[darcs-users] darcs patch: Add forceHashSlurped that hashes the slu... (and 1 more)

Eric Kow eric.kow at gmail.com
Sat Aug 30 18:03:34 UTC 2008


Hi David,

2008/8/30 David Roundy <droundy at darcs.net>:
> It would also be great if he paid attention to feedback on his patches,
> since as I recall they were unacceptable (plastering over a bug, rather
> than fixing it).

Ok.  Petr, I think the most helpful thing here is if you could
motivate the forceHashedSlurp stuff.

In David's last message, I think he said
> What is the purpose of this? The point of caching the hashes was to avoid
> doing this extra IO.  If there's a bug in the caching, we should fix that
> bug rather than working around it!

What are your thoughts on this?

David, reading the archives,
 http://lists.osuosl.org/pipermail/darcs-users/2008-August/012959.html
I got the impression that you over overall favourable to Petr's fix.
Is it just the forceHashedSlurp stuff that you find objectionable?

I'll copy your comments to the remaining patches (the conflicting ones
sadly) below

Tue Aug 12 01:50:39 BST 2008  me at mornfall.net
  * Add Repository.checkPristineAgainstSlurpy.

> Sounds good.

Tue Aug 12 01:54:23 BST 2008  me at mornfall.net
  * First working (albeit slow) version of repair that uses hashed newpristine.

Tue Aug 12 02:06:03 BST 2008  me at mornfall.net
  * Only "update" (sync to disk) the slurpy every 100 patches.

> On the whole, this looks like a nice sequence.  You've got the right idea,
> and apart from minor techincal issues, your code looks great.

Tue Aug 12 02:27:20 BST 2008  me at mornfall.net
  * Fix checkPristineAgainst{Cwd,Slurpy}: we ignored files missing in pristine.

  Add LookForAdds to smart_diff options to fix that and also throw in
IgnoreTimes
  for a good measure and extra paranoia.

> Paranoia is good in check and repair.  It might also be worth having an
> additional old-fashioned check that actually does do the writes to an
> ordinary directory (as the existing code does), just so we can verify that
> there isn't some sort of bug in the writing-to-disk code of darcs (which
> could lead to a very subtly corrupt repository).

So if it's just the forceHashedSlurp stuff that needs working on, it
looks like we're very close!
Thanks,

-- 
Eric Kow <http://www.nltg.brighton.ac.uk/home/Eric.Kow>
PGP Key ID: 08AC04F9


More information about the darcs-users mailing list