Re: [RFC/PATCH v2] ubi: Add ubiblock read-write driver

[Ezequiel Garcia <ezequiel.garcia@free-electrons.com>]

Hi Artem,

On Fri, Apr 19, 2013 at 03:31:13PM +0300, Artem Bityutskiy wrote:
> On Fri, 2013-04-19 at 07:57 -0400, Mike Frysinger wrote:
> > the reason we're talking about not allowing write support at the
> > *block* layer 
> > is because it's questionable how many people actually want this, the 
> > performance isn't good (compared to native flash filesystems), and
> > because the 
> > write/wear characteristics are unknown.
> 
> My opinion that unless people demonstrate that they need this, e.g. in a
> product, etc, we should not merge this stuff.
> 
> We recently merged fastmap, which is a big chunk of code, and it looks
> like no one really needs it. There was a problem report, and Richard
> promised to look, but did not. I do not blame him, he is a busy guy. But
> this shoes that this feature is not really needed, while adds
> maintenance burden.
> 
> To put it differently, I do recommend to merge more UBI-related code
> without a solid user-base.
>

(I know you meant don't)

Of course, I agree that we shouldn't merge stuff just for the sake of it.
I'm always against adding churn.

However, IMHO, fastmap and ubiblock are not comparable **at all**:

  1. Ubiblock is *extremely* simple and *completely* independent from mtd/ubi
     stack.

  2. Fastmap is a cool UBI feature, that's pretty tightly integrated
     with the UBI core and -at least to me- not very easy to understand
     and follow.

-- 
Ezequiel García, Free Electrons
Embedded Linux, Kernel and Android Engineering
http://free-electrons.com