I have a case where pushing a patch from repo A to repo B fails, but pulling in the other direction succeeds.  Igloo tells me this is definitely a bug.

Repo B (ghc-par-gc) is partial, repo A (ghc-head-tmp) is not:

---------------------
~/darcs/ghc-head-tmp > darcs push ../ghc-par-gc -p "whitespace"
Pushing to "../ghc-par-gc"...

Fri Nov 10 18:36:33 GMT 2006  Samuel Bronson <naesten@gmail.com>
  * Zap stray whitespace in lhs formatting
Shall I push this patch? (1/?)  [ynWvpxqadjkc], or ? for help: n

Mon Nov 20 14:25:50 GMT 2006  Simon Marlow <simonmar@microsoft.com>
  * whitespace
Shall I push this patch? (2/?)  [ynWvpxqadjkc], or ? for help: y
darcs: Cannot apply this patch bundle, this is a "--partial repository
We don't have the following patch:
Wed Sep 20 21:48:05 BST 2006  Manuel M T Chakravarty <chak@cse.unsw.edu.au>
  * Correct substitution into theta types (after merge)

~/darcs/ghc-head-tmp > cd ../ghc-par-gc
~/darcs/ghc-par-gc > darcs pull ../ghc-head-tmp -p "whitespace"
Pulling from "../ghc-head-tmp"...

Fri Nov 10 18:36:33 GMT 2006  Samuel Bronson <naesten@gmail.com>
  * Zap stray whitespace in lhs formatting
Shall I pull this patch? (1/?)  [ynWvpxqadjkc], or ? for help: n

Mon Nov 20 14:25:50 GMT 2006  Simon Marlow <simonmar@microsoft.com>
  * whitespace
Shall I pull this patch? (2/?)  [ynWvpxqadjkc], or ? for help: y
Finished pulling and applying.
---------------------

I have saved the exact state of the repos here:

  http://haskell.org/~simonmar/darcs-bug-report/ghc-par-gc.tar.gz
  http://haskell.org/~simonmar/darcs-bug-report/ghc-head-tmp.tar.gz

(30M & 50M respectively)

I have been seeing odd things happen when transferring patches between partial repositories in GHC for a while, so it's nice to nail down a case that is definitely a bug.  With any luck this might help pinpoint the problem.

Cheers,
        Simon

Simon Marlow wrote:
> New submission from Simon Marlow <simonmar@microsoft.com>:
>
> I have a case where pushing a patch from repo A to repo B
> fails, but pulling in the other direction succeeds.  Igloo
> tells me this is definitely a bug.

Of course it's not really "in the other direction", the direction is the same, just that the operation is a pull rather than a push.  Apologies for the confusion.

Cheers,
        Simon

On Mon, Nov 20, 2006 at 02:58:58PM +0000, Simon Marlow wrote:
> I have a case where pushing a patch from repo A to repo B fails, but
> pulling in the other direction succeeds.  Igloo tells me this is
> definitely a bug.
>
> Repo B (ghc-par-gc) is partial, repo A (ghc-head-tmp) is not:

The trouble here I suspect stems from the heuristics we use to minimize
access to remote repositories, which are very crude.  This is an
optimization that doesn't help when the local repository doesn't actually
have the patches.  I'm hoping to be able to somewhat obsolete this bug by
making repositories using the hashed inventory format (not yet working)
able to gradually make themselves un-partial.  Which is a sort of memoized
version of the algorithm in this case that we'd have to used anyways
(i.e. no unnecesary downloads).

So yes, this is a bug, in that we could certainly make the algorithm more
flexible in trying different options for getting the patches we need.  But
it may be a tricky one to fix.  :(
-- 
David Roundy
Department of Physics
Oregon State University

I believe this is considered resolved-in-unstable by using --lazy instead of
--partial, which avoids the bug.