[darcs-users] no more checkpoints, but hashed repos for GHC in Darcs 2.4?

Simon Marlow simonmar at microsoft.com
Mon Sep 21 08:42:44 UTC 2009


> On Sat, 19 Sep 2009, Petr Rockai wrote:
> 
> > Ganesh Sittampalam <ganesh at earth.li> writes:
> >
> >> On Thu, 17 Sep 2009, Petr Rockai wrote:
> >>> For cache, this should be quite simple. For repos, this would be
> incompatible
> >>> change and we probably can't afford to do it just yet.
> >>
> >> Isn't it a straightforward format change? Add support to hashed-
> storage, add an
> >> extra line to the format to lock out older darcs and to drive the
> decision to
> >> read/write the new format, perhaps add a conversion command to darcs
> (otherwise
> >> make people re-get to switch formats), and that's it?
> 
> > Well, yes, but I'm not sure we want to start proliferating formats. It
> > comes at a maintenance cost and code bloat. And it inconveniences users
> > and I'm not sure the performance gain is enough to offset that.
> 
> Personally I think that if the GHC team really consider that they need it
> then we should do it. On the other hand I don't find the performance
> difference that Simon found so far all that compelling given that having a
> warm cache makes the difference go away - Simon, could you comment further
> on how important this is?

You're right - I wouldn't say this is a high priority based on those Linux/ext3 figures I posted.  However, it might be a problem in other setups.  For example on Windows I reported that 'darcs whatsnew' on a hashed repo with 2.3 is much slower than an unhashed repo (don't know if there's a ticket for this one, I couldn't see one).  I don't know whether this is due to the large directories or something else. 

Cheers,
	Simon



More information about the darcs-users mailing list