directly aware of the change and think hopefully about them.

A research on github shows a bunch of people having copied of pqformat.h
as there are a bunch of copies of Postgres so with this much noise it is
not easy to find out what would be broken. In-core contrib and test
modules don't make use of those interfaces as well, except for hstore.
So that could be acceptable.

For pq_sendint there are many matches with printsimple.c.
--
Michael