[darcs-users] hashed-storage / darcs-hs progress

Jason Dagit dagit at codersbase.com
Sat Oct 3 08:05:31 UTC 2009


On Mon, Sep 21, 2009 at 7:53 AM, Eric Kow <kowey at darcs.net> wrote:

> Just a quick update on the current progress with the hashed storage
> work.
>
> Ganesh, Petr and I had a short meeting on the #darcs channel
> (log: http://irclog.perlgeek.de/darcs/2009-09-20#i_1519390 )
> to work out a practical way to work through the disagreement on some
> points about the hashed-storage work.
>
> The two open questions were
>  (a) if we could divorce some darcs-format-specific parts
>     from hashed-storage
>  (b) if we should generalise the current Hash type to
>     something less darcs-specific
>
> For (b) we all seem to agree that generalising the Hash type could be
> useful.  Ganesh is more convinced of this and would prefer that we do
> this from the start.  Petr is less so and therefore feels that we should
> postpone this till the need actually arises.
>
> Since this is something that can be safely postponed for future work,
> we decided to just treat it as such and focus on getting (a) resolved.
>

Point (a) and (b) both sound really important.  It also sounds wise to stall
on them.  Especially if there is going to be a hashed-storage 1.0 release.
Having said that, I'm not sure that Darcs should wait for the hash-storage
1.0 release before switching to it.


>
> Having agreed to this, we took a closer look at the problem.
> Right now there are two kinds of darcs-format-specific information in
> hashed-storage:
>
>  (a1) conventions for admin files like _darcs, _darcs/hashed_inventory and
>      _darcs/pristine.hashed
>  (a2) format for hashed filenames, the file listing used to represent
>      directories, etc
>
> In the same vein as (b), we think of (a2) as being less urgent than
> (a1), so we narrowed the focus further to (a1).  And since this was
> relatively trivial we decided to just put the code into Darcs proper
> for now.
>
> So this gave us a way forward a small plan of action:
>
> 1. Petr: patch hashed-storage to remove the (a1) stuff and likewise
>   Darcs (if relevant)
>

How is this progressing?


>
> 2. Petr: comment on remaining darcs-hs-specific points on the wiki
>
> 3. Ganesh: Merge darcs-hs patches into mainline Darcs
>

I think we're at a point where several of us are really excited about having
these patches merged into mainline Darcs (or darcs.net as I like to call
it).

It seems like both Ganesh and I have needed to write code against both
branches recently.  I know in my benchmarking I've been bouncing back and
forth between darcs.net and darcs-hs sporadically.  It's really kind of a
waste of time though for me to bounce back and forth though.  Oh, I should
also throw in that darcs-hs is now much faster at my pathological record
problem than darcs.net is.  Much less memory used also.  It's quite
pleasing.

In other words, I'm on board with a shift to darcs-hs at this point.  I'd
like for it to replace the darcs.net source; that is the point at which I'll
start using it as my daily darcs so I can ferret out other bugs and
performance issues.  There is a pragmatic concern here that it is much
easier to submit bug reports for the official darcs source than it is to
submit bugs for darcs-hs.

I'd also like to get on with some other tasks that will need to be redone
after a merge if I do them from the current darcs.net source.

May I be so bold as to propose a deadline for the merge?  Let as much work
as can be done happen before the deadline but once it happens we just merge
things whatever the state is, assuming it's a reasonable state.  Say that by
Nov. 1st we need to have all of Petr's work merged so that at the darcs
sprint we can all build on top of it.


>
> After these small steps, I think we can be looking forward to
> index-based diffing and a faster Darcs for hashed repositories.
>

Yes, this should be quite beneficial.

Please, let me know if I can help.

Thanks,
Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osuosl.org/pipermail/darcs-users/attachments/20091003/92417f33/attachment.htm>


More information about the darcs-users mailing list