Yocto Layer strategy for GDP (was: RE: [PATCH] meta-rvi: Add layers for RVI)
rstreif at jaguarlandrover.com
Mon May 9 13:01:23 EDT 2016
I apologize for top-posting.
I cannot really understand where "the too many layers" fear comes from. I
have not heard any compelling argument why one should/must reduce the
number of layers used to build the system. In my opinion:
1. It makes no sense whatsoever to copy recipes from maintained upstream
layers into your own layers ("privatize them, as Gunnar called it). The
argument of reducing complexity does not hold here. It increases complexity
and maintenance effort.
2. If there is a reasonable chance that a component can be reused in
different contexts, it makes good sense to put it into its own layer. For
example, RVI can be reused outside of automotive contexts e.g. for IoT etc.
It would actually hinder the adoption if RVI were commingled into a
meta-ivi etc. layer.
3. The criticism of meta-ivi as a layer that "collects up tech" as Steve
puts it hold true. It's an umbrella or compound layer that includes
multiple sublayers including BSP layers. It makes things simpler as you
only have to clone one repository form a git server but you still have to
include every sublayer individually into your build environment. However,
it commingles things that logically do not belong together: IVI middleware
with BSPs and other stuff.
4. Avoid "kitchen sink" layers putting everything into one or a few
layers because of a perceived inconvenience of including many layers into
the build environment. This is simply turning the clock back to
OpenEmbedded Classic where all recipes were in one directory. Layers
support re-usability and easier maintenance. Maintenance responsibilities
can easily be divided among many maintainers.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the genivi-projects