[Box Backup-dev] COMMIT r245 - box/trunk/infrastructure/m4

[Martin Ebourne]

On Mon, 2005-12-19 at 23:10 +0000, Jonathan Morton wrote:
> > That's odd. What happens if you edit
> >     infrastructure/m4/ax_check_bdb_v1.m4
> > and change the line that says
> >     AC_SEARCH_LIBS([__db185_open],
> > to
> >     AC_SEARCH_LIBS([db185_open],
> 
> No difference, as far as I can tell.
> 
> However, I notice that the code emitted by configure for this test  
> actually tries to call dbopen(), but the linker complains that  
> __db185_open() is missing.  This tells me that db_185.h is doing  
> something weird that breaks.

Interesting. db_185.h unsurprisingly #defines dbopen to __db185_open so
that the same .so can provide v4 and v1.85 support. That's exactly how
the version installed with fedora works. The odd thing is that your .so
doesn't seem to match your header file.

> I think the real fix is to move to proper db4 support, but I have no  
> idea what that entails.

I agree and is what I plan. It should be fairly easy - the db4 stuff is
all in the tree now, so really just a configure issue. However, I didn't
want to do this until after the next release is out because every change
delays it. I think now though I'll do it before.

However I would like all the tests to pass before changing any more
functionality. I've only had a little time to spend on it (I think we
all have day jobs, even if we aren't all emigrating!), but I've worked
out that the raidfile failures are only on 64 bit. No further than that
though. I need to fix the time overflow problem on 64 bit too.

Also I'm on the way to tracking the bbackupd failure down. It happens
when testbbackupd deliberately sets some kind of marker in the backup
store forwards and then checks that bbackupd stops backing up. It
doesn't. About line 706 of the test I think - it checks for a return
code of 2 from bbackupquery 'compare' to indicate differences but it
gets 1 to indicate it all compared equal. No idea any more than that
unfortunately. This is the only failure I'm aware of now on 32 bit
Linux. I tracked the failure down to the first win32 merge onto trunk.
Unfortunately I doubt Chris knows what it is because he didn't write a
lot of the code. A look at the test, bbackupquery, and bbackupd diffs
showed nothing interesting, so must be in one of the libs.

In summary, I'm aware of only 1 test failure on 32 bit, and a further 2
on 64 bit.

Cheers,

Martin.