On 30.07.2008 01:53, Stefan Reinauer wrote:
Carl-Daniel Hailfinger wrote:
On 29.07.2008 22:10, ron minnich wrote:
On Tue, Jul 29, 2008 at 1:05 PM, Stefan Reinauer stepan@coresystems.de wrote:
Peter Stuge wrote:
Yes, I argued strongly for this when they first appeared and even sent a patch. The problem is that the filenames are tied hard into the struct names generated by dtc.
Hm. I definitely want to support your idea here then.
Adding .dts to the filename is about as hard as not doing that. Also, in 2 out of 3 dts files I see struct names.
And what about files which are named dts right now? Do we call them dts.dts (ugly) or .dts (hidden file)? Stripping a given file suffix in dtc before creating the struct name is easy.
Read my initial proposal.
That proposal said:
To make clear what those files are, we should rename them...
dts -> mainboard.dts ide -> ide.dts apic -> apic.dts
I can either take your words literally ( southbridge/amd/cs5536/dts becomes southbridge/amd/cs5536/mainboard.dts ) or I take their intent and conclude the initial proposal was incomplete.
If we'd really autocreate something, we should drop that behavior.
struct name autocreation is a feature I really like.
Absolutely. "That behavior" meant mentioning the struct name manually in the dts files.
Indeed. I'll take a look at those files in the next few months. I'd be happy if someone could tackle this before.
oh no! it's harder! we blew it!
I'd like to disagree. I still haven't fully understood the v2 device tree, while the v3 device tree seems obvious and simple to me.
On a code level beyound the dts they're 100% the same. Now go compare a mainboard Config.lb (minus the makefile stuff) to the scattered dts mentioning struct names for components etc etc. It is really much more complex than in v2. Yet, it does not have more features.
v3 has a few perceived problems and a few real problems. The problem is that everybody has his own idea about which problems are real. I'm not claiming that my version of the story is the absolute truth(tm)
What exactly are you trying to say?
It's like flashrom. Agreeing on features and roadmap is probably more difficult than coding stuff up.
What is your version of the story anyways?
The v2 config files are completely unreadable. In v3, the situation is a lot better. Maybe not optimal, but orders of magnitude better than v2. One thing I see as a problem in both versions is how I can specify different settings for each instance of a chip appearing multiple times on a board. (I may be misinterpreting struct name generation...)
Regards, Carl-Daniel