[SSSD] [PATCH] Implement type-safe getters for primitive types and their arrays

Stef Walter stefw at redhat.com
Thu May 22 13:36:06 UTC 2014


On 22.05.2014 14:00, Jakub Hrozek wrote:
> On Thu, May 22, 2014 at 01:09:06PM +0200, Stef Walter wrote:
>> On 21.05.2014 22:49, Jakub Hrozek wrote:
>>> On Fri, May 16, 2014 at 01:09:12PM +0200, Lukas Slebodnik wrote:
>>>> On (16/05/14 00:48), Jakub Hrozek wrote:
>>>>> On Tue, May 13, 2014 at 07:04:22PM +0200, Stef Walter wrote:
>>>>>> On 12.05.2014 23:27, Jakub Hrozek wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> the attached patchset implement the
>>>>>>> org.freedesktop.DBus.Properties.Get() interface for properties of
>>>>>>> primitive types and arrays of primitive types (except for array of bools
>>>>>>> as DBus bool is of different size than C-sized bool).
>>>>>>
>>>>>> This looks nice to me. A few comments.
>>>>>>
>>>>>> +    dbus_message_iter_init_append(reply, &iter);
>>>>>> +    dbret = dbus_message_iter_open_container(&iter, DBUS_TYPE_VARIANT,
>>>>>> +                                             strtype, &valiter);
>>>>>> +    if (!dbret) {
>>>>>> +        sbus_request_finish(dbus_req, NULL);
>>>>>> +        ret = EINVAL;
>>>>>> +        goto done;
>>>>>> +    }
>>>>>>
>>>>>> You probably want some diagnostics here, as this is where you'll end up
>>>>>> if you pass in an invalid @type.
>>>>>
>>>>> Thanks, I was fooled by the doxygen comment that said the function only
>>>>> returns false on OOM. But the code is pretty conclusive, so I've added a
>>>>> nicer error messages for this case and the others as well.
>>>>>
>>>>>>
>>>>>>> +    for (i = 0; i < len; i++) {
>>>>>>> +        addr = values + i * item_size;
>>>>>>> +        dbret = dbus_message_iter_append_basic(&array_iter, type, addr);
>>>>>>> +        if (!dbret) {
>>>>>>> +            sbus_request_finish(dbus_req, NULL);
>>>>>>> +            ret = ENOMEM;
>>>>>>> +            goto done;
>>>>>>> +        }
>>>>>>> +    }
>>>>>>
>>>>>> Are you planning on using this for byte arrays and such? If so it may be
>>>>>> worth looking into special casing arrays of bytes/ints with
>>>>>> dbus_message_iter_append_fixed_array()
>>>>>>
>>>>>> http://dbus.freedesktop.org/doc/api/html/group__DBusMessage.html#ga6e2d1e936c3c61fe00d80a3f22fd5e76
>>>>>>
>>>>>> But maybe that's not needed in this initial patchset?
>>>>>
>>>>> I haven't changed this part in the attached patchset, but I'm fine with
>>>>> the change you propose -- maybe I just don't see the rationale. Is there
>>>>> a performance boost? Is it worth the additional if-else?
>>>>>
>>>>>>
>>>>>>> From d23886c484bc955bf4bc03998601aeec1650ce02 Mon Sep 17 00:00:00 2001
>>>>>>> From: Stef Walter <stefw at redhat.com>
>>>>>>> Date: Tue, 25 Feb 2014 18:31:03 +0100
>>>>>>> Subject: [PATCH 4/6] WIP properties
>>>>>>
>>>>>> Should I give this a better commit message? :)
>>>>>
>>>>> Oops, that's something I meant to change before sending the patches and
>>>>> then didn't. The attached patchset changes the message to something
>>>>> hopefully better..feel free to propose a different commit message,
>>>>> though.
>>>>>
>>>>>>
>>>>>>> +    if prop.type == "s":
>>>>>>> +        out("    out_val = prop_val == NULL ? \"\" : prop_val;")
>>>>>>
>>>>>> You'll probably also want to special case object paths, returning "/" in
>>>>>> those cases.
>>>>>
>>>>> Yes. I also added unit tests for both these special cases.
>>>>>
>>>>> Thank you very much for the review. New patches are attached.
>>>>
>>>> I let review for others.
>>>>
>>>> I just want to write that problem with builing patches on rhle6 is fixed
>>>> and warnings from static analysers are gone.
>>>>
>>>> LS
>>>
>>> Thanks for running the patches through static analysers and testing the
>>> build!
>>>
>>> I'm attaching a new patchset that has some simplifications in the unit
>>> test warranted by the GetAll patchset and include the Get method in the
>>> introspection XML.
>>>
>>> Any comments on the patches are welcome!
>>
>> ACK.
>>
>> Although I do wonder if all interfaces should have the properties
>> interface in their introspect data unconditionally. I guess pretty much
>> any infopipe interface will have DBus Properties ... so this is fine for
>> now.
>>
>> Stef
> 
> I was wondering that as well and wasn't sure what's the best practice in
> the DBus world. I can only add the interface to Introspection if there
> are properties in the meta, if that's what you'd prefer.
> 
> My reasoning was that you can always send the Get method to the
> interface, even if there are no properties, the sbus handler would just
> kick you out.

Makes sense ... and that's likely how it's implemented in a lot of
places. Haven't checked.

Stef




More information about the sssd-devel mailing list