Re: removing duplicate code

On 12 Oct 2002, Robert Collins wrote:

> Yes,
> IMO in the long run it should be as follows:
> aufs, diskd go away.
> ufs gets the ufscommon code again.
>
> The aufs and diskd and ufs disk access logic become diskio modules, not
> cache_dir modules. Then the ufs store takes a parameter - which diskio
> module to use.

I think having the common directory as a library is fully sufficient for
the purpose.

Trying to abstract into "directory management module" and "iomodule" is
probably too much effort, and I see such move more likely to paint
yourself into a corner than any real benefits. If you only consider the
"plain-stupid-simple" ioimplementations we have today then sure there may
appear to be a benefit, but when you start looking into macro I/O
operations then the boundary becomes fuzzy. Should perhaps also mention
that some of the blocking ufscommon directory management code really
should to be made aufs specific for the aufs store..

What should perhaps be a separate module is the index management today
done in the core code, but I am not convinced on this either.. probably a
internal library is more appropriate there too. When you start looking at
more advanced stores then the boundary between index management and I/O
becomes more and more fuzzy.

Regards
Henrik
Received on Sat Oct 12 2002 - 03:51:46 MDT