On Tue, Apr 21, 2009 at 04:40:04PM +0200, Wolfgang Denk wrote:
Dear Robert,
just to put a few points right:
In message 20090421070431.GX5367@pengutronix.de you wrote:
So our intention was and is:
- Wolfgang has a focus on stability and gradual changes. We respect this political position because it is a *good* one.
This is not quite correct. What I consider important is an evo- lutionary path - this may include bigger changes and reorganizations, but I consider it a bad idea to not provide a reasonable migration path for larger parts of the existing community.
Then start proving your point by removing CONFIG_NET_MULTI. U-Boot carries two incompatible network driver APIs for at least the last seven years and still 19 drivers have not switched to the new API.
It was exactly this kind of stagnation that made me fork U-Boot. I was really tired of this ongoing "no, please don't break existing code". And honestly, I was so blinded by ifdefs that I couldn't even see what the existing code was. So I started hacking on U-Boot in my spare time, not knowing where this would lead, but I must admit that it was really fun to remove everything which I didn't know for what it was good for. You'd be surprised to see how much dead code hides in U-Boot. Additionally I wanted to see what's possible. Is it possible to integrate a driver model without adding much binary space? Is it possible to create a filesystem layer in this limited space? You see I was in a hacking mood and not in a discuss-on-mailing-lists mood. What would have happened if I posted a "I want a driver model" mail on the list? Probably one of the first answers would have been "U-Boot is about size and simplicity". Some mails later I probably would have gone back to business as usual. Even if I tried to integrate a driver model into current U-Boot, how long would it take till all drivers make use of it given the fact that a simple thing like CONFIG_NET_MULTI stays for seven years?
I started this for fun, but I think integrating a driver model into U-Boot without breaking existing stuff is no fun at all and I'm pretty sure none of our customers would have payed us for working on this topic.
Sascha