On Thu, 2012-04-19 at 06:39 +0800, Andy Green wrote:
As I pointed out above, the issue is not that I happen to be looking for WG content, but that if you will bind together multiple LT trees they have to have a chance to converge on what ingredients you are mandating.
CMA is a good example, even in TI people have multiple trees with their stuff working only on particular versions of CMA series. Because I guess all the LT trees will currently have their own CMA, you stand no chance to bind them together in a unified tree unless they all agree to work on a single CMA version. It's not just CMA but other prerequisites like dmabuf, UMM pieces or whatever else is not-quite-in-mainline-yet. Sometimes that's turnkey but other times, uplevel in LT tree to what you want them to use may be nontrivial or involve politics or domain experts that don't want to do it. If it interacts with third-party code like 3D unit driver, it may not even be possible to do it on a given timescale for a given LT.
Since there will likely be multiple common ingredients like that, you need to provide the LTs with a flavouring tree for test so they can move towards sync against the mandated ingredient versions (hopefully, the latest stuff that works in each case). Otherwise, you don't give LTs a chance to converge or even test with your ingredients (not to mention forward planning so we can recruit help from domain experts at better than zero notice). That's the chicken-and-egg if your ingredients are only coming at post-unification tree how are we meant to prep our LT trees to combine smoothly?
I totally agree with Andy.
I don;t believe the common Linaro tree can work unless teams contributing to it know, and have access to, the common ingredient topics they depend on. This requires agreement, or at least an official announcement each month, of what version of a topic is going in; and for this to be done in good time for people to integrate, test and fix this.
If theses topics people depended on were also available in a pre-megered and tested flavouring branch then this would save duplicated effort and time.
(For the current ARM LT situation we have a simple life because we don't depend on anything other than mainline, but still, we could do better and earlier testing of our code and board if a Linaro flavouring topic was available.)