Re: plpython issue with Win64 (PG 9.2)

On 27/06/12 13:57, Jan Urbański wrote:
> On 27/06/12 11:51, Asif Naeem wrote:
>> Hi,
>>
>> On Windows 7 64bit, plpython is causing server crash with the following
>> test case i.e.
>>
>> CREATE PROCEDURAL LANGUAGE 'plpython3u';
>>> CREATE OR REPLACE FUNCTION pymax (a integer, b integer)
>>> RETURNS integer
>>> AS $$
>>> if a> b:
>>> return a
>>> return b
>>> $$ LANGUAGE plpython3u;
>>> SELECT pymax(1, 2);
>
>>
>> I think primary reason that trigger this issue is when Function
>> PLyUnicode_Bytes() calls "PyUnicode_AsEncodedString( ,WIN1252 /*Server
>> encoding*/, ) " it fails with null. I built latest pg 9.2 source code
>> with
>> python 3.2.2.3 by using Visual Studio 2010. Thanks.
>
> I'll try to reproduce this on Linux, which should be possible given the
> results of your investigation.

Your analysis is correct, I managed to reproduce this by injecting

serverenc = "win1252";

into PLyUnicode_Bytes. The comment in that function says that Python
understands all PostgreSQL encoding names except for SQL_ASCII, but
that's not really true. In your case GetDatabaseEncodingName() returns
"WIN1252" and Python accepts "CP125".

I'm wondering how this should be fixed. Just by adding more special
cases in PLyUnicode_Bytes?

Even if we add a switch statement that would convert PG_WIN1250 into
"CP1250", Python can still raise an exception when encoding (for various
reasons). How about replacing the PLy_elog there with just an elog? This
loses traceback support and the Python exception message, which could be
helpful for debugging (something like "invalid character <foo> for
encoding cp1250"). OTOH, I'm uneasy about invoking the entire PLy_elog
machinery from a function that's as low-level as PLyUnicode_Bytes.

Lastly, we map SQL_ASCII to "ascii" which is arguably wrong. The
function is supposed to return bytes in the server encoding, and under
SQL_ASCII that probably means we can return anything (ie. use any
encoding we deem useful). Using "ascii" as the Python codec name will
raise an error on anything that has the high bit set.

So: I'd add code to translate WINxxx into CPxxx when choosing the Python
to use, change PLy_elog to elog in PLyUnicode_Bytes and leave the
SQL_ASCII case alone, as there were no complaints and people using
SQL_ASCII are asking for it anyway.

Cheers,
Jan