[Cubicweb] Python module for Apache : how to use cubicweb.dbapi

[Dimitri Papadopoulos Orfanos]

Hi,

I'm attemting to write an Apache authentication module that would use 
the CWusers of a CubicWeb instance.


We had already discussed the following solution:
* bypass CubicWeb,
* directly connect to the PostgreSQL database associated to the CubicWeb 
instance,
* read the "cw_login" and "cw_upassword" columns of the "cw_cwuser" 
PostgreSQL table,
* understand the encoding of the "cw_upassword" column (SHA1?),
* compare the password to the contents of the "cw_upassword" column.

Note that essential information such as the encoding of the 
"cw_upassword" column is missing.


Anyway, our sysadmins would rather use a different solution:
* use an Apache module written in Python,
* use cubicweb.dbapi to attempt an authentication against CubicWeb.

Can you help me? We believe we should be using dbapi.connect(), but I 
have a hard time finding the correct arguments. Here is my code:
   try:
       cnx = dbapi.connect(database, login='foo', password='bar')

What should the "database" variable look like? I've tried a few values, 
here are my findign so far:

1. database = 'instanceid'
Of course I'd rather use a simple ("in memory"?) connection. Are such 
connections available only through "cubicweb-ctl shell"? Is there a way 
to avoid "cubicweb-ctl shell"?

2. database = 'zmqpickle-tcp://neurospin-cubicweb:8181'
This works but the connection is slow. It also method requires ZMQ 
enabled in the CubicWeb instance.

3. database = 'zmqpickle-tcp://127.0.0.1:8181'
The loopback address does not work! Any clues? Is this a Python or a 
cubicweb.dbapi bug/feature? I need to address both interfaces:
- the loopback interface is used by scripts running on the server,
- the regular interface is used by scripts running on other machines.

Regards,
-- 
Dimitri Papadopoulos
CEA/Saclay
I2BM, NeuroSpin
F-91191 Gif-sur-Yvette cedex, France