Re: sqlite3

[Tim Cross]

Po Lu <luangruo@yahoo.com> writes:

> Arthur Miller <arthur.miller@live.com> writes:
>
>> I suggested hashmap serialzied to/from file. I used such approach
>> myself for key-value pairs. It worked fine for me, it was quite simple
>> to read/write it. Currently I am testing a thing, and I use just assoc
>> list to read/write it to a file, and it works fine for me too.
>
> I think the ideal choice would either be dbm (and the GNU variant, gdbm,
> which improves upon it in every area) or recutils.
>
> Which still raises the other important question of how to serialize
> data, and how to do it better than `read' and `print'.
>
>> Just because everyone uses sqlite is maybe not the best argument, but
>> anyway, sqlite is maybe faster when serializing that reading/writing
>> lisp objects.  I don't know, I am not familiar so much with elisp vs
>> sqlite. Also, there are many uses for relational databse than just
>> persisting user settings.  I am thinking of desktop applications a lá
>> Access/Excel and similar.
>
> What is special about the settings of MS Access or MS Excel that warrant
> a relational database for storing them?

It probably wouldn't be particularly hard to define a layer which sits
between the Emacs code and the 'database', which could be either sqlite3
or gdbm? users could decide which backend they prefer (either as a
compile time option or possibly even as a user configuration option).

However, I do agree with other comments about simply serialising lisp
objects (i.e. hash maps) probably being sufficient for a majority of
cases. Isn't this essentially what packages like bbdb and applications
like vm (mail) did?

In othe words, I don't know if we are really talking about a wholesale
replacement and move towards sqlite3/gdbm, but instead perhaps providing
an interface to (lets say sqlite3) which could be useful when you do
need a more relational model and limited SQL support.