Dropping TDB

[Matthias Andree]

Stefan Bellon <sbellon at sbellon.de> writes:

> Gyepi SAM wrote:
>> On Mon, Oct 25, 2004 at 06:50:35PM +0200, Matthias Andree wrote:
>> > Speaking of RISC-OS, when we switch over to 0.93, making the new
>> > BerkeleyDB Transactional Store code the default (to avoid data base
>> > corruption), we're considering discontinuing support for TDB (this
>> > is almost sure) and QDBM.
>
> Matthias' original message hasn't arrive here yet, therefore I'm
> replying to Gyepi's message.

I'll Bcc: you just in case.

> What does dropped mean?

Removed, all traces. The datastore_qdbm.c file, the corresponding
configure option and relevant documentation.

> You may have noticed that QDBM didn't need much attention over the
> last time. If the interface stays like it is at present, then I'm
> willing to make occasional changes if necessary. If however the
> interface changes completely and is about to change frequently in the
> future, then I won't update bogofilter for RISC OS anymore and just
> fork the current version out. I had hoped that this step will only
> come some time after 1.0 and not before ...

The question is if the reason to offer several DBs at all still
persists. I'd be more comfortable with a streamlined code that had less
options to build and run, which will ease support, results in better
testing and usually better code quality.

I have added stubs (dummy functions that do nothing) to satisfy the TXN
linker "interface" for TDB and QDBM on the TXN branch (untested though);
obviously, this will not provide the user with transactional interface
features, commonly named ACID, Atomicity (very important), consistency
(very important), integrity (also important), durability (importance is
a matter of personal attitude...), and I cannot judge how many of these
traits TDB and QDBM can offer, considering power outages or
software/kernel crashes. The TXN branch offers all four traits when the
requirements of BerkeleyDB WRT atomicity and durability of writes
(various synchronous I/O completion points in POSIX) are met.

-- 
Matthias Andree