On Tue, Oct 18, 2011 at 12:26 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Tue, Oct 18, 2011 at 10:06 AM, Peter Eisentraut <peter_e@gmx.net> wrote:
>>> No, I believe we are OK everywhere else. We are only ignoring the
>>> result in cases where we are trying to report errors in the first place.
>
>> The relevant code is:
>
>> while (len > PIPE_MAX_PAYLOAD)
>> {
>> p.proto.is_last = (dest == LOG_DESTINATION_CSVLOG ? 'F' : 'f');
>> p.proto.len = PIPE_MAX_PAYLOAD;
>> memcpy(p.proto.data, data, PIPE_MAX_PAYLOAD);
>> write(fd, &p, PIPE_HEADER_SIZE + PIPE_MAX_PAYLOAD);
>> data += PIPE_MAX_PAYLOAD;
>> len -= PIPE_MAX_PAYLOAD;
>> }
>
>> Which it seems to me we could change by doing rc = write(). Then if
>> rc <= 0, we bail out. If not, we add and subtract rc, rather than
>> PIPE_MAX_PAYLOAD. That would be barely more code, probably safer, and
>> would silence the warning.
>
> And it would break the code. The whole point here is that the message
> must be sent indivisibly.
How is that different than the chunking that the while loop is already doing?
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company