incorrect PostgreSQL version, failed to connect to PG10/Debian

[ldgeo]

• asyncpg version: 0.14.0
• PostgreSQL version: 10.1
• Python version: 3.6.4
• Platform: Linux Debian sid
• Do you use pgbouncer?: no
• Did you install asyncpg with pip?: yes
• If you built asyncpg locally, which version of Cython did you use?:
• Can the issue be reproduced under both asyncio and
  uvloop?: not related to event loop

I experience an issue when trying to connect to my local postgresql cluster. The error is :

  File "/home/ludo/.virtualenvs/pyvt/lib/python3.6/site-packages/asyncpg/serverversion.py", line 49, in <listcomp>
    versions = [int(p) for p in parts][:3]
ValueError: invalid literal for int() with base 10: '1 (Debian 10'

PostgreSQL version given by the server itself :

[ludo@postgres] # select version();
                                                 version                                                 
---------------------------------------------------------------------------------------------------------
 PostgreSQL 10.1 (Debian 10.1-3) on x86_64-pc-linux-gnu, compiled by gcc (Debian 7.2.0-19) 7.2.0, 64-bit
(1 ligne)

PostgreSQL was installed from the main debian repositories :

apt-cache policy postgresql-10
postgresql-10:
  Installé : 10.1-3
  Candidat : 10.1-3
 Table de version :
 *** 10.1-3 500
        500 http://httpredir.debian.org/debian sid/main amd64 Packages
        100 /var/lib/dpkg/status

I looked into the code but failed to find where the version is fetched from PG.
The version_string received in the split_server_version_string function is equal to "10.1 (Debian 10.1-3)" but it should be only "10.1" according to the rest of the function.

Any ideas to fix it ?

[elprans]

Sigh, arguably it's a Debian bug. There's no good reason why they should adulterate the PG_VERSION risking breakage like this. #251 has a workaround.

[ldgeo]

thanks for the quick fix.
For my understanding how do you retrieve the PG version ?

[elprans]

The version is reported by the server when the connection is established via a ParameterStatus protocol message.

[jodal]

It seems like this problem is now present in new/upgrades PostgreSQL databases on Heroku, so anyone running asyncpg on Heroku with a recent PostgreSQL version now needs to run asyncpg from master. It would be lovely to have a new release cut which includes this workaround.