[PATCH] LogFS take three

[Jörn Engel]

On Tue, 15 May 2007 19:37:36 -0700, Roland Dreier wrote:
> 
> There are rather a lot of of FIXME comments, including scary stuff like
> 
>  > +	/*
>  > +	 * FIXME: this cannot be right but it does "fix" a bug of i_count
>  > +	 * dropping too low.  Needs more thought.
>  > +	 */
>  > +	atomic_inc(&old_dentry->d_inode->i_count);

Any thoughts on this would be appreciated.

> and
> 
>  > +int __logfs_write_inode(struct inode *inode)
>  > +{
>  > +	/*
>  > +	 * FIXME: Those two inodes are 512 bytes in total.  Not good to
>  > +	 * have on the stack.  Possibly the best solution would be to bite
>  > +	 * the bullet and do another format change before release and
>  > +	 * shrink the inodes.
>  > +	 */
>  > +	struct logfs_disk_inode old, new;
> 
> are you going to change the format?  or fix this some other way?

I would love to put my inodes on a diet.  It is just a matter of time
and priorities.  To me the 512 bytes on the stack are unfortunate, but
not a show stopper.  Crash behaviour is, so that has priority.

> I think a sweep through the code searching for FIXME and at least
> rewriting all such comments to look like stuff that can be deferred
> would be warranted ;)

Are you asking me to hide known problems under a rug? ;)

Will see if I can easily fix some of these.  In particular the
eat-your-data FIXME that can cause LogFS to not survive a crash.

Jörn

-- 
Correctness comes second.
Features come third.
Performance comes last.
Maintainability is easily forgotten.