[arc-discuss] Re: ARB rabbit holes (was: Requesting PSARC/2007/124 details)

[Dale Ghent]

On Mar 28, 2007, at 5:53 PM, Bill Sommerfeld wrote:

> On Wed, 2007-03-28 at 16:37 -0400, Dale Ghent wrote:
>>    3) (Paraphrasing yourself) "ARC documents are not documentation"
>
> that's an incorrect paraphrase.
>
> ARC documents most certainly are documentation but they do not
> automatically create a commitment that an interface will not change
> again in the future.  you need to carefully review the interface
> taxonomy for that.

Right, I understand that. When I said "are not documentation" I was  
referring to what James said in that man pages are definitive  
documentation for a public interface (which VFS isn't, as has been  
pointed out).

Those 3 points I made are to point out the practical dilemma that  
third party FS developers are put into in that we need to use VFS to  
implement our fs. Doing this necessitates the need to use non-Public  
interfaces, so we must fall back to using source code as a form of  
reference documentation, however unsavory that may be... but that's  
the way it is. When an ARC case such as this one comes along, it  
would help us greatly to have more than just a code diff to see what  
that chance means to us. (and that spawned the default-to-open  
discussions)

It's kind of surreal to be (figuratively) rapped across the knuckles  
when I say that we must use these sorts of non-Public interfaces to  
implement a file system. I mean, how else could one do it!

So for this particular problem, VFS/vnode need to go through whatever  
process they need to go through (be it technical in nature or  
otherwise) to make them Public. I don't believe there's a lack of  
demonstrated need for this, and this can't be the first time that  
this type of issue has arisen, though it may have historically been a  
rare occasion.

/dale