[Python-Dev] PPC Leopard buildbot failing sqlite test for Python 3.2

[Bill Janssen]

R. David Murray <rdmurray at bitdance.com> wrote:

> On Sat, 16 Oct 2010 13:14:20 -0700, Bill Janssen <janssen at parc.com> wrote:
> > There was a test added to the sqlite suite between 3.1 and the current
> > 3.x trunk, TestInTransaction.  It's failing pretty consistently on the
> > PPC Leopard buildbot (2.7 and 3.1 are testing OK).
> > Calling sqlite3.sqlite_version returns 3.4.0.
> > 
> > A couple of things come to mind:
> > 
> > * Does this require a different flavor of the sqlite libraries?  I
> >   looked through the 3.x docs and couldn't find any mention of a
> >   specific sqlite version.
> 
> That's what I was wondering (I see you found the relevant issue...),
> but it doesn't *seem* to be a version issue, but it would be
> cool if you could check it against 3.7 just for confirmation.
> 
> > * Is this a big-endian bug?  Might fail on SPARC, too, then.
> 
> It might be, except for two things: it passes on the Sparc Solaris 10
> buildbot,

For 3.x, the "SPARC Solaris gcc" buildbot says,

  [ 87/349] test_sqlite
  test_sqlite skipped -- No module named _sqlite3

> and the io module also makes use of a T_BOOL field and
> its tests pass on all buildbots.  On the flip side, the only other
> failure I've see was on the Debian Sparc buildbot.  Which claims
> it is running sqlite 3.7.2-1.  It works for me using 3.7.2 on Gentoo
> i86.
> 
> > What should I look at?
> 
> Good question :)
> 
> When I added this it seemed like such a simple thing.  Maybe the
> endianness issue is the one to look at, despite the pass on Solaris.

The SPARC architecture supports either big-endian and little-endian use
(as does the PowerPC -- actually, I think any program can switch
endianness on the PowerPC, except...); the OS gets to decide.  I'd
expect Solaris on SPARC to run big-endian, like OS X on PPC, and
Wikipedia says Ubuntu on SPARC runs big-endian, too.

Except: PPC Leopard is running on a G5 process, a PowerPC 970.  As far
as I know, that's the only PowerPC chip that couldn't run either-endian
-- it only runs big-endian.  So that code has to be running big-endian
on PPC Leopard.

Looks like it's not passing on any big-endian architecture.

Bill