Re: [RFC][PATCH 4/5] ar9170: USB frontend driver

[Michael Buesch <mb@bu3sch.de>]

On Saturday 21 March 2009 13:54:39 Johannes Berg wrote:
> On Sat, 2009-03-21 at 13:49 +0100, Michael Buesch wrote:
> 
> > > > Well I think it would currently generate lots of errors, because we have
> > > > code like this in the kernel:
> > > > 
> > > > if (x)
> > > > 	mutex_lock();
> > > > ...
> > > > if (x)
> > > > 	mutex_unlock();
> > > 
> > > That's sloppy code anyway. Not to be encouraged.
> > 
> > That's not true. Sometimes it is the cleanest way to do things.
> > Look at drivers/ssb/main.c. To make this mutex-sparse compliant, we'd
> > need to introduce quite a few sub-functions.
> > 
> > It simply is a limitation of sparse. Nothing else.
> 
> No, I still think it's sloppy code;

Patches that change the code into code with _better_ style are accepted.
However, I don't think introducing more foo(); calls __foo(); style stuff is better.
It's just a workaround to sparse.

> some future work will in most cases 
> invariably move the conditions further apart, at which point it becomes
> more and more unlikely that the invariant that the "x" doesn't change
> inbetween is maintained.

Did you look at the example I gave?

-- 
Greetings, Michael.