Got an unexpected number of seconds in datetime from asyncpg

[rossnomann]

• asyncpg version: 0.17.0
• PostgreSQL version: 10.3
• Do you use a PostgreSQL SaaS? If so, which? Can you reproduce
  the issue with a local PostgreSQL install?: reproduced with a local install
• Python version: 3.6.5
• Platform: Linux x86_64
• Do you use pgbouncer?: no
• Did you install asyncpg with pip?: yes

from datetime import datetime

import asyncio
import asyncpg


async def main():
    pool = await asyncpg.create_pool(
        database='postgres',
        user='postgres',
        host='127.0.0.1',
        password='1234',
        port=2000
    )
    async with pool.acquire() as connection:
        dt_in = datetime(1970, 1, 1, 20, 31, 23, 648000)
        dt_out = await connection.fetchval("SELECT '%s'::timestamp" % dt_in)
        op = '==' if dt_in == dt_out else '!='
        print('%s %s %s' % (dt_in, op, dt_out))


loop = asyncio.get_event_loop()
loop.run_until_complete(main())

Output:

1970-01-01 20:31:23.648000 != 1970-01-01 20:31:24.648000

But in psql all works fine:

postgres=# select '1970-01-01 20:31:23.648000'::timestamp;
        timestamp        
-------------------------
 1970-01-01 20:31:23.648
(1 row)

I guess there is a problem with time parsing in asyncpg.
Maybe seconds were rounded somewhere, like this:

>>> d
datetime.datetime(1970, 1, 1, 20, 31, 23, 648000)
>>> d.timestamp()
63083.648
>>> int(d.timestamp()) % 60
23
>>> round(d.timestamp()) % 60
24