[Python-checkins] SQLite test hangs - was: Re: Python Regression Test Failures basics (1)

[Neal Norwitz]

On Sun, Mar 9, 2008 at 3:39 AM, Gerhard Häring <gh at ghaering.de> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
>  Hash: SHA1
>
>  Neal Norwitz wrote:
>  > Gerhard,
>  >
>  > There is another problem with test_sqlite on the trunk.  In
>  > Lib/sqlite3/test/transactions.py two functions cause hangs:
>  >
>  >   CheckLocking
>  >   CheckRaiseTimeout
>  >
>  > which is causing the builtbots to fail.
>  >
>  > Could you take a look?
>
>  I tried to investigate this one.
>
>  First, I couldn't find the hangs you described in the buildbot status
>  pages.

It's only one machine.  Here's the most recent:
 http://www.python.org/dev/buildbot/all/x86%20gentoo%20trunk/builds/3169/step-test/0

>  Second, my first guess was that it was caused by old SQLite versions
>  rather than SQLite. I set up a test environment in which I build and
>  test pysqlite trunk (which is almost exactly the same code as the
>  sqlite3 module) against all SQLite versons from 3.0.8 to 3.5.6.
>
>  I couldn't reproduce the hang (this is Linux on x86; Ubuntu 7.10). I'm
>  currently out of ideas, as I don't have a hardware/software
>  combination available in which I can reproduce the hang.

If you give me some ideas for where to look/how to debug, I can try to
find the problem on this machine.  I don't know if it's specific to
o/s, sqlite version, compiler version, etc.  It could be indicative of
a larger problem or not.  No way to know without finding the cause, so
I'd like to try if possible.

>  What I did find, however, was that two aggregate-related tests fail
>  with SQLite verson 3.5.5 and 3.5.6. I'm 99.999 % sure it's caused by
>  SQLite swapping out its stack-based VM in favour of a new
>  register-based one.  So I can wait out this particular problem ;-)

:-)

n