2.4 and sqlite...

[Phil Longstaff]

On Sun, 2011-01-02 at 08:23 +0100, Gour wrote:
> On Sat, 1 Jan 2011 16:17:06 -0500
> >>>>>> "Derek" == wrote:
> 
> Derek> Because it's not *our* bug, it's a bug in libdbi, 
> 
> I'm aware it's not Gnucash bug...
> 
> Derek> and it's a bug Derek> that only affects certain builds of libdbi, 
> 
> but I didn't know about the latter.

I'm not positive, but I think it might only affect 64-bit builds or
builds on powerPC architecture.

> This is something I do not understand fully...Based on what I see at
> libdbi site, it looks that it's not in such a rapid development
> producing new releases so often, iow the site says:
> 
> libdbi-0.8.4  2010-09-01 
> libdbi-0.8.3  2008-02-06 
> 
> which means there are two releases in last (almost) three years.
> 
> Now, which version of the libdbi is broken?

0.8.3 is broken.  0.8.4 is probably also broken, since the 0.8.4 release
notes mention that there have only been build system changes, not code
changes.  The proper fixes have been put into the libdbi code repository
but have not been released yet.  When I ask about a new libdbi release,
they mention test improvements which do not pass for all platforms
and/or databases (they support more than just sqlite3, mysql and pgsql).
So, if you want a version of libdbi that works properly, the only sure
way is to build it from source.  On win32, we do build it from source,
but only for 32-bit windows, not win64.  We don't build for all possible
linux distros.  Once an new libdbi is released with the fixes, then
users who don't want to build their own need to wait for their disro to
pick up the new source and build it.

Libdbi was chosen for the sql backend because it was thought to be the
best maintained and most stable.  An alternative, libgda, was undergoing
a major architectural and API change.  If we had to, I suppose we could
pull a copy of libdbi into gnucash, but I would only do that as an
absolute last resort.

> I'll try to build gnucash on my Archlinux system, but for now I can
> only say that libdbi is built with:  ./configure --prefix=/usr
> so I do not know what does it mean in regard to '-fast-math' option.

I'm not sure what the exact issue is, so I can't help.

> Still (although I very much admire Gnucash and its devs), I believe
> that it could be that not-too-many devs were testing with Sqlite
> back-end which is important considering that SQL storage is some of
> the 'hot stuff' proudly announced as 'major changes' in 2.4.0.

One of the reasons for all of the 2.3.X releases was to try to get the
sql backend tested better than the developers were able to.  I'm the
major sql backend developer and have been using it while it was being
developed.  I have never used all of the business features, so my
testing for those features has been minimal.  I use 32-bit linux, so
never tested at all on a 64-bit system.  I do dual-boot windows so have
done minimal windows testing.  I don't have a mac so have to rely on
others for mac testing.  By releasing development versions, we hope that
more people testing will cover more configurations and provide feedback
which we can't easily test on our own.

Phil