Using a command-line program (sqlite3) as a backend

[florian]

It seems to me I have a rather basic question:

I would like to use an SQLite database, and be able to query and
modify it from Elisp functions. It seems that nobody has written Elisp
bindings for the SQLite library yet, and I am definitely not smart
enough to do that myself.

So I gather I will have to use SQLite's command-line interface,
sqlite3. I don't see much of a problem in parsing the output (in fact,
I've already written that), but I am wondering about the differences
between running sqlite3 as a synchronous process for every query (in a
one-shot fashion), and starting it up once, as an asynchronous
process, and then having it linger in the background, sending it
commands and parsing the output it returns as needed.

To me, the synchronous method seems more robust, but I am wondering
whether it will scale well (supposing I get fond of using the database
and start to use it for more and more complex things). As to the
asynchronous method, I seem to have heard that I/O via stdin and
stdout is prone to, er, I don't know: hang? lock? I've forgotten, but
I would much appreciate to be warned if that is the case.

(I have tried to find hints in ispell.el (since that seems to be a
similar situation), but only found out that it implements indeed
both.)

Can anybody offer me any advice here? Thanks so much!

Florian