On Fri, Nov 01, 2013 at 07:33:00AM -0400, Stephen Gallagher wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 10/31/2013 01:40 PM, Jakub Hrozek wrote:
> On Thu, Oct 31, 2013 at 12:54:04PM +0100, Michal Židek wrote:
>> On 10/27/2013 07:32 PM, Jakub Hrozek wrote:
>>> On Fri, 2013-10-25 at 20:11 -0400, Simo Sorce wrote:
>>>> On Fri, 2013-10-25 at 17:42 +0200, Michal Židek wrote:
>>>>> I decided not to only document the behaviour, but also add
>>>>> option to configure it (so I could remove one FIXME from
>>>>> the code).
>>>>>
>>>>>
https://fedorahosted.org/sssd/ticket/1718
>>>>
>>>> Thanks a lot, glad to see a long standing FIXME go away.
>>>>
>>>> Ack, Simo.
>>>>
>>>
>>> The code itself is good, but you also need to add the parameter
>>> to the confAPI. In particular, src/config/etc/sssd.api.conf and
>>> then make sure make check passes.
>>>
>>
>> Done.
>>
>> Michal
>>
>
> Sorry I didn't think about it the first time around, but I think
> the code should also initialize offline_timeout to the default
> value so that if confdb_get_int fails for whatever reason, we still
> decide based on the default value. Ignoring the result if
> confdb_get_int is fine, but we could warn with a DEBUG message.
If confdb_get_int() fails, it can pretty much only mean that the value
in the config file was not an integer. In that case, we should
*really* be quitting.
Not really, there can be a number of internal errors as well. I think
the proper solution for the problem you describe is a configuration file
validator, but that's really out of scope of this change.
I'd rather see us check this option at startup and store it in the
be_ctx, that way we can abort (or loudly log that it's wrong and fall
back to defaults) at startup.
I don't like extending be_ctx with per-option attributes. What if we
converted all generic backend options into a dp_option array instead?