Re: [PATCH 0/3][RFC v2] gdbus: Preliminary kdbus-support patches

[Lennart Poettering <mztabzr 0pointer de>]

On Wed, 08.01.14 14:44, Lukasz Skalski (l skalski partner samsung com) wrote:

Heya!

Thanks for preparing those patches! I only had a quick look, but one
thing I noticed: there's no need to process the bus driver messagesin
the library really, we provide a bus driver anyway on the bus, and
people can use it. We do not consider the bus driver legacy or anything,
since it is quite valuable when connecting to kdbus via some other
transport where you don't have direct access to the ioctls.

Hence, I'd recommend not to try to process the bus driver messages
locally, just let them go through and allow the bus driverd service to
handle them.

That said, I think it makes a lot of sense to issue the ioctl()s locally
if there are high-lvel function calls for them defined in gdbus. So to
make this clear: I have nothing agagainst issuing ioctl()s directly,
that's actually even a good thing, but I am not convinced that trying to
parse the bus driver mesages locally is the best idea, we should just
leave that to the bus driver service that we ship anyway.

This would then be very close to what libsystemd-bus does too: if people
manually put together bus driver method calls, we just send them out,
and don't do anything special for them. However, if people call a
high-level function, such as sd_bus_request_name() then we will try to
do the ioctl() directly if we can, only deferring to bus driver messages
if we are not on kdbus...

I hope that makes some sense...

Lennart

-- 
Lennart Poettering, Red Hat